Weekend GT730 Project

[masser]

I have successfully run mfaktc and cudalucas on my first gpu, a GT730. For GIMPS, I have found a few factors AND completed a LL-DC. Now, I would like to complete the following four tasks:

1. Get my gpu listed on mersenne.ca gpu benchmarks page
2. Tune mfaktc for optimal performance
3. Tune cudalucas for optimal performance
4. Answer the TF vs. LL-DC question for this gpu

Any help or insights would be greatly appreciated.

[masser]

Is this the right information for a CUDALucas benchmark?

------- DEVICE 0 -------
name GeForce GT 730
Compatibility 3.5
clockRate (MHz) 901
memClockRate (MHz) 2505
totalGlobalMem 2146762752
totalConstMem 65536
l2CacheSize 524288
sharedMemPerBlock 49152
regsPerBlock 65536
warpSize 32
memPitch 2147483647
maxThreadsPerBlock 1024
maxThreadsPerMP 2048
multiProcessorCount 2
maxThreadsDim[3] 1024,1024,64
maxGridSize[3] 2147483647,65535,65535
textureAlignment 512
deviceOverlap 1

Using threads: square 256, splice 128.
Starting M57885161 fft length = 3136K
| Date Time | Test Num Iter Residue | FFT Error ms/It Time | ETA Done |
| Nov 04 20:16:34 | M57885161 10000 0x76c27556683cd84d | 3136K 0.25000 30.3519 303.51s | 20:07:57:04 0.01% |
| Nov 04 20:21:38 | M57885161 20000 0xfd8e311d20ffe6ab | 3136K 0.25781 30.3301 303.30s | 20:07:41:28 0.03% |
| Nov 04 20:26:41 | M57885161 30000 0xce0d85ab0065a232 | 3136K 0.24609 30.3214 303.21s | 20:07:30:07 0.05% |
^C SIGINT caught, writing checkpoint. Estimated time spent so far: 15:34

[masser]

I hope this works for the mfaktc benchmarks:

GPU brand: NVIDIA
GPU model: GT 730 (GDDR5)
Stock core clock: 901 MHz
Actual core clock: 901 MHz
Assignment: Factor=134473637,71,72
Running Time: 1h 28m 56.640s
SievePrimes: 25000
GPU usage (avg): 99%
mfaktc version: mfaktc v0.21 (64bit built)

[Gordon]

Originally Posted by masser

I hope this works for the mfaktc benchmarks:

Why are you bothering?

What took you 1h 29m I ran in 12m 33s...

[Mark Rose]

Originally Posted by Gordon

Why are you bothering?

What took you 1h 29m I ran in 12m 33s...

Because it's fun? The distributed effort all adds up.

I have a 30 GHz-d/d card that has put in over 20 THz-d.

[Xyzzy]

Originally Posted by Gordon

Why are you bothering?

Maybe he is using it as a learning exercise?

[masser]

Originally Posted by Gordon

Why are you bothering?

What took you 1h 29m I ran in 12m 33s...

I bought the GT 730 for a few reasons:

1. It's one of the more powerful fanless gpu available on the market

2. It only burns a few watts (less than 15W with 1 instance of mfaktc at 99% GPU utilization)

3. It's absent from the mersenne.ca benchmarks; if there are other folks interested in quiet, energy-efficient systems that want to contribute to the project, they too might wonder about the GT 730. By adding this data point, we might help them make a better purchase.

[VBCurtis]

Originally Posted by Gordon

Why are you bothering?

What took you 1h 29m I ran in 12m 33s...

If you're consuming over 110W for your GPU on this run, why are *you* bothering? 8x the speed for 8x the power is available for anyone from Amazon instances. If the GT730 produces similar production per watt-hour as your GPU, it is precisely as useful to the global effort, merely at a smaller (slower?) scale.

[masser]

Originally Posted by masser

2. Tune mfaktc for optimal performance

Any help or insights would be greatly appreciated.

Fiddling with GPUSieveProcessSize in mfaktc.ini:

value : GhzD/D
32 : 34.8
24 : 57.6
16 : 57.6
8 : 58.5

Is this expected behavior?

[Mark Rose]

Originally Posted by masser

Fiddling with GPUSieveProcessSize in mfaktc.ini:

value : GhzD/D
32 : 34.8
24 : 57.6
16 : 57.6
8 : 58.5

Is this expected behavior?

Yes. The best number will depend on your card.

[masser]

I think I've settled on:

GPUSievePrimes=87400
GPUSieveSize=128
GPUSieveProcessSize=8

The GhzD/D increased to about 58.6 from 57.6 with the defaults.

Should I try changing any of the other options in the mfaktc.ini file?