[QUOTE=owftheevil;339389]The different kernels run synchronously, the cutilSafeThreadSync call is so the cpu doesn't do busy waiting and eat up an entire cpu core.[/QUOTE]

:busted:

I should have remembered that....

Anyone keep track of the largest LL that has been run on CUDALucas so far? Any problems with bigger numbers?

I might have a candidate for largest such test in about 15 hours...:smile:

[QUOTE=chalsall;339372]Perhaps a new GIMPS slogan: "Our software tests hardware like no other!".... :smile:[/QUOTE]

Proof that projects like GIMPS aren't just esoteric artifacts of geekdom. Remember that it took Nicely's derivation of the twin prime constant to find the FDIV bug, and even then, Intel tried to weasel out of it by arguing that "no one pushes their computers that hard"!

Hi owftheevil,

I just tested your memtest on some GPU:
It doesn't make any output when iterations is <10 and doesn't show 100% if iterations is not a multiple of 10.
Fix:
[CODE] 48 if(j != pos) compare_kernel<<<n / 512, 512>>> (&g_ttp[j * n], &g_ttp[pos * n], g_compare);
49 cutilSafeThreadSync();
50 }
51 if(k%10 == 0[B][COLOR="Red"] || k == iter[/COLOR][/B])
52 {
53 cutilSafeCall (cudaMemcpy (&compare, g_compare, sizeof (int), cudaMemcpyDeviceToHost));
54 cutilSafeCall (cudaMemset (g_compare, 0, sizeof (int)));
[/CODE]

Additionally there is a floating point exeception when size is 1:[CODE] 55 read_total += compare /((s - 1) * 10);
56 compare %= (s - 1) * 10;
57 read_total += compare / (s - 1);
58 compare %= (s - 1);
[/CODE]
I didn't spent time thinking about what should be done here, sorry. :redface:

Oliver

With size = 1, the test doesn't actually do anything, so I should probably default to a minimum of 2 for size.

Thanks for pointing those out.

Carl

I think this might be the largest CUDALucas run to date:
M( 82090249 )C, 0xb8398d26ebabea__, n = 4718592, CUDALucas v2.03

Total running time was somewhere around 325 hours or so, running some of the time on a K10 and the rest on a K20. Will be interesting to get a double-check back on this one.

This is the first LL residue I can remember turning in that features the string "babe", not to mention "6ebabe" ("sexy babe"), LOL. At least it's not DEADBEEF. :smile:

For s & g I threw that exponent (82090249) on a i5-2500. Should be done 2013-05-20.

[QUOTE=NBtarheel_33;339917]I think this might be the largest CUDALucas run to date:
M( 82090249 )C, 0xb8398d26ebabea__, n = 4718592, CUDALucas v2.03

Total running time was somewhere around 325 hours or so, running some of the time on a K10 and the rest on a K20. Will be interesting to get a double-check back on this one.

This is the first LL residue I can remember turning in that features the string "babe", not to mention "6ebabe" ("sexy babe"), LOL. At least it's not DEADBEEF. :smile:[/QUOTE]

was it a prime ? Just wondering before I will start double check on it with my Titans..

[QUOTE=owftheevil;339740]With size = 1, the test doesn't actually do anything, so I should probably default to a minimum of 2 for size.

Thanks for pointing those out.

Carl[/QUOTE]

Here's a trial Windows x64 binary:
[URL="https://www.dropbox.com/s/4lh34niqddm5tf8/CUDAmemtest_20130509.zip"]https://www.dropbox.com/s/4lh34niqddm5tf8/CUDAmemtest_20130509.zip[/URL]

[QUOTE=Manpowre;339927]was it a prime ? Just wondering before I will start double check on it with my Titans..[/QUOTE]

I wouldn't be typing this right now if it were, for they'd have had to surgically remove my lower jaw from the keyboard, LOL. You need all zeroes in your residue for a prime, remember? And my residue had "babe" in it, LOL. Would have been nice, though, because the first six digits - 820902 - are my birthday in big-endian format. I do have one other chance with the one other number of this form, which I have 71% complete on Prime95. I am going to move it over to the big iron to get it finished in the next few days. If you hold up on the Titans, you could double-check the Prime95 result on *that* number...

[QUOTE=sdbardwick;339924]For s & g I threw that exponent (82090249) on a i5-2500. Should be done 2013-05-20.[/QUOTE]

Thanks! I am interested, as I am sure all the GPU code authors are, in seeing if we get a match.

How many cores of the i5 are you using? 11 days is actually faster than CUDALucas! Must be better FFT size selection on Prime95.