Prime95 version 26.5

[Prime95]

Originally Posted by oilwarzonedotco

Has the new software been activated to work on the servers?

The server is supposed to auto-register new prime95 versions. It did so with the Windows and Linux 64-bit build 5, but didn't for Linux 32-bit. Of course, as soon as I tried to debug the problem the problem went away.

[James Heinrich]

Originally Posted by Commaster

Logical CPUs detection is clearly broken in b5, compared to b3 (I missed b4). Using Core-i7 920 with Windows 7 x64

I noticed the same thing in b4, and George has no fix at the moment:

Originally Posted by Prime95

I don't know how to proceed. For now, I'll just gather more data unless a brainstorm hits me.

[James Heinrich]

In 25.6.5, my Q6600 no longer respects MaxHighMemWorkers=2 in local.txt. I hadn't upgraded that machine since v25.6.2, when it worked, now it doesn't (all 3 workers that want to use RAM are using RAM).
edit: Q6600 is on Vista64, and is broken. i7-920 is on Seven64, and is working properly.

[drh]

I saw this happen a little while ago on one of my other machines as well ... any ideas?

[Feb 22 13:02] M53730367 stage 2 is 96.50% complete. Time: 1299.093 sec.
[Feb 22 13:24] M53730367 stage 2 is 97.48% complete. Time: 1302.961 sec.
[Feb 22 13:46] M53730367 stage 2 is 98.45% complete. Time: 1299.585 sec.
[Feb 22 14:05] Available memory is 6904MB.
[Feb 22 14:05] Using 401MB of memory. Processing 2 relative primes (286 of 288 already processed).
[Feb 22 14:12] M53730367 stage 2 is 99.30% complete. Time: 1565.172 sec.
[Feb 22 14:30] M53730367 stage 2 is 99.38% complete. Time: 1089.810 sec.

Thanks,
Doug

[Prime95]

Originally Posted by drh

[Feb 22 14:05] Using 401MB of memory. Processing 2 relative primes (286 of 288 already processed).

There were only 2 relative primes left to process - not much memory was needed to do that

[drh]

Makes sense, but my other machine dropped with 89 remaining, and delays completion by an hour on this machine, and a few hours on the other ... just an observation.

Thanks!

[Commaster]

Uhm... It does work sometimes. I suspect, it works only if there's no other serious load on the CPU (b3 worked in both cases):

Optimizing for CPU architecture: Core i3/i5/i7, L2 cache size: 256 KB, L3 cache size: 8 MB
Test clocks: 153882
Logical CPU 1 clocks: 299840
Test clocks: 153882
Logical CPU 3 clocks: 296388
Test clocks: 153878
Logical CPU 5 clocks: 301070
Test clocks: 153882
Logical CPU 7 clocks: 299716
Logical CPUs 0,1 form one physical CPU.
Logical CPUs 2,3 form one physical CPU.
Logical CPUs 4,5 form one physical CPU.
Logical CPUs 6,7 form one physical CPU.

[Prime95]

Originally Posted by Commaster

Uhm... It does work sometimes. I suspect, it works only if there's no other serious load on the CPU (b3 worked in both cases):

I assume all these not working complaints are on Windows?

As explained earlier, b3 was not working on Windows. A bug caused it to always think 0,1 paired, 2,3 paired, etc.

The only explanation I have for the output I've seen on the failed runs is that the Windows scheduler is not running the thread on logical 0 and the thread running on logical 1 at the same time even though I've set affinity to do so.

In any case, these startup warnings are not serious as Prime95's default assumptions match how the OS is assigning logical CPUs.

[ixfd64]

The failure to detect hyperthreaded cores may be a software regression that was introduced when George patched the CPU #31 bug. The feature stopped working in build 4, and the CPU #31 issue seems to be the most closely related.

George, I would recommend looking at your code for the CPU #31 fix.

[Prime95]

Originally Posted by ixfd64

The failure to detect hyperthreaded cores may be a software regression that was introduced ...

As explained before, detection was NOT working in b3/4 on Windows. The clock counting always returned zero. Twice zero is zero - therefore it thought it found the logical CPU running at half speed.

[ixfd64]

FYI: Prime95 26.5 is now a stable release. Check out the GIMPS home page!