Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Aggie The Pew message board
: Winter Solstice - GFN21
| Author |
Message |
|---|
|
|
|
So the last challenge of the year is coming up. GFN21 - this should be fun.
Anyone who would like to participate is more than welcome to join in.
Cheers
And now a question. So I was working on the cpu side.. got 5 systems all setup but forgot to include the <plan_class> token. Fixed that and then let some units run. Every single one seems to have error'd out right around the 3600 sec mark. As far as I know nothing on the cpu is over clocked so I'm pretty sure I've either left out some more tokens or something is really wrong with how I am running them.
I was double checking my app_config and found that it contained the following and on the boards it's different, so is the -nt important? I'm thinking it might be
<cmdline>-t 2\</cmdline> - mine
<cmdline>-nt 2</cmdline> - boards | |
|
|
|
|
Under 8 days away and I haven't even started to think about what to run... like, are CPUs competitive to GPUs with multi-threads? Is this affected by the nvidia 400 driver thing too? | |
|
 Michael Goetz Volunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
                           
|
So the last challenge of the year is coming up. GFN21 - this should be fun.
Anyone who would like to participate is more than welcome to join in.
Cheers
And now a question. So I was working on the cpu side.. got 5 systems all setup but forgot to include the <plan_class> token. Fixed that and then let some units run. Every single one seems to have error'd out right around the 3600 sec mark. As far as I know nothing on the cpu is over clocked so I'm pretty sure I've either left out some more tokens or something is really wrong with how I am running them.
I was double checking my app_config and found that it contained the following and on the boards it's different, so is the -nt important? I'm thinking it might be
<cmdline>-t 2\</cmdline> - mine
<cmdline>-nt 2</cmdline> - boards
3600 is one hour...
The Genefer error recovery is designed to do the following when an error occurs...
1) Print something to stderr
2) Pause 10 minutes
3) Try again
4) After 6 tries (1 hour) give up for good and abort the task.
The reason is that there's a lot of transient errors that can break a GPU task, some of which may go away by themselves if you wait. If you wait, you may still be able to complete the task.
Also, even if the computer is configured such that it's completely impossible to run a GPU task, the 1 hour delay means that we're burning a task once an hour instead of once every three seconds.
So... look in stderr to see what isn't working.
____________
My lucky number is 75898524288+1 | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
|
Looking at the stdout on those tasks, it's complaining about the -t flag being incompatible with the -q flag.
You tried something like "-t 4", right?
Wrong.
This ain't LLR.
Multithreading would be "-nt 4". "-t" was already in use by Genefer for something else.
EDIT: (I didn't remember what it was either. I had to look it up.)
____________
My lucky number is 75898524288+1 | |
|
|
|
|
Thanks Mike. Will do that. Even though CPUs will be slow I'd like to get a few extra units in.
Does -t vs -nt make any difference? | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
Thanks Mike. Will do that. Even though CPUs will be slow I'd like to get a few extra units in.
Does -t vs -nt make any difference?
Well, -nt does what you want (uses multiple threads on a CPU), while -t does something completely different (it runs a test suite to verify the hardware and software is operating correctly.) So, yeah, I'd say it makes a difference. :)
____________
My lucky number is 75898524288+1 | |
|
|
|
|
Then I suppose I've been lucky. All the llr tasks ran fine. Thanks again. | |
|
|
|
... like, are CPUs competitive to GPUs with multi-threads?
I remember running GFN20's on my Skylakes with really good turn around times and RAC. For whatever its worth:
(1) i7 6700k - 25 hrs/unit - RAC 120,691 credits/day
(2) i3 6300 - 31.9 hrs/unit - RAC 47,278 credits/day
(3) i3 6100 - 31.9 hrs/unit - RAC 47,278 credits/day
(4) i7 3630QM - 50.64 hrs/unit - RAC 44.687 credits/day
I would not shy away from using recent CPU's with GFN21. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
Then I suppose I've been lucky. All the llr tasks ran fine. Thanks again.
Not luck. LLR is not Genefer. The parameter when using LLR is '-t'. The parameter when using Genefer is '-nt'. Different programs use different parameters.
____________
My lucky number is 75898524288+1 | |
|
|
|
Then I suppose I've been lucky. All the llr tasks ran fine. Thanks again.
Not luck. LLR is not Genefer. The parameter when using LLR is '-t'. The parameter when using Genefer is '-nt'. Different programs use different parameters.
That makes sense. Thanks again. | |
|
|
|
|
So after looking at the stderr and adding the plan_class and updating to -nt x I have 2 systems running a gfn21. My 4 core mac says it will take 1 day 16 hrs and my 2 core i5 a little over 3 days. The important part, they are running and haven't error'd and it looks like I should be able to get a few cpu tasks in during the challenge.
Thanks again to Mike for pointing out the important -nt flag. I of course had just copied the app from an llr. Here's what I have running now on my i5
<app>
<name>genefer</name>
<max_concurrent>1</max_concurrent>
<fraction_done_exact/>
</app>
<app_version>
<app_name>genefer</app_name>
<cmdline>-nt 2</cmdline>
<avg_ncpus>2</avg_ncpus>
<plan_class>cpuGFN21</plan_class>
</app_version> | |
|
|
|
I remember running GFN20's on my Skylakes with really good turn around times and RAC. For whatever its worth:
(1) i7 6700k - 25 hrs/unit - RAC 120,691 credits/day
(2) i3 6300 - 31.9 hrs/unit - RAC 47,278 credits/day
(3) i3 6100 - 31.9 hrs/unit - RAC 47,278 credits/day
(4) i7 3630QM - 50.64 hrs/unit - RAC 44.687 credits/day
I would not shy away from using recent CPU's with GFN21.
To confirm, the above was your testing for GFN20 not GFN21? What ram is in it?
I'm doing a GFN21 test unit using 4 cores on a 6700k right now, stock settings (4.0 GHz all cores) with 3200 ram. As of when I last looked, it was estimating around 24 hours but I should have a clearer picture tonight. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
|
My i5-4670K running 3 cores will do about 44 hours, and 4 cores will do about 38 hours. Both run a bit slower if the GPU is also running GFN21, which, of course, it will be doing. I suspect memory bandwidth to be the culprit because even the 3 core version slows down.
I'll probably run 3 cores.
The 1060 is claiming about 20 hours, so it will probably do about 11 or 12 tasks.
I've also got 2 laptops that will run -mt 2. I haven't benchmarked them but they should certainly be able to do at least 1 or 2 each. The lesser CPUs will be sitting this one out.
All in all, I'll probably get about 18 to 20 tasks done.
____________
My lucky number is 75898524288+1 | |
|
|
|
I remember running GFN20's on my Skylakes with really good turn around times and RAC. For whatever its worth:
(1) i7 6700k - 25 hrs/unit - RAC 120,691 credits/day
(2) i3 6300 - 31.9 hrs/unit - RAC 47,278 credits/day
(3) i3 6100 - 31.9 hrs/unit - RAC 47,278 credits/day
(4) i7 3630QM - 50.64 hrs/unit - RAC 44.687 credits/day
I would not shy away from using recent CPU's with GFN21.
To confirm, the above was your testing for GFN20 not GFN21? What ram is in it?
Yes, with GFN20 using dual channel RAM (2x4 Gb) running at 2133 MHz due to mobo restriction and with the CPU's running at stock (I never overclock my CPU's or RAM). | |
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2165
ID: 1178
Credit: 8,777,295,508
RAC: 0
                                     
|
The 1060 is claiming about 20 hours, so it will probably do about 11 or 12 tasks.
My two GTX 1060s 6GB variety do GFN21 in about 70,000 seconds, so just a bit under 20 hours. The 3GB variety runs about 1 hour slower than that.
I'll be all in on the GPU side from the RTX 2070 FE (just over 28,000 seconds) all the way down to a GT 540M card to see if the latter can get through one unit at least. Will run as many FMA3 and AVX CPUs as I can get MT set up on.
AtP is going to need an all hands on deck for this one if we are going to compete for a top 3 position!
| |
|
|
|
AtP is going to need an all hands on deck for this one if we are going to compete for a top 3 position!
I am actively searching for an i5 or i7 Coffee Lake CPU. If I get it before the start of the challenge, I will be all in as well with both CPU's and GPU's since the extra CPU will easily make up for the ten days otherwise lost. Otherwise I must stay focused with GCW Sieve so to finish mid-March while also participating fully with TdP a month earlier. | |
|
|
|
AtP is going to need an all hands on deck for this one if we are going to compete for a top 3 position!
I am actively searching for an i5 or i7 Coffee Lake CPU. If I get it before the start of the challenge, I will be all in as well with both CPU's and GPU's since the extra CPU will easily make up for the ten days otherwise lost. Otherwise I must stay focused with GCW Sieve so to finish mid-March while also participating fully with TdP a month earlier.
Of course you could move your gpus over if you wanted :) Just a thought. | |
|
|
|
AtP is going to need an all hands on deck for this one if we are going to compete for a top 3 position!
I am actively searching for an i5 or i7 Coffee Lake CPU. If I get it before the start of the challenge, I will be all in as well with both CPU's and GPU's since the extra CPU will easily make up for the ten days otherwise lost. Otherwise I must stay focused with GCW Sieve so to finish mid-March while also participating fully with TdP a month earlier.
Of course you could move your gpus over if you wanted :) Just a thought.
That crossed my mind as well but I do not want to impact the CPU processing and which would eventually push back my end date. End of March will be the one-yr anniversary when the suspension/termination was first brought up as a possibility for GCW Sieve.
I can also do what I did last year with TdP by starting later into the month but I do want the TdP 2019 badge. I came close last year in spite of my late start (Feb 20th?) but I was a double-checker.
I have put a lot of work into GCW Sieve and I do want it to come to an end. It has been a long exercise. I will be at 80M mid-month and well into 90M+ by the end of January. TdP thereafter and 19-20 days in March for the final sprint to 100M.
I will be able to shave a lot of time if I can get my hands on a Coffee Lake CPU. I am very active in my search.
Edit: I could be done with GCW Sieve once and for all (100M) by January if I can get it sooner than later. In fact, I would be able to just squeeze in the GFN21 challenge as well. It makes that much of a difference.
Edit 2: I just checked my Excel spreadsheet: from an RAC perspective, GFN20 was my best CPU project when I ran my credits for the GFN badge to Sapphire. No question I will be looking for another shield at some point but now with GFN. | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 2829
ID: 130544
Credit: 954,793,678
RAC: 42
              
|
|
I'll be all-in with CPU & GPU having previously set up app_config & benchmarked a few weeks ago while you were all sleeping. | |
|
|
|
|
In preparation for the upsoming challenge I decided to run a few gfn21 work units on some slow gpu's as they've still got to be a lot faster than doing the cpu thing, and to see how they do as to issues/no-issues.I'll be adding some 1060's & 970's for the challenge. Will be rooting for the team!! :-) | |
|
|
|
|
I am looking to keep my app_config file as complete as possible. Can someone please confirm for me if the following addition is complete and correct for a four-core processor such as the i7 6700k Skylake:
<app>
<name>genefer</name>
<fraction_done_exact>1</fraction_done_exact>
</app>
<app_version>
<app_name>genefer</app_name>
<cmdline>-nt 4</cmdline>
<avg_ncpus>4</avg_ncpus>
<max_ncpus>4</max_ncpus>
</app_version> | |
|
|
|
|
You also need the following in the <app_version> section
<plan_class>cpuGFN21</plan_class>
If I am incorrect I'm sure someone will speak up.
Cheers | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
I am looking to keep my app_config file as complete as possible. Can someone please confirm for me if the following addition is complete and correct for a four-core processor such as the i7 6700k Skylake:
<app>
<name>genefer</name>
<fraction_done_exact>1</fraction_done_exact>
</app>
<app_version>
<app_name>genefer</app_name>
<cmdline>-nt 4</cmdline>
<avg_ncpus>4</avg_ncpus>
<max_ncpus>4</max_ncpus>
</app_version>
The <max_ncpus> block has exactly the effect as a <big_mac> block and a <dilbert> block, i.e., none at all. "max_cpus" is not a valid tag inside app_version.
The good news is it will just be ignored and won't hurt anything. The bad news is everyone who reads your post will think that tag is necessary unless someone else corrects you.
And, as Rick said, you need the <plan_class> tag. It won't work without it.
After correcting those tags, please test it to make sure it works. Especially if you have multiple computers, the start of a challenge is not the time you want to be debugging your config files.
Edit: This is what I'm using on my computers. Notice there's no <app> section. It's not really necessary.
<app_config>
...
<app_version>
<app_name>genefer</app_name>
<cmdline>-nt 4</cmdline>
<plan_class>cpuGFN21</plan_class>
<avg_ncpus>4</avg_ncpus>
</app_version>
...
</app_config>
____________
My lucky number is 75898524288+1 | |
|
|
|
The <max_ncpus> block has exactly the effect as a <big_mac> block and a <dilbert> block, i.e., none at all.
I had to read this one several times before I had a good laugh. Very good ! | |
|
|
|
|
Depending on your system(s) the following might not make any difference. However on a few of mine, I had been running both GFN15 on the gpu and GCW sieves on the cpus. I turned off the cpu and my gfn15 times dropped a ton. ie, on my 1070 the tasks went from 60 to 40 seconds-ish. That's a big drop in time. My other systems not such a noticeable improvement but some. So I won't be running any cpu tasks on the systems with gpus but will be on my others. And of course this might just be the gfn I'm running and gfn21 is not as cpu intensive. But the old "leave a core free to feed the gpu" might hold true.
And again, I have really old memory, bad cache and slow what nots. | |
|
|
|
The <max_ncpus> block has exactly the effect as a <big_mac> block and a <dilbert> block, i.e., none at all. "max_cpus" is not a valid tag inside app_version.
The good news is it will just be ignored and won't hurt anything. The bad news is everyone who reads your post will think that tag is necessary unless someone else corrects you.
And, as Rick said, you need the <plan_class> tag. It won't work without it.
After correcting those tags, please test it to make sure it works. Especially if you have multiple computers, the start of a challenge is not the time you want to be debugging your config files.
Edit: This is what I'm using on my computers. Notice there's no <app> section. It's not really necessary.
<app_config>
...
<app_version>
<app_name>genefer</app_name>
<cmdline>-nt 4</cmdline>
<plan_class>cpuGFN21</plan_class>
<avg_ncpus>4</avg_ncpus>
</app_version>
...
</app_config>
Thanks for cleaning up the various sections for me.
I have now modified the app_config file across all of my systems. | |
|
|
|
The <max_ncpus> block has exactly the effect as a <big_mac> block and a <dilbert> block, i.e., none at all.
I had to read this one several times before I had a good laugh. Very good !
I have seen it before and it is quite good. | |
|
|
|
|
I'm hoping to set up one of my rigs this evening to join the Challenge. I'll be making full use of the <dilbert> block.
<dilbert>Out, out!! You demons of stupidity!!</dilbert>
____________
Proud member of Team Aggie the Pew
"Wir müssen wissen. Wir werden wissen."
"We must know, we shall know."
- David Hilbert, 1930 | |
|
|
|
|
So I just logged in a few minutes ago, and the big event timer sais 5hrs 32mins to go, yet my internet clock sayid 0848, PST. Woulda thot the minutes counter should be very close like matching. Am I missing something?? | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 2829
ID: 130544
Credit: 954,793,678
RAC: 42
|
|
Always go off the small font UTC time at the top of that section. The big counter takes it from your own PC's clock. Correct your clock manually if you need to. Start is 22:23UT. | |
|
|
|
|
If you are using windows, you can use gpedit and set your time server under the computer configuration, administrative templates, system, windows time service.
There you will enable the time server and client side.
I use us.pool.ntp.org as the time server.
In the configure windows ntp client (first option) use us.pool.ntp.org,0x9 and make sure the server type is NTP
In enable windows ntp client (2nd option) make sure you enable it.
then to verify in a cmd prompt running under admin rights, you can type w32tm /query /status and it will show you if the time server has been contacted and is in use.
This way the time on my systems stay on the correct time.
of course if you are at work, they may have a group policy already setup
and you might need to allow it through the firewall if it's on | |
|
|
|
|
There is also a neat utility which is freeware that pings to well-known time servers at intervals you specify ahead of time (ex every ten minutes), thus keeping your Windows systems right up to date.
See http://www.thinkman.com/dimension4/. | |
|
|
|
|
My systems are ready to go. All on GFN16 (GPU)/PPSE (CPU). I should have a pretty clean start. | |
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2165
ID: 1178
Credit: 8,777,295,508
RAC: 0
|
My systems are ready to go. All on GFN16 (GPU)/PPSE (CPU). I should have a pretty clean start.
Me, too. But GFN15 (GPU)/SGS (CPU). Nice side effect is finding 3 SGS and 4 GFN15 primes in the last 30 hours or so. :)
| |
|
|
|
|
Today is monthly Windows Update day isn't it? So if you haven't blocked those, watch out for reboots, or do the updates manually before the start. | |
|
|
|
Today is monthly Windows Update day isn't it? So if you haven't blocked those, watch out for reboots, or do the updates manually before the start.
Ugh. I forgot about that. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
Today is monthly Windows Update day isn't it? So if you haven't blocked those, watch out for reboots, or do the updates manually before the start.
Thanks. I'm going to all the computers and updating them now, before the challenge starts.
____________
My lucky number is 75898524288+1 | |
|
|
|
Nice side effect is finding 3 SGS and 4 GFN15 primes in the last 30 hours or so. :)
Not that I had any doubts but I decided to take a quick peek at SGS: the last six consecutive finds have Z-san as primary discoverer and you are the double checker on three of those. Those are followed by your three consecutive finds with Z-san being double checker on one of them.
Glad to see you two keeping things in the family. :-)
Next time I pursue SGS (for the march towards the Turquoise badge), I will make sure the two of you are busy with some lengthy LLR challenge (sorry RR).
Edit: Going back further ever so slightly, Z-san has a string of ten consecutive SGS finds as primary discoverer. A one-man wrecking crew. Respect! | |
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2165
ID: 1178
Credit: 8,777,295,508
RAC: 0
|
My systems are ready to go. All on GFN16 (GPU)/PPSE (CPU). I should have a pretty clean start.
Me, too. But GFN15 (GPU)/SGS (CPU). Nice side effect is finding 3 SGS and 4 GFN15 primes in the last 30 hours or so. :)
UPDATE: now up to 6 GFN15 finds, all on 12/11!
| |
|
|
|
UPDATE: now up to 6 GFN15 finds, all on 12/11!
Ummm be quiet ! LOL nice job.. you must be killing those things. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
Today is monthly Windows Update day isn't it? So if you haven't blocked those, watch out for reboots, or do the updates manually before the start.
Thanks. I'm going to all the computers and updating them now, before the challenge starts.
The Win 10 boxes seem to be downloading a new version of Windows 10, 1809.
____________
My lucky number is 75898524288+1 | |
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2165
ID: 1178
Credit: 8,777,295,508
RAC: 0
|
UPDATE: now up to 6 GFN15 finds, all on 12/11!
Ummm be quiet ! LOL nice job.. you must be killing those things.
I upgraded the GPU farm considerably over the last couple of months:
GTX 1080 --> RTX 2070
GTX 960 --> GTX 1080
GTX 960 --> GTX 1080
GTX 970 --> GTX 1070
RX 470 --> GTX 1070
Also, I am just getting really lucky on these today.
| |
|
|
|
UPDATE: now up to 6 GFN15 finds, all on 12/11!
Who ate all the primes? :)
Looking on T5K you're "only" ranked 4th by number... so there are 3 more quantity producers out there. One other seems (or has been) obviously active on PrimeGrid.
Just noticed GFN15... so not T5K anyway. Doh! | |
|
|
|
The Win 10 boxes seem to be downloading a new version of Windows 10, 1809.
The notorious 1809 update. MS had a few false starts with that but I have previously manually updated most of my systems at home (the important ones anyway) and haven't had any problems with it. It even seems to break the GPU driver less than previous major updates, but I'd still recommend a reinstall of that just in case. | |
|
|
|
|
@Scott - yes I would say that's some upgrading !
@Mack & @Mike - glad you mentioned updating now. Working on it as well. | |
|
|
|
|
Ant info as to how long the latest update takes? | |
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2165
ID: 1178
Credit: 8,777,295,508
RAC: 0
|
Ant info as to how long the latest update takes?
Under 15 minutes on FX-8350 with 7200rpm drive.
| |
|
|
|
Always go off the small font UTC time at the top of that section. The big counter takes it from your own PC's clock. Correct your clock manually if you need to. Start is 22:23UT.
okay, that explains it -- I was assuming that start was at xx.00 rather than 22.23. :-)
| |
|
|
|
Ant info
Small insect - normally found on ocean bordering countries and loves to infiltrate electrical equipment. | |
|
|
|
Ant info
Small insect - normally found on ocean bordering countries and loves to infiltrate electrical equipment.
Hey you, small typo. :-) | |
|
|
|
|
Out.
All computers dead (won't boot) except for a couple old Macs which can't run GFN. Used all my magic spells to no effect. Time for a trip to the hazmat disposal place. Sorry.
Gary | |
|
|
|
|
Preparations done... now to wait for start time. Initial firepower:
2070, 1070, 1060 3GB, Vega 56, 6700k, 5775C
I'll monitor house temps for the first day before deciding if I can pour more fuel on. | |
|
|
|
Out.
All computers dead (won't boot) except for a couple old Macs which can't run GFN. Used all my magic spells to no effect. Time for a trip to the hazmat disposal place. Sorry.
Gary
oh man I hate that for you !!! | |
|
|
|
Nice side effect is finding 3 SGS and 4 GFN15 primes in the last 30 hours or so. :)
Next time I pursue SGS (for the march towards the Turquoise badge), I will make sure the two of you are busy with some lengthy LLR challenge (sorry RR).
Ha - just found an SGS | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 2829
ID: 130544
Credit: 954,793,678
RAC: 42
|
|
Basically NNT but have 2 21s in progress, 1 will just about finish in time, the other < 10 mins after start.
Sorry to hear about your problems Gary :(. What on earth caused multiple problems at this crucial time? | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 2829
ID: 130544
Credit: 954,793,678
RAC: 42
|
|
Aaand go!
Some higher quora values with these than I've been used to recently. | |
|
|
|
|
Lost the start flag by 7 minutes.
Had to wait for an AP27 task to finish.
____________
"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment. | |
|
|
|
Nice side effect is finding 3 SGS and 4 GFN15 primes in the last 30 hours or so. :)
Next time I pursue SGS (for the march towards the Turquoise badge), I will make sure the two of you are busy with some lengthy LLR challenge (sorry RR).
Ha - just found an SGS
Congrats. Z-san double checker. | |
|
|
|
|
Best of luck everyone! | |
|
|
|
|
After running gfn15's and shorter cpu tasks, my computers reporting in is going to be slow now. Patience, patience and more. Not something I'm good at.
Good luck AtP !!! as Mr Ant Man said :)
Cheers
and again sorry Gary... hate this for you | |
|
|
|
|
I got my machine unpacked, reconnected, and running. I'm impressed by the quality of notes my past self left for me two months ago when I dismantled everything to be moved!
I'm fighting off some errors, and hoping to get stable before going to bed.
I'm hoping to set up one of my rigs this evening to join the Challenge. I'll be making full use of the <dilbert> block.
<dilbert>Out, out!! You demons of stupidity!!</dilbert>
____________
Proud member of Team Aggie the Pew
"Wir müssen wissen. Wir werden wissen."
"We must know, we shall know."
- David Hilbert, 1930 | |
|
|
|
I'm fighting off some errors, and hoping to get stable before going to bed.
Best of luck! | |
|
TimT
Send message
Joined: 2 Dec 11
Posts: 408
ID: 121414
Credit: 1,427,240,743
RAC: 0
                     
|
|
I've been under the weather for the last day or so, and might have missed the starting gun, but I'm in and running now.
btw, has anyone done any testing with gfn multicore on 8-core CPUS? I have mine set to use all 8, just wondering if there is some known reason to change that...
--Tim | |
|
|
|
|
Look on the main challenge thread. Yves reported a 17 thread task. It is impressive.
Sorry on iPad or I'd paste the link. | |
|
|
|
btw, has anyone done any testing with gfn multicore on 8-core CPUS? I have mine set to use all 8, just wondering if there is some known reason to change that...
I don't have one, but I suspect it will be ram bandwidth limited. What's the estimated total run time? 9700k with 3000 ball park ram, I'd guess could be in ball park of 24h-30h. If it could make use of all the cores, could be closer to 12h. If 24h+, you might not see lower speeds reducing the core count.
The 9900k might have had enough cache to speed up a lot, but again this is in theory as I've not tested it. I'm assuming it behaves similar to LLR, which is not guaranteed. | |
|
TimT
Send message
Joined: 2 Dec 11
Posts: 408
ID: 121414
Credit: 1,427,240,743
RAC: 0
|
btw, has anyone done any testing with gfn multicore on 8-core CPUS? I have mine set to use all 8, just wondering if there is some known reason to change that...
I don't have one, but I suspect it will be ram bandwidth limited. What's the estimated total run time? 9700k with 3000 ball park ram, I'd guess could be in ball park of 24h-30h. If it could make use of all the cores, could be closer to 12h. If 24h+, you might not see lower speeds reducing the core count.
The 9900k might have had enough cache to speed up a lot, but again this is in theory as I've not tested it. I'm assuming it behaves similar to LLR, which is not guaranteed.
hmm. my 9700k might have some kind of problem.. I set it to 8 use 8 cores, but its showing 3d time remaining... I might have to go kick that machine once or twice to see whats happening. I do notice in boinctasks, it does not show '8 cores', I may need to reboot or restart boinc
| |
|
|
|
|
What ram does it have? Single/dual channel? Speed?
BTW I couldn't resist, putting 1080Ti on it. I wasn't going to, but it's nearly 3 units a day and to hell with the electricity bill!
Edit: make it 2.5 units... has the task size increased that much in the last days? | |
|
TimT
Send message
Joined: 2 Dec 11
Posts: 408
ID: 121414
Credit: 1,427,240,743
RAC: 0
|
What ram does it have? Single/dual channel? Speed?
BTW I couldn't resist, putting 1080Ti on it. I wasn't going to, but it's nearly 3 units a day and to hell with the electricity bill!
Edit: make it 2.5 units... has the task size increased that much in the last days?
the 9700 has 8 gb total 4200 running dual channel -- I dont recall offhand the true speed, but the CPU's max memory speed was lower than 4200, so it's running at the cpu max I believe, using the XMP profile that was available in BIOS
edit: the machine seemed to be running only one core when I visited it -- restarted boinc, and CPU usage went up to 97%ish, but the boinc time remaining is now 23 days!
edit2: I didnt like the 23 day timer, so I changed the machine to use 4 cores for GFN21. now, the original WU shows as 2+days with timer slowly going up, and the new task shows as 4 days with the timer going down (sometimes a few minutes per tick) -- I also notice that even though I have near 100 percent CPU utilization, it's not producing anywhere near the heat that it had been while running ESP tasks (temp was pretty hot at 80-85C, and now a more reasonable 70C. Unless anyone has ideas for things to try, I'll let it run like this and at least let the timers settle down.
...and yeah, I know I should have played with this before the challenge, but was rushing to finish the next esp badge before it started :) | |
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2165
ID: 1178
Credit: 8,777,295,508
RAC: 0
|
What ram does it have? Single/dual channel? Speed?
BTW I couldn't resist, putting 1080Ti on it. I wasn't going to, but it's nearly 3 units a day and to hell with the electricity bill!
Edit: make it 2.5 units... has the task size increased that much in the last days?
No, I think that is about right. A 2070 is running a bit faster on these than a 1080 Ti and should do 3.1 per day or so. A 1080 is about 2.0 per day depending on clocks.
| |
|
|
|
|
Looks like my trusty 460 might get one done in a tad over 5 days. | |
|
|
|
edit2: I didnt like the 23 day timer, so I changed the machine to use 4 cores for GFN21. now, the original WU shows as 2+days with timer slowly going up, and the new task shows as 4 days with the timer going down (sometimes a few minutes per tick) -- I also notice that even though I have near 100 percent CPU utilization, it's not producing anywhere near the heat that it had been while running ESP tasks (temp was pretty hot at 80-85C, and now a more reasonable 70C. Unless anyone has ideas for things to try, I'll let it run like this and at least let the timers settle down.
...and yeah, I know I should have played with this before the challenge, but was rushing to finish the next esp badge before it started :)
MT GFN seems to be less stressful than MT LLR. I have witnessed the same on my machines. | |
|
mfl0pProject administrator
Volunteer developer
Send message
Joined: 5 Apr 09
Posts: 224
ID: 38042
Credit: 860,159,741
RAC: 66
                  
|
|
i started a little late but should be able to complete about 17 units a day | |
|
|
|
No, I think that is about right. A 2070 is running a bit faster on these than a 1080 Ti and should do 3.1 per day or so. A 1080 is about 2.0 per day depending on clocks.
First units are in from 2070 and 1080Ti.
2070: 29536s (8.2h or 2.9 units/day))
1080Ti: 30783s (8.55h or 2.8 units/day)
Guess the initial estimate was a bit lower than reality.
Temps still seem tolerable at home. Debating if I should put more power on, but anything I haven't already started will take day+/unit. | |
|
|
|
edit: the machine seemed to be running only one core when I visited it -- restarted boinc, and CPU usage went up to 97%ish, but the boinc time remaining is now 23 days!
edit2: I didnt like the 23 day timer, so I changed the machine to use 4 cores for GFN21. now, the original WU shows as 2+days with timer slowly going up, and the new task shows as 4 days with the timer going down (sometimes a few minutes per tick) -- I also notice that even though I have near 100 percent CPU utilization, it's not producing anywhere near the heat that it had been while running ESP tasks (temp was pretty hot at 80-85C, and now a more reasonable 70C. Unless anyone has ideas for things to try, I'll let it run like this and at least let the timers settle down.
...and yeah, I know I should have played with this before the challenge, but was rushing to finish the next esp badge before it started :)
One core could be you not restarting boinc after editing the file. I've fallen for that one before.
You're running two units of 4 cores each now? When I was saying to reduce cores, it was still for running a single unit. Also the estimates will be thrown off from earlier slow running so it'll take a long while to get a true reflection of finish time. | |
|
|
|
I'm fighting off some errors, and hoping to get stable before going to bed.
Best of luck!
Got it running right, and the first batch of WUs is in! Not pretty, but it'll do for now on the floor until the Challenge is over.
____________
Proud member of Team Aggie the Pew
"Wir müssen wissen. Wir werden wissen."
"We must know, we shall know."
- David Hilbert, 1930 | |
|
|
|
|
What's wrong with that? I think it looks rather stylish! I have setups much like that but not nearly as cool looking. I mean open space, plenty of air flow and an easy way to pick it up. What's NOT to like!
Glad you got it running.
Cheers | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 2829
ID: 130544
Credit: 954,793,678
RAC: 42
|
|
Nice to hear the good news. Nothing wrong with hardcore during a challenge, especially one with tasks this big! | |
|
|
|
|
Well we are definitely off to a good start on the challenge!!!! :-) | |
|
|
|
|
I had good intentions to post the 1 day results. Now I am stuck working then I can't from home. Unless someone would like to post them it will be tomorrow.
However yes AtP is doing very well at the moment!!
Go AtP ! | |
|
|
|
|
OK, here it goes.
After Day 1:
Last update: 2018-12-12 23:15:03
Rank Name Team Score
5 Van Zimmerman Aggie The Pew 3,883,393.84
9 mfl0p Aggie The Pew 2,543,113.35
11 yank Aggie The Pew 2,241,583.37
18 Scott Brown Aggie The Pew 1,792,776.93
23 mackerel Aggie The Pew 1,345,150.33
28 SkyHighWeFly Aggie The Pew 1,194,968.68
45 Renix1943 Aggie The Pew 896,021.12
47 shinozk Aggie The Pew 747,977.59
54 zunewantan Aggie The Pew 598,327.15
72 Rick Reynolds Aggie The Pew 448,029.39
91 TimT Aggie The Pew 298,936.20
96 mattozan Aggie The Pew 298,681.29
114 Brook Harste Aggie The Pew 149,659.13
133 Eudy Silva Aggie The Pew 149,463.39
149 Michael Goetz Aggie The Pew 149,295.69
Rank Name Score
1 Sicituradastra. 18,088,000.00
2 Aggie The Pew 16,737,377.45
3 SETI.Germany 13,596,214.21
4 Czech National Team 12,706,966.40
5 Crunching@EVGA 9,270,732.47
6 AMD Users 7,783,612.16
7 TeAm AnandTech 6,128,815.80
8 Rechenkraft.net 4,033,498.83
9 Storm 3,435,641.34
10 BOINC@MIXI 3,286,140.36
____________
"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment. | |
|
|
|
|
@Eudy wonderful. Thank you | |
|
|
|
|
Good start for the group. Keep it going! | |
|
|
|
|
Nice little grouping
56 zunewantan Aggie The Pew 1495681.51
57 TimT Aggie The Pew 1495625.25
58 Rick Reynolds Aggie The Pew 1494443.01 | |
|
|
|
|
Stats after Day 2:
Last update: 2018-12-13 22:30:04
Rank Name Team Score
5 Van Zimmerman Aggie The Pew 8,972,941.51
10 yank Aggie The Pew 5,686,455.62
13 mfl0p Aggie The Pew 5,388,663.95
16 Scott Brown Aggie The Pew 5,085,540.48
28 mackerel Aggie The Pew 3,140,582.44
37 SkyHighWeFly Aggie The Pew 2,842,284.72
46 Renix1943 Aggie The Pew 2,093,357.96
47 zunewantan Aggie The Pew 1,944,700.03
52 shinozk Aggie The Pew 1,646,785.76
57 TimT Aggie The Pew 1,495,625.25
59 Rick Reynolds Aggie The Pew 1,494,443.01
85 ILW8 Aggie The Pew 898,481.60
124 mattozan Aggie The Pew 597,811.21
130 Dave Aggie The Pew 596,894.09
143 Brook Harste Aggie The Pew 448,683.78
167 Dirk Aggie The Pew 299,267.66
169 Eudy Silva Aggie The Pew 299,253.43
181 Michael Goetz Aggie The Pew 298,951.47
Rank Name Score
1 Sicituradastra. 46,534,115.15
2 Aggie The Pew 42,781,828.99
3 SETI.Germany 40,523,666.63
4 Czech National Team 38,439,526.12
5 Crunching@EVGA 23,346,528.13
6 AMD Users 16,773,986.23
7 TeAm AnandTech 14,365,843.07
8 Rechenkraft.net 12,258,606.72
9 BOINC@MIXI 11,511,317.44
10 Storm 10,919,720.52
____________
"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment. | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 2829
ID: 130544
Credit: 954,793,678
RAC: 42
|
|
Name Score
1 Aggie The Pew 57447182.74
2 Sicituradastra. 56721179.05
Floor it!! | |
|
|
|
|
This one looks like it is going to be quite the race. | |
|
|
|
|
Great seeing the good guys in first place overall. Really good effort! | |
|
|
|
|
Looked like Z-san dropped several units and everyone is bringing it ! I have one pc that's running very slowly but don't want to mess with it. But turning in cpu units is making a big difference in credit and standing.
Might be one of those challenges where the last few units turned in make the difference. Hoping the timing is right.
Go AtP ! | |
|
|
|
But turning in cpu units is making a big difference in credit and standing.
If someone was crunching simply on the basis of CPU RAC alone, I think that Genefer is the best project to this effect. It even beats sieving.
I am going by memory but a good portion of the credit (if not all of it) for my Sapphire GFN badge was a result of running GFN20's on my CPU's.
Edit: When the time comes to upgrade my Sapphire GFN badge, I foresee myself having all of my systems running GFN21's (CPU). | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
If someone was crunching simply on the basis of CPU RAC alone, I think that Genefer is the best project to this effect. It even beats sieving.
Only for GFN-21. Everything lower than that must use the x87 transform, which is at least ten times slower and therefore has ten times less RAC.
...my Sapphire GFN badge was a result of running GFN20's on my CPU's.
GFN-20 switched to x87 about a year and a half ago, so if you were getting good GFN-20 CPU credits, it was before July 2017. I would not recommend running GFN-20 on CPUs today. On a CPU, GFN-20 takes over 5 times as long to run but yields less than a quarter of the credit.
____________
My lucky number is 75898524288+1 | |
|
|
|
If someone was crunching simply on the basis of CPU RAC alone, I think that Genefer is the best project to this effect. It even beats sieving.
Only for GFN-21. Everything lower than that must use the x87 transform, which is at least ten times slower and therefore has ten times less RAC.
...my Sapphire GFN badge was a result of running GFN20's on my CPU's.
GFN-20 switched to x87 about a year and a half ago, so if you were getting good GFN-20 CPU credits, it was before July 2017. I would not recommend running GFN-20 on CPUs today. On a CPU, GFN-20 takes over 5 times as long to run but yields less than a quarter of the credit.
Many thanks for the clarification and insights. I am quite certain I flipped to Sapphire well over 18 months ago and the switchover point you describe.
GFN21 (CPU) will be my project of choice when the times comes to go after another shield. | |
|
|
|
|
Here's another nice grouping.
11 Scott Brown Aggie The Pew 8679260.73
12 mfl0p Aggie The Pew 8086195.94
13 yank Aggie The Pew 8084610.62 | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 2829
ID: 130544
Credit: 954,793,678
RAC: 42
|
Here's another nice grouping.
11 Scott Brown Aggie The Pew 8679260.73
12 mfl0p Aggie The Pew 8086195.94
13 yank Aggie The Pew 8084610.62
Looks like a mega fl0ppy sandwich to me. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
|
Copied from the main challenge thread:
Michael Goetz wrote: Furthermore, there's actually no need for the <app> section. All you need is the <app_version> part. This is what I use:
<app_config>
<app_version>
<app_name>genefer</app_name>
<cmdline>-nt 4</cmdline>
<plan_class>cpuGFN21</plan_class>
<avg_ncpus>4</avg_ncpus>
</app_version>
</app_config>
EDIT: That's for 4 cores. Change the numbers as necessary.
After a bit of experimentation, I'm going to change that.
This is what I now recommend:
<app_config>
<report_results_immediately/>
<app>
<name>genefer</name>
<fraction_done_exact/>
</app>
<app_version>
<app_name>genefer</app_name>
<cmdline>-nt 4</cmdline>
<plan_class>cpuGFN21</plan_class>
<avg_ncpus>4</avg_ncpus>
</app_version>
</app_config>
I added two things.
I added the app section because it seems we need to explicitly specify <fraction_done_exact/> for each app in app_config. The server itself sets this flag, but it appears that if an app appears in app_config it ignores what the server says. This flag improves the accuracy of BOINC's time estimates.
I also added <report_results_immediately/>, which goes outside of any <app> or <app_version> block. You only need this once. I'm not actually sure that this is necessary, but it won't hurt. This too is set on the server, but if the BOINC client is ignoring the server's <fraction_done_exact/> perhaps it's also ignoring the server's <report_results_immediately/>. By including it explicitly you can insure it's turned on.
____________
My lucky number is 75898524288+1 | |
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2165
ID: 1178
Credit: 8,777,295,508
RAC: 0
|
Here's another nice grouping.
11 Scott Brown Aggie The Pew 8679260.73
12 mfl0p Aggie The Pew 8086195.94
13 yank Aggie The Pew 8084610.62
Looks like a mega fl0ppy sandwich to me.
Hmm...now I am in the middle:
11 yank Aggie The Pew 8834896.42
12 Scott Brown Aggie The Pew 8828812.71
13 mfl0p Aggie The Pew 8236212.63
I feel like I'm being yanked and flopped! ;P
| |
|
|
|
Here's another nice grouping.
11 Scott Brown Aggie The Pew 8679260.73
12 mfl0p Aggie The Pew 8086195.94
13 yank Aggie The Pew 8084610.62
Looks like a mega fl0ppy sandwich to me.
Hmm...now I am in the middle:
11 yank Aggie The Pew 8834896.42
12 Scott Brown Aggie The Pew 8828812.71
13 mfl0p Aggie The Pew 8236212.63
I feel like I'm being yanked and flopped! ;P
https://www.youtube.com/watch?v=8BF95hdIuOg&feature=youtu.be | |
|
|
|
Here's another nice grouping.
11 Scott Brown Aggie The Pew 8679260.73
12 mfl0p Aggie The Pew 8086195.94
13 yank Aggie The Pew 8084610.62
Looks like a mega fl0ppy sandwich to me.
Hmm...now I am in the middle:
11 yank Aggie The Pew 8834896.42
12 Scott Brown Aggie The Pew 8828812.71
13 mfl0p Aggie The Pew 8236212.63
I feel like I'm being yanked and flopped! ;P
https://www.youtube.com/watch?v=8BF95hdIuOg&feature=youtu.be
Oh man - ya'll are just too funny! | |
|
|
|
|
Brook Harste wrote: https://www.youtube.com/watch?v=8BF95hdIuOg&feature=youtu.be
Mel Brooks Harste :)
____________
"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment. | |
|
|
|
|
After Day 3:
Last update: 2018-12-14 23:30:05
Rank Name Team Score
5 Van Zimmerman Aggie The Pew 16,168,255.28
8 zunewantan Aggie The Pew 12,111,787.38
11 Scott Brown Aggie The Pew 9,578,030.58
12 yank Aggie The Pew 8,984,796.07
17 mfl0p Aggie The Pew 8,236,212.63
34 mackerel Aggie The Pew 5,240,060.73
37 SkyHighWeFly Aggie The Pew 4,791,689.02
52 Renix1943 Aggie The Pew 3,142,503.94
62 shinozk Aggie The Pew 2,544,717.62
63 Rick Reynolds Aggie The Pew 2,542,947.38
64 TimT Aggie The Pew 2,393,715.60
91 ILW8 Aggie The Pew 1,498,103.26
122 Dave Aggie The Pew 1,046,368.37
138 mattozan Aggie The Pew 897,181.31
153 Brook Harste Aggie The Pew 748,364.81
169 Dirk Aggie The Pew 599,048.10
179 Michael Goetz Aggie The Pew 598,155.50
192 Eudy Silva Aggie The Pew 449,206.74
238 Kouhki Aggie The Pew 149,987.07
Rank Name Score
1 Aggie The Pew 81,272,236.44
2 Sicituradastra. 79,949,649.12
3 Czech National Team 69,280,217.06
4 SETI.Germany 68,984,713.62
5 Crunching@EVGA 38,938,668.04
6 AMD Users 25,772,978.05
7 TeAm AnandTech 23,659,010.58
8 Rechenkraft.net 20,193,761.94
9 BOINC@MIXI 18,849,652.50
10 Storm 18,412,425.91
19 crunching rats
AtP leading !
____________
"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment. | |
|
TimT
Send message
Joined: 2 Dec 11
Posts: 408
ID: 121414
Credit: 1,427,240,743
RAC: 0
|
|
I was just looking at the first task that finally finished on my shiny new 9700k and it's not at all what I expect. Does anyone have ideas of what I should look at to improve the performance?
here's the result. You can see where I re-started the task several times with different core counts, and very different estimated compute times that are confusing.
http://www.primegrid.com/result.php?resultid=952930522
overall, I expect that the 9700k should perform faster than my other CPUs, but that's not the case for some reason:
9700k: 2d 18:43:15 (7 cores mostly, 1x1060GPU)
7700k: 1d 19:35:26 (4 cores, HT Off, 2x1060GPUs)
4790k (estimate): 2d 20h (4 cores, HT Off)
Even the GTX1060 GPU in the 9700k system is performing slower than the 1060s in my 7700k system (~20 hours vs 27-32 hours per task) | |
|
|
|
I was just looking at the first task that finally finished on my shiny new 9700k and it's not at all what I expect. Does anyone have ideas of what I should look at to improve the performance?
The task is not visible and, presumably, the double checker has yet to report his result.
What has me puzzled is why the i7 6700k is more efficient than the i7 7700k? I would have expected the Kaby Lake to be faster than the Skylake but this is not the case with the Kaby Lake requiring approximately 30% more elapsed time (with "nt -4"). There may be system differences at play (ex speed of memory etc) but I do not know since my observations are based on someone else's results. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
You can see where I re-started the task several times with different core counts, and very different estimated compute times that are confusing.
I have observed that the time estimate from the CPU version of Genefer, at least when multi-threading, seems to have "issues". I wouldn't put much faith in that estimate. It's something we need to look into.
The GPU version of Genefer produces rock-solid estimates, however.
____________
My lucky number is 75898524288+1 | |
|
|
|
overall, I expect that the 9700k should perform faster than my other CPUs, but that's not the case for some reason:
9700k: 2d 18:43:15 (7 cores mostly, 1x1060GPU)
7700k: 1d 19:35:26 (4 cores, HT Off, 2x1060GPUs)
4790k (estimate): 2d 20h (4 cores, HT Off)
Even the GTX1060 GPU in the 9700k system is performing slower than the 1060s in my 7700k system (~20 hours vs 27-32 hours per task)
Earlier you said you had 4200 ram in, can you double check it is running at that? For example, you can use CPU-Z to report on running ram speed, but note it reports true MHz, which is exactly half marketing MHz. e.g. for 4200 ram you should see it report 2100. Also note if it says it is running dual channel or not.
These units are likely ram limited on Intel quad cores and up (AMD cores are much slower so you could run much more before limiting). If you're running 4200 dual channel, single rank ram, I'd estimate times in the ball park of 23 hours as long as you're running at least 4 cores. If the ram was left at base speeds like 2133, I'd estimate around 45 hours.
As a suggestion, while a unit is running, also run CPU-Z and screenshot the CPU, mainboard, memory and SPD tabs. For SPD, you might have to manually select a slot that has ram in it so it shows the info. Upload it somewhere and link to it. That'll help give a picture of what's going on.
Note I never tested a mixed CPU+GPU environment, so I don't know if GPU has much of an impact.
Also for comparison, I have a 1060 3GB, and that does units in 22h or less. Mine has a rubbish cooler on it though, and it is constantly on thermal limiter. A better cooled one could likely go faster. | |
|
|
|
|
After Day 4:
Last update: 2018-12-15 22:30:06
Rank Name Team Score
5 Van Zimmerman Aggie The Pew 20,821,529.55
9 Scott Brown Aggie The Pew 16,018,579.96
11 zunewantan Aggie The Pew 14,658,815.76
12 yank Aggie The Pew 12,287,070.67
16 mfl0p Aggie The Pew 11,087,153.52
36 mackerel Aggie The Pew 7,041,911.13
39 SkyHighWeFly Aggie The Pew 6,443,346.08
55 Renix1943 Aggie The Pew 4,191,741.81
59 Rick Reynolds Aggie The Pew 3,891,937.30
62 TimT Aggie The Pew 3,593,881.99
64 shinozk Aggie The Pew 3,295,708.11
87 ILW8 Aggie The Pew 2,547,307.55
123 Dave Aggie The Pew 1,495,676.63
144 mattozan Aggie The Pew 1,196,639.27
150 Brook Harste Aggie The Pew 1,048,668.11
159 Dirk Aggie The Pew 899,137.99
170 Michael Goetz Aggie The Pew 897,491.42
194 Eudy Silva Aggie The Pew 599,340.12
207 Kouhki Aggie The Pew 450,154.12
291 Nilay Khandelwal Aggie The Pew 149,738.89
316 Deano Montreuil Aggie The Pew 149,537.04
Rank Name Score
1 Aggie The Pew 112,316,472.02
2 Sicituradastra. 112,208,463.64
3 Czech National Team 98,053,052.30
4 SETI.Germany 96,572,372.15
5 Crunching@EVGA 52,593,441.81
6 TeAm AnandTech 33,266,034.34
7 AMD Users 31,479,610.04
8 Rechenkraft.net 29,187,029.62
9 BOINC@MIXI 26,350,114.96
10 Storm 24,564,913.58
21 AtP members are crunching
____________
"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment. | |
|
|
|
|
Wow great participation so far, we are in for a dawg fight it appears. Anyone care to jump in the deep end? Would love to see ATP win the last challenge of the year.
Go ATP! | |
|
TimT
Send message
Joined: 2 Dec 11
Posts: 408
ID: 121414
Credit: 1,427,240,743
RAC: 0
|
Earlier you said you had 4200 ram in, can you double check it is running at that?
I think you hit the nail on the head there -- I double checked my BIOS settings, and indeed, I told it to load the xmp2.0 profile for 4200 ram, the rest of the gazillion settings were left on 'auto' -- however, cpu-z shows it running at some slow default speed.
After playing around far longer than I planned, I believe I found that you have to select the XMP profile, but also have to select the frequency that you want to run at (at least this is my take given my experience) -- I found the memory was running at its default of roughly 1800, so I did a few tests, increasing the frequency gradually, doing a quick benchmark / test for each. When I finally got up to 4200, I saw an error message at POST, and the machine again had memory running at the default speed, so one last time, I selected the speed just under 4200 and it worked. I'll declare this 'close enough' -- I will monitor it for the next day or so and see if there are any issues. | |
|
|
|
I think you hit the nail on the head there -- I double checked my BIOS settings, and indeed, I told it to load the xmp2.0 profile for 4200 ram, the rest of the gazillion settings were left on 'auto' -- however, cpu-z shows it running at some slow default speed.
After playing around far longer than I planned, I believe I found that you have to select the XMP profile, but also have to select the frequency that you want to run at (at least this is my take given my experience) -- I found the memory was running at its default of roughly 1800, so I did a few tests, increasing the frequency gradually, doing a quick benchmark / test for each. When I finally got up to 4200, I saw an error message at POST, and the machine again had memory running at the default speed, so one last time, I selected the speed just under 4200 and it worked. I'll declare this 'close enough' -- I will monitor it for the next day or so and see if there are any issues.
It is normal for a mobo to try and auto-recover if it fails to boot. Normally it should give a warning and give you the option of going into bios to fix it before booting, rather than silently revert back to safe speeds and boot anyway.
4200 ram is VERY fast. Compatibility is going to be a minefield at that speed. Depending on the ram module manufacturer, some put in two XMP profiles, one a bit slower to increase the chance of compatibility. Even if you can boot manually selecting a lower speed, that doesn't mean it is stable. | |
|
|
|
After Day 4:
Rank Name Score
1 Aggie The Pew 112,316,472.02
2 Sicituradastra. 112,208,463.64
The difference is less than ONE work unit, right?! | |
|
|
|
|
It seems so.
One task is worth around 149k - 150k credits.
____________
"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment. | |
|
|
|
It seems so.
One task is worth around 149k - 150k credits.
I will change the cache on all of my systems in the event we need a last-minute push.
Edit: They passed us ... 800k difference right now or about six units. | |
|
|
|
|
It's rather fun having a close race for the end of the year.
Go ATP! | |
|
|
|
|
We've passed them
Last update: 2018-12-16 05:00:05
Rank Name Score
1 Aggie The Pew 121,920,460.90
2 Sicituradastra. 121,216,837.25
____________
"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment. | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 2829
ID: 130544
Credit: 954,793,678
RAC: 42
|
|
Wowsers. 1 unit is 149k so I can't see how a difference between teams can be ~108k. Hmm maybe it's 'drift' becuase the overall amassed credit is so high.
My rate of processing will be slightly reduced today but I do not expect it to have any fundamental impact i.e just 1 GPU will be paused for 13 hours & I may amend app_config to free up 1 core. | |
|
yank
Send message
Joined: 14 May 07
Posts: 111
ID: 8367
Credit: 11,474,812,476
RAC: 0
         
|
|
GREAT RACE.
____________
 | |
|
|
|
|
yup, really a closely matched race. You can bet folks on each team are trying to maximize their output. Really a fun challenge! :-) | |
|
|
|
After Day 4:
21 AtP members are crunching
Is this from the extended version of "The 12 Days of Christmas?"
"21 members crunching,
20 fans a-churning,
19 keyboard tapping,
18 GPUs smoking..."
____________
Proud member of Team Aggie the Pew
"Wir müssen wissen. Wir werden wissen."
"We must know, we shall know."
- David Hilbert, 1930 | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
After Day 4:
21 AtP members are crunching
Is this from the extended version of "The 12 Days of Christmas?"
"21 members crunching,
20 fans a-churning,
19 keyboard tapping,
18 GPUs smoking..."
"... and an Amdahl in a B-Tree."
____________
My lucky number is 75898524288+1 | |
|
|
|
After Day 4:
21 AtP members are crunching
Is this from the extended version of "The 12 Days of Christmas?"
"21 members crunching,
20 fans a-churning,
19 keyboard tapping,
18 GPUs smoking..."
"... and an Amdahl in a B-Tree."
That is some good stuff there! Now I'm going to have that tune in my head all day. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
|
A CALL TO ARMS
I'm not sure how many of you have noticed, but the top two people on the challenge leaderboard are up there primarily because they're using Amazon AWS p3 servers with the insane Tesla V100-XSM2-16GB GPUs. These are server-only, special form-factor GPUs that are a lot faster than the PCIE version of the GPU. According to our fastest GPUs page, an RTX 2080 Ti is only about 62% as fast as these beasts.
GFN-21 takes about 3 hours and 20 minutes.
At the best possible pricing these servers cost less than $1 per hour to rent.
As it turns out, neither of these two are members of a team that's competing for the top team position, but it's possible, even likely, that other teams have noticed this. I know for a fact that SETI.Germany has spotted it. It's therefore possible that some people and some teams can greatly increase their performance for the second part of the challenge simply by throwing a few bucks at Amazon.
So don't get complacent because we're in the lead. As of right now, we're up by about a million points. That's only 7 tasks -- which $25 and 200 minutes at AWS can erase.
We have a chance to win this, but we really need every task we can get. If you've got anything that can complete a GFN21 before the challenge ends, now's the time to crank it up.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
After Day 5 of 10:
Last update: 2018-12-16 22:30:04
Rank Name Team Score
5 Van Zimmerman Aggie The Pew 26,530,942.96
7 zunewantan Aggie The Pew 23,056,677.86
9 Scott Brown Aggie The Pew 20,971,610.52
13 yank Aggie The Pew 16,642,540.93
21 mfl0p Aggie The Pew 13,790,381.41
35 SkyHighWeFly Aggie The Pew 8,996,698.36
36 mackerel Aggie The Pew 8,996,291.79
55 Renix1943 Aggie The Pew 5,393,560.39
61 Rick Reynolds Aggie The Pew 4,642,443.99
63 TimT Aggie The Pew 4,492,208.89
67 shinozk Aggie The Pew 4,198,048.89
91 ILW8 Aggie The Pew 2,998,464.45
121 Dave Aggie The Pew 1,945,947.05
147 mattozan Aggie The Pew 1,497,222.02
153 Dirk Aggie The Pew 1,349,982.38
157 Brook Harste Aggie The Pew 1,349,400.57
173 Michael Goetz Aggie The Pew 1,197,636.15
206 Eudy Silva Aggie The Pew 748,132.91
211 Kouhki Aggie The Pew 600,535.03
262 Nilay Khandelwal Aggie The Pew 299,932.76
349 Deano Montreuil Aggie The Pew 149,537.04
Rank Name Score
1 Sicituradastra. 150,805,218.11
2 Aggie The Pew 149,399,301.39
3 SETI.Germany 132,579,716.65
4 Czech National Team 131,799,113.47
5 Crunching@EVGA 65,513,565.70
6 TeAm AnandTech 49,646,372.25
7 Rechenkraft.net 39,537,532.91
8 AMD Users 37,042,946.41
9 BOINC@MIXI 35,209,948.19
10 Storm 31,325,544.87
____________
"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment. | |
|
|
|
A CALL TO ARMS
I'm not sure how many of you have noticed, but the top two people on the challenge leaderboard are up there primarily because they're using Amazon AWS p3 servers with the insane Tesla V100-XSM2-16GB GPUs. These are server-only, special form-factor GPUs that are a lot faster than the PCIE version of the GPU. According to our fastest GPUs page, an RTX 2080 Ti is only about 62% as fast as these beasts.
GFN-21 takes about 3 hours and 20 minutes.
At the best possible pricing these servers cost less than $1 per hour to rent.
As it turns out, neither of these two are members of a team that's competing for the top team position, but it's possible, even likely, that other teams have noticed this. I know for a fact that SETI.Germany has spotted it. It's therefore possible that some people and some teams can greatly increase their performance for the second part of the challenge simply by throwing a few bucks at Amazon.
So don't get complacent because we're in the lead. As of right now, we're up by about a million points. That's only 7 tasks -- which $25 and 200 minutes at AWS can erase.
We have a chance to win this, but we really need every task we can get. If you've got anything that can complete a GFN21 before the challenge ends, now's the time to crank it up.
Noted. I tried to spawn a P3.2xlarge instance, but BOINC kept saying "No coprocessors found."
I used this AMI:
suse-sles-11-sp4-v20180816-hvm-ssd-x86_64 (ami-09a652af9ed1e0dad)
And I followed these copy/paste lines to install the CUDA drivers, then rebooted:
sudo zypper install boinc-client
wget http://developer.download.nvidia.com/compute/cuda/repos/sles11/x86_64/cuda-repo-sles11-5.5-0.x86_64.rpm
sudo rpm --install cuda-repo-sles11-5.5-0.x86_64.rpm
sudo zypper refresh
sudo zypper install cuda
export PATH=/usr/local/cuda-5.5/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-5.5/lib64:$LD_LIBRARY_PATH
sudo zypper install cuda-cross
cd /usr/local/cuda-5.5/
cuda-install-samples-5.5.sh ~
cd ~
Still no coprocessors in BOINC.
Can anyone tell me what I'm doing wrong?
____________
Proud member of Team Aggie the Pew
"Wir müssen wissen. Wir werden wissen."
"We must know, we shall know."
- David Hilbert, 1930 | |
|
JamesProject administrator
Volunteer tester
Send message
Joined: 19 Sep 14
Posts: 95
ID: 366225
Credit: 523,713,437
RAC: 0
   
|
A CALL TO ARMS
I'm not sure how many of you have noticed, but the top two people on the challenge leaderboard are up there primarily because they're using Amazon AWS p3 servers with the insane Tesla V100-XSM2-16GB GPUs. These are server-only, special form-factor GPUs that are a lot faster than the PCIE version of the GPU. According to our fastest GPUs page, an RTX 2080 Ti is only about 62% as fast as these beasts.
GFN-21 takes about 3 hours and 20 minutes.
At the best possible pricing these servers cost less than $1 per hour to rent.
As it turns out, neither of these two are members of a team that's competing for the top team position, but it's possible, even likely, that other teams have noticed this. I know for a fact that SETI.Germany has spotted it. It's therefore possible that some people and some teams can greatly increase their performance for the second part of the challenge simply by throwing a few bucks at Amazon.
So don't get complacent because we're in the lead. As of right now, we're up by about a million points. That's only 7 tasks -- which $25 and 200 minutes at AWS can erase.
We have a chance to win this, but we really need every task we can get. If you've got anything that can complete a GFN21 before the challenge ends, now's the time to crank it up.
Noted. I tried to spawn a P3.2xlarge instance, but BOINC kept saying "No coprocessors found."
I used this AMI:
suse-sles-11-sp4-v20180816-hvm-ssd-x86_64 (ami-09a652af9ed1e0dad)
And I followed these copy/paste lines to install the CUDA drivers, then rebooted:
sudo zypper install boinc-client
wget http://developer.download.nvidia.com/compute/cuda/repos/sles11/x86_64/cuda-repo-sles11-5.5-0.x86_64.rpm
sudo rpm --install cuda-repo-sles11-5.5-0.x86_64.rpm
sudo zypper refresh
sudo zypper install cuda
export PATH=/usr/local/cuda-5.5/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-5.5/lib64:$LD_LIBRARY_PATH
sudo zypper install cuda-cross
cd /usr/local/cuda-5.5/
cuda-install-samples-5.5.sh ~
cd ~
Still no coprocessors in BOINC.
Can anyone tell me what I'm doing wrong?
Is there a specific reason why you chose this AMI and manually installing CUDA? There are some AMIs out there pre-made for machine learning which includes CUDA pre-installed. Perhaps you could try that instead?
As for your installation, it seems right to me. | |
|
|
|
We have a chance to win this, but we really need every task we can get. If you've got anything that can complete a GFN21 before the challenge ends, now's the time to crank it up.
Following this request, I thought I would satisfy my curiosity and look up the performance of the Tesla GPU. I came across a mining video with 8 Tesla 100 GPU's using AWS.
In the video, the fellow paid $30 for 24 hrs worth of mining using the 8 Tesla GPU's.
I do not remember the date of the video but (1) if the price today is similar (or lower) and (2) if someone can look after the software installation for me, I would be willing to spend the $30 for 24 hrs to run GFN21 across the 8 GPU configuration available at AWS. Doing some very rough calculations in my head, that should translate to 10 million credits or about 60 units and which should secure first place for us.
Off to find the video I saw yesterday on youtube ....
Edit: CANCEL! I just found the video and the fellow quotes $30 per hour: https://www.youtube.com/watch?v=t_0uHQKOqeU
Edit 2: Current pricing for the 8-GPU option is at $24.48/hr as per https://aws.amazon.com/ec2/pricing/on-demand/ (see plan "p3.16xlarge") | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
Edit: CANCEL! I just found the video and the fellow quotes $30 per hour: https://www.youtube.com/watch?v=t_0uHQKOqeU
Edit 2: Current pricing for the 8-GPU option is at $24.48/hr as per https://aws.amazon.com/ec2/pricing/on-demand/ (see plan "p3.16xlarge")
There's three different pricing levels with each server (on-demand and two flavors of spot pricing) which vary from less than $1 per hour to about $3 per hour for the single GPU server. It depends on how you feel about your server suddenly vanishing because someone else is willing to pay more for it. It's a nuisance to have to rebuild the server if you lose it -- but it's a third the cost. I know people who have automated the entire process, including purchasing the servers, to run fleets of spot-pricing AWS servers at very low cost. When they lose a server, they just create another one automatically.
Or you can pay $3 and "own" it until you decide to give it up. You can also pay $1.60 and reserve it for up to 6 hours.
Also, check all the zones. The prices differ.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
This is something I half considered before in the past. Don't know if still current, but at the time I understood:
* You can get shared persisting storage as an option, which could be used to checkpoint and not lose work, especially for longer units.
* They don't just kill your instance instantly if you get outbid, but you get a short notice. If this could be detected, you could shut down gracefully rather than rely on checkpoints.
* Ideal scenario would then be set a price level, and allow some management to open/close instances according to price.
* For a challenge, you might up the price to keep running.
I probably should revisit this as I'm sure pricing has changed. At least, my unit cost for electricity has gone up 50% in last two years. Is cloud computing cheaper than buying and running my own hardware? Putting aside for now I have no idea how to do any of the software things above... | |
|
|
|
|
Silly question to the group: is anyone running both CPU and GPU units simultaneously on the same system?
I should be able to produce 1.5 million credits per day if I dump everything I have onto GFN21 but I do not know what kind of RAM bandwidth issues there will be. My mobo's are restricted to 2133MHz using dual channel RAM (2x4Gb).
I am about 4 hrs away from hitting 80M confirmed credit with GCW Sieve and, conceivably, I could park for the duration of the challenge (four days) by picking up GFN21. I could make these days up during TdP so as not to endanger my Mar 19-20, 2019 finish (100M). | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
Is cloud computing cheaper than buying and running my own hardware?
It depends on what you figure into the cost. Running new electrical lines from the power company, dedicated air conditioning, building one or more rooms specifically for the servers, etc.
I've seen servers like these on ebay for about $25K. I'm guessing they need a special, dedicated circuit -- I doubt you can't run it off a standard US 120V 15A or 20A circuit.
If you're looking at the 'mere mortal' version (i.e., PCIE) of the GPU, you're probably better off getting the much less expensive RTX 2080 Ti, which seems to have similar performance. My guess would be that buying your own would be much less expensive, unless you only want the computing power for a very short period of time, or you want so many of them that keeping them at home just isn't practical.
I see this kind of cloud computing as a "surge" resource, to give you that extra push to get you over the hump. But there are some people who run them for extended periods of time.
____________
My lucky number is 75898524288+1 | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
Silly question to the group: is anyone running both CPU and GPU units simultaneously on the same system?
Pretty much everyone is. :)
Just kidding! Seriously, though, a lot of people are doing that. I am. Using multi-threading, the CPU can do these tasks in a reasonable amount of time. The trick is figuring out what's most efficient, running all the cores, leaving a core free, splitting the CPU to run more than one task, etc.
I should be able to produce 1.5 million credits per day if I dump everything I have onto GFN21 but I do not know what kind of RAM bandwidth issues there will be. My mobo's are restricted to 2133MHz using dual channel RAM (2x4Gb).
I ran benchmarks before the challenge to see what worked best, and to see how much contention there was between the CPU and GPU tasks. Short answer: There's some. I don't have a really good rule of thumb for what's best in a generic system.
On this computer (Haswell i5-4670K, GTX 1060) I decided to run 3-threaded GFN on the CPU and GFN on the GPU, leaving a core free for the GPU. At the halfway mark, I switched to running all 4 threads on the CPU. I might be able to pick up one extra CPU task before the challenge ends by doing that, and I probably won't lose a GPU task. This is due to the exact timing of the tasks on this system: I'll complete 11 GPU tasks, but I have a lot of leeway for those 11, so slowing down a GPU a little won't stop me from finishing the 11th task on time. On the other hand, the CPU is going to do 4 tasks, and isn't that much short of being able to do 5 tasks, and that extra core cuts a significant number of hours off each task.
As you can see, I had to go down to the level of "This task takes X hours and I have Y hours remaining" in order to make the decision of how to set it up. I doubt anyone can tell you what's going to work best on your computer.
____________
My lucky number is 75898524288+1 | |
|
|
|
On this computer (Haswell i5-4670K, GTX 1060) I decided to run 3-threaded GFN on the CPU and GFN on the GPU, leaving a core free for the GPU.
If the logical cores are sufficient to look after the necessary CPU cycles needed by the GPU, my question would be moot since all of my processors do have HT available, thus potentially being able to supply cycles to the GPU while the all of the physical cores are busy.
My mistake for totally ignoring this challenge and not doing some testing earlier! | |
|
|
|
|
Assuming both groups are dumping units at a stable rate (ex no Z-san bombs), we have a small problem with the given spread at this point.
1 Sicituradastra. 179213343.63
2 Aggie The Pew 175400479.59
| |
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2165
ID: 1178
Credit: 8,777,295,508
RAC: 0
|
Assuming both groups are dumping units at a stable rate (ex no Z-san bombs), we have a small problem with the given spread at this point.
1 Sicituradastra. 179213343.63
2 Aggie The Pew 175400479.59
I am running with HT on machines (4 or 6 real cores crunching with HT free for the GPU). The effect on GPU times seems minimal in those cases.
As for the stable dumping rate, both Z-san and I are CPU heavy, so the rates are far from constant. :) | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 2829
ID: 130544
Credit: 954,793,678
RAC: 42
|
Silly question to the group: is anyone running both CPU and GPU units simultaneously on the same system?
Yes.
Icould park for the duration of the challenge (four days) by picking up GFN21.
Please do. Floor it! | |
|
|
|
Icould park for the duration of the challenge (four days) by picking up GFN21.
Please do. Floor it!
I just modified my preferences and I will start the GPU units within a few minutes. The CPUs will kick in about 90 minutes and when my GCW Sieve reaches 80M confirmed credits.
Hopefully my contribution at this stage will prove to be meaningful. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 903
|
Icould park for the duration of the challenge (four days) by picking up GFN21.
Please do. Floor it!
I just modified my preferences and I will start the GPU units within a few minutes. The CPUs will kick in about 90 minutes and when my GCW Sieve reaches 80M confirmed credits.
Hopefully my contribution at this stage will prove to be meaningful.
Every bit counts, thanks!
This is especially true for this challenge. Despite the large numbers of credits, the actual number of tasks is very low. Each additional task is important.
____________
My lucky number is 75898524288+1 | |
|
|
