Prime95 beta version 28.3

[lalera]

Originally Posted by kladner

Were you unable to upload at-
http://www.mersenne.org/manual_result/
?

i tried often to upload at http://www.mersenne.org/manual_result/
but the server does not take it

[kladner]

What messages are you getting from the server?

EDIT: I'm not sure that I will have answers, but I am trying to ask some of the right questions.

There might be responses from people who know more, by the time we have enough data collected.

EDIT2: Those Who Know have stepped in while I was tweaking my comments.

[lalera]

Originally Posted by kladner

What messages are you getting from the server?

i get this message:
No factor lines found: 0
Mfaktc no factor lines found: 0
Mfakto no factor lines found: 0
CUDAPm1-factor lines found: 0
CUDAPm1-nofactor lines found: 0
Factors found: 0
P-1 lines found: 0
LL lines found: 0
Mlucas lines found: 0
Glucas (G29) lines found: 0
Glucas lines found: 0
MacLucasFFTW lines found: 0
CUDALucas lines found: 0
clLucas lines found: 0
ECM lines found: 0

[Prime95]

Originally Posted by lalera

username : america64
i do use v28.3
i reserved M65378393 for p-1 at gpu72 and i can not upload the result
P-1 found a factor in stage #2, B1=545000, B2=10355000, E=12.
UID: ANONYMOUS, M65378393 has a factor: 53709915590644747022698801 (P-1 B1=545000, B2=10355000, E=12)

Delete everything in the parentheses. I'll debug the manual web page.

[lalera]

i sent the following:
P-1 found a factor in stage #2, B1=545000, B2=10355000, E=12.
UID: ANONYMOUS, M65378393 has a factor: 53709915590644747022698801

and the Report of the server is:
No factor lines found: 0
Mfaktc no factor lines found: 0
Mfakto no factor lines found: 0
CUDAPm1-factor lines found: 0
CUDAPm1-nofactor lines found: 0
Factors found: 1
Processing result: M65378393 has a factor: 53709915590644747022698801
Insufficient information for accurate CPU credit.
For stats purposes, assuming factor was found using P-1 with B1 = 800000.
CPU credit is 2.9322 GHz-days.
P-1 lines found: 0
LL lines found: 0
Mlucas lines found: 0
Glucas (G29) lines found: 0
Glucas lines found: 0
MacLucasFFTW lines found: 0
CUDALucas lines found: 0
clLucas lines found: 0
ECM lines found: 0

[TheMawn]

[Mon Jan 27 22:16:17 2014]
Iteration: 20656201/32482543, POSSIBLE ERROR: ROUND OFF (0.40625) > 0.40
Continuing from last save file.
[Mon Jan 27 22:32:16 2014]
Disregard last error.  Result is reproducible and thus not a hardware problem.
For added safety, redoing iteration using a slower, more reliable method.
Continuing from last save file.

Another one. Twice in twelve DC's. Both errors were actually fine, so I'm not really worried in that regard. Again, just letting you know.

All twelve DC's matched perfectly, so I have swapped my regular work to version 28.3.

[bayanne]

I have had 2 factors which were not reported.
I have sent an e-mail to George.

[lavalamp]

Using version 28.3 of Prime95 on both an Ivy Bridge (3570k) and Haswell (4570) system, I find that the Haswell system is significantly slower at stage 1 ECM, about 40% slower clock-for-clock.

I've attached a screenshot, but basically, the 3.4 GHz Haswell takes ~47k seconds for 1 curve, while the 4.3 GHz Ivy Bridge takes 26.2K seconds.

3.4 * 47 / (4.3 * 26.2) = 1.418

So 42% extra time accounting for clock speed. The Haswell system has slightly faster RAM and runs no other tasks, while the Ivy Bridge system has me pestering it most of the time. They have the same size L1/2/3 CPU caches.

I do not know why there should be such a big difference in performance (and had I known it was such a big gap I would have got another Ivy Bridge!). If anyone can shed light on this I'd be interested.

[Prime95]

Originally Posted by lavalamp

Using version 28.3 of Prime95 on both an Ivy Bridge (3570k) and Haswell (4570) system, I find that the Haswell system is significantly slower at stage 1 ECM.

I will look into it. Have you tried setting CPUSupportsFMA3=0 in local.txt? Have you tried 27.9?

[lavalamp]

I don't have a record to back this up, but I think v27.9 ran at the same speed. I do have numbers to show that v28.1 also ran at this speed.

I've set CpuSupportsFMA3=0 in local.txt and started it up again (still v28.3), it will be several hours before it's possible to see the effect though, so I'll report back tomorrow.

[lavalamp]

Doesn't seem to have had an effect unfortunately.

It was part way through 3 curves when I started it last night, all of which have now finished. But the time it took to run the remaining proportion of the curves is still around the same 47k seconds / curve level.