GTX570 SC (running on 266.xx drivers against instructions)

running 121*2^4207281-1 Time per iteration: 1.303ms.
currently at 40% completion

same # running on one cpu/1thread======>.2.437ms.
i7-920 @ 3.75Gz

the 570 is using about 33% of one cpu

Do you have the final results? (timings)

EDIT:

By the iterations you post I see the GPU client is 1.9 x faster than the CPU client but the GPU uses, as you stated, 33 % of one core of your 4-core machine.
Regarding that the TDP of the GTX570 SC is 214 W and of the CPU is 130 W at stock speed ( 130*3.75/2.66 overclocked, if consumption linear), and that the GPU uses 33 % of one core, then I can determine that the GPU client is 2.7 x less energy efficient than the CPU client running on all 4 cores.
For the GPU to have the same energy efficiency than the CPU it should have a time iteration of 0.476 ms for the current size of number, meaning that the GPU client should be at least 5.1 x faster than the CPU client.

Carlos
____________

I agree.

But remember I was running the older drivers, so not
a true test.

LLRCUDA is in it's earliest forms.
Remember that about a year ago people
where saying that LLR would never run on a GPU.

We are still in the earliest code.

My Kill--A-Watt meter says 442 watts
running the GTX and 4 Cores at once.

By the way, final timing was
EVGA GTX570 SC -------5522 seconds.
CPU -----------------------10494 seconds.
CPU running 4 cores ---14792 seconds average

There should be a better result running an SoB unit
which I will try to do.

Cheers.

Pilgrim,

With your new results I got:

Although the GPU is 2,68x faster than one core of the CPU it is 1,91x less energy efficient than the CPU with 4 cores running.

I used the standard TDP for both CPU and GPU. Can you make a test? First see how much watts consumes the machine with only the GPU crunching and then GPU off with all 4 cores on. In this way calculations will be more realistic.

Carlos
____________

o.k.

I was away for a few days so will get you the info......

sorry for the delay Carlos....

i7/920 o/c'd to 3.75Ghz

running only 4 CPU's (4 threads)...290 watts 15,000 seconds/unit

running only GTX570 SC ...............360 watts 5,700 seconds/unit

running all 5 above ......................460 watts

Cheers

sorry for the delay Carlos....

i7/920 o/c'd to 3.75Ghz

running only 4 CPU's (4 threads)...290 watts 15,000 seconds/unit

running only GTX570 SC ...............360 watts 5,700 seconds/unit

running all 5 above ......................460 watts

Cheers

Kill-A-Watt says 200 watts with idle mode.

The GPU always runs a unit in the same amount of time
regarless of the load on the CPU.

I am not sure what you mean about "both processors"
because I have an i7-920 but I am not using HT.

And to add the excitement, while Folding at home uses
less than 10% of a CPU core LLRCUDA uses around 60-70%
of a core.

The GPU always finishes around 5700 seconds.

The CPU's (4) are now taking over 17,000 seconds
for 4 cores/units. Of course, divide by four and it seems/is
the GPU is energy inefficient BUT I could never get
below 12000 seconds/unit running just one Core/unit
(without GPU) so it seems to my brain dead mind
that even though the GPU is a power hog it still beats
One CPU core in a race, even considering the power usage
of the GPU. What am I missing?

I think the GPU might have a better comparison on
REALLY big numbers, like 14 million digits?

Let me know what else you need.

Cheers

p.s....the numbers in previous post are when
all four cores AND GPU are running...just to clear that up.

GPU not running:

CPU running 4 cores ---17192 seconds average
CPU % utilization:50%

==================================

GPU running:

EVGA GTX570 SC -------5722 seconds.
CPU running 4 cores ---17192 seconds average
CPU % Utilization: 57%

The GPU always finishes around 5700 seconds.

The CPU's (4) are now taking over 17,000 seconds
for 4 cores/units. Of course, divide by four and it seems/is
the GPU is energy inefficient BUT I could never get
below 12000 seconds/unit running just one Core/unit
(without GPU) so it seems to my brain dead mind
that even though the GPU is a power hog it still beats
One CPU core in a race, even considering the power usage
of the GPU. What am I missing?

I think the GPU might have a better comparison on
REALLY big numbers, like 14 million digits?

yes, my computers are in a warm environment and
even though they are dust free, have 6-8 fans each
and are air cooled with the Noctura 14 my RealTemp
runs about 75 degrees with four cores. 8 threads
pushes the temperature into the mid 80's unless
I cut back on the O/C.

I will start running 6 threads soon and find I lose only about
20% throughput versus 8 threads. The law of diminshing returns
with the HT feature.

@Carlos: the calculations are quite easy; the CPU is using 290 W and the GPU 170 W (based on the reasoning that GPU always runs together with CPU)

P(W) t(s) E(Wh) E(Wh/task)
CPU 290 17192 1385 346
GPU 170 5722 270 270