Question: Will P95 stop the ECM (and switch to the next assignment) if a factor is found (in stage 1? in stage 2?), or it will continue until the number of pledged curves is finished?

(edit: doing 150 curves for each of the first 5 exponents over 7M, currently at curve 136, 138, 141, 139 and 0 respectively, on a 4 core machine)

It will stop unless the factor found was supplied as part of the known_factors list for the assignment.

AffinityScramble2 doesn't work on Bulldozer
 

Hi,

Mprime v27.7 (Linux 64bit) doesn't honor AffinityScramble2 on Bulldozer CPUs? I didn't try older mprime versions

local.txt[CODE]
[...]
WorkerThreads=1
ThreadsPerTest=2
Affinity=0
AffinityScramble2=24
[...]
[/CODE]
works fine on all CPUs I've tested except Bulldozer CPUs? It works as expected on the older Opteron 61xx and on all Xeon systems I've tested.

[CODE]
Mersenne number primality test program version 27.7
Optimizing for CPU architecture: AMD Bulldozer, L2 cache size: 2 MB, L3 cache size: 12 MB
Starting worker.
Worker starting
Setting affinity to run worker on logical CPU [COLOR="Red"][B]#1[/B][/COLOR]
Optimal P-1 factoring of M59xxxxxx using up to 59615MB of memory.
Assuming no factors below 2^73 and 2 primality tests saved if a factor is found.
Optimal bounds are B1=555000, B2=11377500
Chance of finding a factor is an estimated 3.76%
Using AMD K10 FFT length 3200K, Pass1=640, Pass2=5K, 2 threads
Setting affinity to run helper thread 1 on logical CPU [COLOR="Red"][B]#2[/B][/COLOR]
[/CODE]
Any idea how to debug? I quick view into the sources didn't help, I've tried DebugAffinityScramble=2 but got no additional screen outputs.

Oliver

In my reading of the code AffinityScramble2 is only used on machine that support hyperthreading. Try setting NumCPUs=n and CpuNumHyperthreads=2
in local.txt

Adding "CpuNumHyperthreads=2" to local.txt is enough.
AffinityScramble2 is working fine on Opteron 61xx which don't have Hyperthrashing.

Oliver

[QUOTE=TheJudger;319524]
AffinityScramble2 is working fine on Opteron 61xx which don't have Hyperthrashing.[/QUOTE]

It shouldn't. Does Options/CPU report a hyperthreaded CPU?

My fault (wrong observation), on Opteron 61xx AffinityScramble2 is ignored, too. Adding CpuNumHyperthreads=2 to local.txt does the trick, too.

Oliver

26.6 and 27.7 x64 torture test difference
 

I'm using x64 versions of Prime95. There is difference in torture test memory consumption between 26.6 and 27.7.
Sometimes on 4GB Windows 7 x64 SP1 (without any updates, high perf profile) 26.6 can eat up to 3,29GB of 3,5GB usable memory (4GB installed but 512MB used by int video) in Blend test, 27.7 can occupy only 1.6GB, or 26.6 can't start the 1st of 4 threads (Starting, Self-Test, Self-test, Self-test) while using all 4 cores.
Also under 26.6 Blend test computer is barely usable, perhaps because of no free memory. I don't remember I've seen such sluggish system, it seems like frozen at first, but responds in 5-10 seconds on your actions. It can take some time to stop the test. Under 27.7 Blend test it can run unnoticed. Unfortunately it doesn't happens all the time. Frequently 26.6 starts testing consuming 1,6GB of memory and running 100% CPU unnoticed like 27.7.
I haven't found changes in whatsnew.txt regarding tests, except one: "Multi-threaded tests might be a little bit faster especially when using a lot of threads."
Do you suggest using 27.7 for AMD K10?
Can I help narrowing down the issue?

results.txt of 27.7 is:
AMD Athlon(tm) II X4 640 Processor
CPU speed: 3013.38 MHz, 4 cores
CPU features: 3DNow!, SSE, SSE2
L1 cache size: 64 KB
L2 cache size: 512 KB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
L1 TLBS: 48
L2 TLBS: 512
Prime95 64-bit version 27.7, RdtscTiming=1

26.6:
AMD Athlon(tm) II X4 640 Processor
CPU speed: 8863.04 MHz, 4 cores
CPU features: Prefetch, 3DNow!, MMX, SSE, SSE2
L1 cache size: 64 KB
L2 cache size: 512 KB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
L1 TLBS: 48
L2 TLBS: 512
Prime95 64-bit version 26.6, RdtscTiming=1

When version 26.6 uses 3.2GB of memory and your system slows to a crawl - this is not a good thing. Windows is likely thrashing -- paging memory to and from disk. This may be stressful to the disk, but not the CPU. Thus, you are not getting a good torture test of the CPU.

Somewhere between 26.6 and 27.7 either a bug was fixed or the algorithm for computing the default amount of memory to use in a blend torture test changed.

I suggest using 27.7. If you want to try to get prime95 to use more than 1.6GB, try using a custom torture test giving prime95 more memory to use.

I am getting the following on every iteration:

[Dec 7 16:29] Iteration: 1193000 / 29810063 [4.00%]. Per iteration time: 0.055 sec.
[Dec 7 16:29] Possible hardware errors have occurred during the test!
[Dec 7 16:29] 2 ROUNDOFF > 0.4 of which 1 were repeatable (not hardware errors).
[Dec 7 16:29] Confidence in final result is fair.

Prime95 v27.7, build 2 on a box that I have not had any bad results. The other three workers are fine.

How do I specify a larger FFT? Do I stop and start over again since I'm only at 4.00%. Thanks!

Can you post the roundoff error reports from results.txt?
Some roundoff values (such as 0.4375) are really OK when running near the upper limit of an FFT size.

If you have one of the suspicious error values (0.5 or very close to it) then I'd restart the test since you're only at 4%.