GPU LL Testing FAQ - Page 10

Dubslow · 2012-07-13, 22:03

Quote:

Originally Posted by henryzz

Wouldn't this mean that if we got some sort of GPU LL test running on a ivy bridge it would run really fast?

It's not the CPU that's the bottleneck, it's the transfers between host and device over (typically) the PCI slot. Those take a long time. (And besides, like I said, the CPU code is light enough that you don't notice it. Extra speed there won't help the GPU.)

You know the polite function in CUDALucas, which causes an iteration to take ~10+% longer? That's implemented by transferring one float from device to host. One float transfer causes a 10% performance hit.

sdbardwick · 2012-07-13, 22:17

I interpreted henryzz's question like this:
Wouldn't this mean that if we got some sort of GPU LL test running on a ivy bridge's integrated GPU it would run really fast?

Dubslow · 2012-07-13, 22:32

Quote:

Originally Posted by sdbardwick

I interpreted henryzz's question like this:
Wouldn't this mean that if we got some sort of GPU LL test running on a ivy bridge's integrated GPU it would run really fast?

Ooooohhhhhhhhhhhh!!!!

Probably not. Getting it running on IB-IG requires the same work as getting it running on AMD cards, namely someone needs to write a fast OpenCL FFT library.

henryzz · 2012-07-14, 11:30

Quote:

Originally Posted by Dubslow

Ooooohhhhhhhhhhhh!!!!

Probably not. Getting it running on IB-IG requires the same work as getting it running on AMD cards, namely someone needs to write a fast OpenCL FFT library.

I was sort of ignoring the work required for writing the program and just thinking about the results. The bandwidth for an integrated gpu should be much higher.

bcp19 · 2012-07-15, 14:08

I've run more than 1 instance of CL in the past and can attest to the fact that it gives no benefit(and possibly a bit of detriment). CPU usage is minimal with all the calaculation being done by the GPU (3% or less) so CPU speed does not seem to affect the result. That brings up an interesting test idea though, I know I tried to run a GTX 480 in an 8200 Quad and even with all 4 cores could not max out the GPU with Mfaktc. It would be interesting to test the 480 doing CL to see how the timings are using the Quad vs the 2500K.

kladner · 2012-07-16, 04:38

@bcp19: in regard to the GTX 480, what % did you consider "maxed out"? Two instances of mfaktc run my GTX 460 ~97% avg, depending on other demands on the system. This is on a Phenom II x6 1090T with 2 cores dedicated to mfaktc.

I have been sniffing around Gtx 570s, with the thought of running mfaktc on such a thing and giving the 460 to CUDALucas. I'm just wondering if two 1090T cores would do it justice, or if I would have to give up a P-1 worker to properly feed it.

TObject · 2012-07-18, 01:50

I followed the instructions in the first post PDF.

My CUDALucas file name is CUDALucas.cuda4.0.sm_21.WIN64.exe, it reports v1.4.2. Do I have the latest version?

If not, where do I get the latest version?

Thanks.

LaurV · 2012-07-18, 02:42

Quote:

Originally Posted by TObject

where do I get the latest version?

That should be v2.04 with few different flavors already. The version you have is much behind. Now CL has its own web page and it is much more clever, it can handle "variable" FFT sizes (not powers of 2) and it is MUCH MUCH faster on newer cards. Check the "cuda lucas" thread in this subforum (it may be page 60 or before).

edit: the pdf (v07) in the first post of this thread which you used is also old, maybe some mod can actualize it??.
I remeber I saw v10 or v11 already.

edit 2: I have v10 on the disk but I can't edit thread head in this subforum.

TObject · 2012-07-18, 02:59

Thank you. Which one is the latest non-beta for a GTX 580 on Windows 64 bit?

I started a test with v1.4.2. Are the save files compatible?

Thanks again.

Dubslow · 2012-07-18, 03:05

I don't think the save files are compatible, though that version is so old I'm more guessing than remembering. 2.03 is the latest stable version.

http://sourceforge.net/projects/cudalucas/files/

TObject · 2012-07-18, 03:58

Perfect. Thank you.

2012-07-13, 22:17	#101
sdbardwick Aug 2002 North San Diego Coun 821 Posts	I interpreted henryzz's question like this: Wouldn't this mean that if we got some sort of GPU LL test running on a ivy bridge's integrated GPU it would run really fast? Last fiddled with by sdbardwick on 2012-07-13 at 22:18 Reason: still learning how to type...

2012-07-18, 02:59	#108
TObject Feb 2012 110010101₂ Posts	Thank you. Which one is the latest non-beta for a GTX 580 on Windows 64 bit? I started a test with v1.4.2. Are the save files compatible? Thanks again. Last fiddled with by TObject on 2012-07-18 at 03:08

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Anti-poverty drug testing vs "high" tax deduction testing	kladner	Soap Box	3	2016-10-14 18:43
What am I testing?	GARYP166	Information & Answers	9	2009-02-18 22:41
k=243 testing ??	gd_barnes	Riesel Prime Search	20	2007-11-08 21:13
Testing	grobie	Marin's Mersenne-aries	1	2006-05-15 12:26
Speed of P-1 testing vs. Trial Factoring testing	eepiccolo	Math	6	2006-03-28 20:53

2012-07-15, 14:08	#104
bcp19 Oct 2011 7·97 Posts	I've run more than 1 instance of CL in the past and can attest to the fact that it gives no benefit(and possibly a bit of detriment). CPU usage is minimal with all the calaculation being done by the GPU (3% or less) so CPU speed does not seem to affect the result. That brings up an interesting test idea though, I know I tried to run a GTX 480 in an 8200 Quad and even with all 4 cores could not max out the GPU with Mfaktc. It would be interesting to test the 480 doing CL to see how the timings are using the Quad vs the 2500K.

2012-07-16, 04:38	#105
kladner "Kieren" Jul 2011 In My Own Galaxy! 2×3×1,693 Posts	@bcp19: in regard to the GTX 480, what % did you consider "maxed out"? Two instances of mfaktc run my GTX 460 ~97% avg, depending on other demands on the system. This is on a Phenom II x6 1090T with 2 cores dedicated to mfaktc. I have been sniffing around Gtx 570s, with the thought of running mfaktc on such a thing and giving the 460 to CUDALucas. I'm just wondering if two 1090T cores would do it justice, or if I would have to give up a P-1 worker to properly feed it.

2012-07-18, 01:50	#106
TObject Feb 2012 3⁴·5 Posts	I followed the instructions in the first post PDF. My CUDALucas file name is CUDALucas.cuda4.0.sm_21.WIN64.exe, it reports v1.4.2. Do I have the latest version? If not, where do I get the latest version? Thanks.

2012-07-18, 03:05	#109
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 7221₁₀ Posts	I don't think the save files are compatible, though that version is so old I'm more guessing than remembering. 2.03 is the latest stable version. http://sourceforge.net/projects/cudalucas/files/

2012-07-18, 03:58	#110
TObject Feb 2012 3⁴·5 Posts	Perfect. Thank you.