Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Problems and Help :
New monster PC too slow!
| Author |
Message |
|---|
|
|
|
I've just built a new PC on Intel's X299 platform running an i9-7900 CPU, and was hoping this would be a killer number cruncher... but I'm not at all happy with the performance.
Samsung 960 M.2 drive, 32G RAM, GEForce 1070 GPU, and have overclocked to 4.4GHz. I've set the BOINC Manager parameters to 50% CPU, up to 100% of the time. The trouble is, my old i7-4771 is crunching faster!
Times for ESP are over 2 days! I had BOINC Manager rerun the CPU benchmarks, and the estimated times got worse!!
Clearly I'm doing something wrong!
Here are my system specs... Appreciate any suggestions.
TIA
MB: ASRock X299 Taichi
CPU: Intel i9-7900K Skylake-X 10 core 3.3GHz LGA 2066
CPU Cooler: Corsair Hydro H80i
Memory: G.Skill TridentZ 4x8GB DDR4 3200 (PC4 25600)
HDD (OS drive primary Win10): Samsung SSD 960 Pro (NVMe) 1TB
HDD (OS drive secondary Win7): SSD Samsung SSD 850 EVO 500GB
HDD (Data): 2x4TB WD Gold Enterprise 7200 rpm as RAID1
PSU: Corsair HX1000 CP-90201-NA 1000W Platinum
Graphics: MSI GEForce GTX 1070 Gaming X
Optical drives: 2x Pioneer Blu-Ray burner BDR-209DBK
Card reader: AFT XM-37U USB 3.0 Kiosk
Case: Cooler Master CM Storm Scout 2 Advanced SGC-2100-KWN3
Fans (additional): 2xCorsair ML120 Pro LED CO-9050043-WW 120mm Blue
Windows 10 Build: Windows 10 Pro 64bit, version 1709, OS Build 16299.192
Note: The 960 can dual boot into Win10 or Win7, but I'm running BOINC in the Win10 partition.
The 950 drive is Win7 only and is simply an "emergency" bootable drive in case the NVMe drive has a catastrophic failure. | |
|
 Michael Goetz Volunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 236,922,854
RAC: 3,199
                           
|
|
You probably know this, but I don't like to assume anything, so:
First of all, simply ignore BOINC's time estimates, and the benchmarks are utterly useless.
Once a task is running for a few minutes, the estimated time remaining should be pretty reasonable, but juts ignore the estimated times for tests that haven't started yet.
With the obvious stuff out of the way, what are the CPU temperatures? It's possible the CPU is overheating. If it is, then it will lower its clock speed to protect the CPU, and that will really slow things down.
You're definitely running way too slow. In this work unit, for example, your wingman's computer, a Sandy Bridge, finished the task in half the time. Both computers were running single threaded.
____________
My lucky number is 75898524288+1 | |
|
mackerelVolunteer tester
Send message
Joined: 2 Oct 08
Posts: 2460
ID: 29980
Credit: 442,802,854
RAC: 10,291
                        
|
|
If you're running one task per core, you're likely hitting ram bandwidth limitations. Running multi-thread should help a lot.
The L3 cache on Skylake-X is really slow and likely holds back the ram potential. You can really overclock that too. Mine does an easy 50% from 2000 to 3000, stock voltage. I've heard but not repeated myself that voltage doesn't seem to really help get more out.
Double check the CPU isn't over heating. That is a very small cooler. My 7800X has a custom loop with 360mm rad, and that I'd rate as barely adequate. It's also been delidded and had liquid metal applied.
What core voltage are you running anyway? For comparison, my 7800X does 4.3 Prime95 stable at 1.10v. It will go higher, but needs more voltage and gets hot fast.
It is also possible to hit other thermal throttling effects. Check also VRM temperature if your mobo allows. | |
|
|
|
|
Update: Well, I uninstalled BOINC, figuring that BOINC was just confused at all my setting changes, then went back and reset all the BIOS settings to default. Then I re-enabled just the ones I actually needed in order to keep both Win7 and Win10 happy.
THEN I tweaked things a bit... Used the built-in ASRock overclocking suite to bump it up to the recommended "safe" level of 4.4GHz. I also enabled TurboBoost v3.0, though I can't really see what that buys me over the overclocking. I ran into a number of issues going through all this, and I had to reset the BIOS multiple times before it all finally settled down.
I don't care that Tom's Hardware rated this board "highly recommended," I'm never doing another ASRock board again. But I digress...
The system seemed fast again, as measured by my "seat of the pants" measurer. So then I reinstalled BOINC again. I'd already switched tasks to PPS Mega for the February Tour de Primes, so I don't have an exact comparison. But the PC, if left to its own, is taking about 47 minutes to execute 10 Mega tasks. It takes one of my team members about 2 hours to run the same task, and of course, he can only run 4 of them in that time. I don't know what times others are taking, but it seems as quick as it should from my perspective.
The CPU cores themselves are running at 55 -71 degrees C with my cooling. Not sure if that's too hot or not, but at least it's all working for the moment. Appreciate the responses. | |
|
KeithVolunteer tester
Send message
Joined: 8 Dec 13
Posts: 436
ID: 284516
Credit: 412,432,810
RAC: 6,580
            
|
|
Looking at your computers, the times on a per-core basis are just about in line with the other two systems you're running. The big advantage to the 7900X is that it has so many more physical cores. So instead of going 4 tasks ever 3000 seco0nds you can get 10 tasks every 3500 seconds.
I might recommend experimenting with multi-threading via app_config.xml. I've seen great improvements in throughput on just about all projects by running 1 task on 4 cores vs 1 task on each core.
I'd probably see whether 1 task on 10 cores, 2 tasks on 5 cores, or 5 tasks on 2 cores gives you the best throughput on that CPU.
____________
My Primes
Badge Score: 2*1 + 3*1 + 4*2 + 6*4 + 7*9 + 9*1 + 10*2 = 129
 | |
|
axnVolunteer developer
Send message
Joined: 29 Dec 07
Posts: 285
ID: 16874
Credit: 28,027,106
RAC: 0
           
|
I'd probably see whether 1 task on 10 cores, 2 tasks on 5 cores, or 5 tasks on 2 cores gives you the best throughput on that CPU.
Considering that MEGA is running 256K FFT (=2MB), and that 7900X has 13.75MB of L3, best throughput will be achived by 2threads/task * 5 tasks (taking 10MB total+overhead). But it will be touch and go, since it also has quad channel memory, so the net impact of running outside L3 cache is minimized. Nonetheless, 2 threaded will improve the chance of being the primefinder as well, so that would be the sweetspot. | |
|
Message boards :
Problems and Help :
New monster PC too slow! |
