Website update, new client, Hi Louie

Website update, new client, Hi Louie
" SPEEDUP: NEW SB CLIENT v27.7 AVAILABLE
(posted by Louie Helm)
Wednesday, 16 May 2012

UPDATED Prime95 client available for SB

There's a new version of SB available thanks to even more great work by George Woltman.


Major Enhancements

25% speed increase on Intel i5/i7 processors
Better multi-threaded performance
Faster FFT for Core 2 w/ 1MB L2 cache
Mac OS X client now GUI instead of CLI
"

If I remember correctly, if I have an existing setup, open the .gz, extract mprime to working directory and fire off as usual

Thanks for the heads-up. :thumbs:
I rarely look at the SoB home page.

Anybody else having possible hardware but it was reproducible error message on 27.7 build 2 on AMD cpus???? No error message on core 2 cpus.

Quote:

---

Originally Posted by engracio

Anybody else having possible hardware but it was reproducible error message on 27.7 build 2 on AMD cpus???? No error message on core 2 cpus.

---

If you are getting small reproducible roundoff error values like 0.40625 and 0.4375 then it is likely due to testing a value near the limits of what that FFT size can test. In other words, nothing to worry about.

If you are getting roundoff errors like 0.49999 then it is a hardware issue.

P.S. The FFT limits changed ever so slightly between 26.6 and 27.7. So it is entirely possible that the same number would use a different FFT size in the two versions.

Quote:

---

Originally Posted by prime95

If you are getting small reproducible roundoff error values like 0.40625 and 0.4375 then it is likely due to testing a value near the limits of what that FFT size can test. In other words, nothing to worry about.

If you are getting roundoff errors like 0.49999 then it is a hardware issue.

P.S. The FFT limits changed ever so slightly between 26.6 and 27.7. So it is entirely possible that the same number would use a different FFT size in the two versions.

---

THanks George you are correct, it was small reproducible roundoff error values like 0.40625 and 0.4375. It also states to disregard the error. Your expanation seems to match the problem which was nothing compare to the speedup added on the new build. Thank you again.:thumbs:

Yep, I noticed this on mine too shortly after upgrading (Intel i7 here, not AMD)

One of my tests seems to keep getting errors. It says:

18 ROUNDOFF > 0.4 of which 8 were repeatable (not hardware errors).
Confidence in final result is very poor.

Does this mean 10 of the 18 errors it thinks are hardware errors? Yesterday it said confidence was "fair", then "poor", now "very poor" so it's getting worse. I am thinking maybe I just need to abort this test and start a new one. (This test was started in v. 26.6) I'm not overclocked, I'm not overheating, and I've never had issues with this computer before, so I am hesitant to call it a hardware issue. I am running two workers (each with 2 threads) and the 2nd worker does not have any problems. The 2nd worker was very near the end of the test when I upgraded, so it completed the test that was started in 26.6 and started a new test in 27.7 within an hour or two of upgrading, with no errors.

Quote:

---

Originally Posted by enderak

Yep, I noticed this on mine too shortly after upgrading (Intel i7 here, not AMD)

One of my tests seems to keep getting errors. It says:

18 ROUNDOFF > 0.4 of which 8 were repeatable (not hardware errors).
Confidence in final result is very poor.

Does this mean 10 of the 18 errors it thinks are hardware errors? Yesterday it said confidence was "fair", then "poor", now "very poor" so it's getting worse. I am thinking maybe I just need to abort this test and start a new one. (This test was started in v. 26.6) I'm not overclocked, I'm not overheating, and I've never had issues with this computer before, so I am hesitant to call it a hardware issue. I am running two workers (each with 2 threads) and the 2nd worker does not have any problems. The 2nd worker was very near the end of the test when I upgraded, so it completed the test that was started in 26.6 and started a new test in 27.7 within an hour or two of upgrading, with no errors.

---

Ya everything you said. The most I had was 10 errors with 7 repeatable. I think the more error you get the less confidence it has. My 10 error msg was rated poor. It was so close to finishing I let it finish. A fair rating I normally finish unless it just started.

OK, well I've put the troublesome test on hold for now and started a brand new test for that worker. We'll see how it does over the weekend. If the new test doesn't have any issues I'll just abort the screwed up test and not worry about it. If I keep getting errors, I guess I'll figure out where to go from there.

Can you post the numbers that are getting roundoff errors? I'll look into whether the FFT size selection code needs a little fine tuning.

Quote:

---

Originally Posted by enderak

18 ROUNDOFF > 0.4 of which 8 were repeatable (not hardware errors).

---

Again look at the size of the roundoff errors. 0.40625 and 0.4375 are OK, 0.4999 is really bad.

I think a roundoff error of 0.40625 and 0.4375 may sometimes not be reproducible because there could be a slight change in the FFT data when the number is reloaded from the save file. If so, your test would be fine and I should change the wording of the "confidence is poor" warnings.

This is the error msg I was getting "Iteration: 16159691/22839108, ERROR: ROUND OFF (0.4375) > 0.40
Continuing from last save file." I think we should let the wu complete from 26.6 if it is almost finish as prime95 suggested. Have 27.7 pick a new wu, keep an eye on the next couple wu until each is satisfied that the issue was just due to ver transition. Otherwise just keep on crunching.:D

Quote:

---

Originally Posted by engracio

ERROR: ROUND OFF (0.4375) > 0.40

---

That 0.4375 is just the kind of round off error I wouldn't worry about. BTW, what number are you testing.

Quote:

---

Originally Posted by prime95

That 0.4375 is just the kind of round off error I wouldn't worry about. BTW, what number are you testing.

---

22839108

engracio, you might need to post your k as well.

My trouble-maker is 21181•2^22976900+1

I don't think I can get my exact error values any more, because I restarted prime95 since they occurred and I don't see them in the log. I don't think it's anything to worry about like you said, but I'll probably just cancel the test just in case. It's only at 35%, so I'm not losing that much work. I'd hate myself forever if it turned out to be a bad test that was actually prime :P

EDIT: On second thought - I have an extra machine available, maybe I'll disable reporting & run the test to see what happens.

It was 22699•2^22839108+1. In all honesty I have not seen the error msg again after it all completed or restarted with 27.7 build.

If louie is amiable and have a little time, maybe he can drop the wu with errors during the transition from 26.6 to 27.7 to the first pass queue.

Otherwise I think this was a non issue in the first place.:D