Prime95 version 27.4

Prime95 · 2012-03-09, 02:57

Prime95 beta version 27.4 is available.

This version fixes a number of bugs in 27.3 - mostly related to LLR and PFGW issues. There are no speed improvements over 27.3. Hopefully, Bulldozer processors now properly run SSE2 FFTs.

Unless you have a Sandy Bridge CPU, there is little reason to upgrade to this version.

Download links:
Windows 64-bit: ftp://mersenne.org/gimps/p64v274.zip
Windows 32-bit: ftp://mersenne.org/gimps/p95v274.zip
Linux 64-bit: ftp://mersenne.org/gimps/mprime274-linux64.tar.gz
Linux 32-bit: ftp://mersenne.org/gimps/mprime274.tar.gz
Mac OS X: ftp://mersenne.org/gimps/Prime95-MacOSX-274.zip
FreeBSD 64-bit: ftp://mersenne.org/gimps/mprime274-FreeBSD64.tar.gz
Source code: ftp://mersenne.org/gimps/source274.zip

I have one more significant change planned that should speed up multithreaded FFTs somewhat.

Prime95 · 2012-03-09, 02:57

Known bugs and fixes:

1. Does not work on AMD Bulldozer processors. Fixed in gwnum 27.5 (there will not be a prime95 27.5).
2. Length 1280K FFT crashes on Pentium 4s with 512K L2 cache or less. Fixed in 27.6.
3. Hyperthread detection is not working properly. Fixed in 27.6.
4. The 6K 2^N-1 and 8K, 10K, 12K, 16K 2^N+1 AVX FFTs are missing. Fixed in 27.6.
5. Some exponents above 536 million generate roundoff errors and incorrect results. Fixed in 27.6.
6. FFT crossover points need retuning. Many can be more aggressive, but a few need to be more conservative. Fixed in 27.6.

Dubslow · 2012-03-09, 03:37

No point in going from 27.3-.4 for production GIMPS?

Prime95 · 2012-03-09, 03:59

Quote:

Originally Posted by Dubslow

No point in going from 27.3-.4 for production GIMPS?

Correct.

Xyzzy · 2012-03-09, 04:05

The mersenneforum mirror has been synced.

flashjh · 2012-03-09, 04:15

Quote:

Originally Posted by Prime95

Prime95 beta version 27.4 is available.

Hopefully, Bulldozer processors now properly run SSE2

Tried with Opteron 6272:

All 8 workers reported a FFT error.

Code:

[Main thread Mar 8 21:22] Mersenne number primality test program version 27.4
[Main thread Mar 8 21:22] Optimizing for CPU architecture: AMD Bulldozer, L2 cache size: 2 MB, L3 cache size: 12 MB
 
 
[Thu Mar 08 21:12:20 2012]
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002

Dubslow · 2012-03-09, 04:49

Were they trying AVX or SSE FFTs?

flashjh · 2012-03-09, 04:58

Quote:

Originally Posted by Dubslow

Were they trying AVX or SSE FFTs?

I don't know because it doesn't tell me. It fails before that step. Here's a worker example:

Code:

[Mar 8 21:56] Worker starting
[Mar 8 21:56] Setting affinity to run worker on any logical CPU.
[Mar 8 21:56] Optimal P-1 factoring of M52859099 using up to 117950MB of memory.
[Mar 8 21:56] Assuming no factors below 2^72 and 2 primality tests saved if a factor is found.
[Mar 8 21:56] Optimal bounds are B1=520000, B2=10920000
[Mar 8 21:56] Chance of finding a factor is an estimated 4.03%
[Mar 8 21:56] Cannot initialize FFT code, errcode=1002
[Mar 8 21:56] Worker stopped.

It properly detects the Bulldozer now, but the L3 cache is still wrong. I set it manually in 26.6 with CpuL3CacheSize=16384.

If I try running a torture test (on 1 worker blend test) this is what I get:

Code:

[Mar 8 21:59] Worker starting
[Mar 8 21:59] Setting affinity to run worker on any logical CPU.
[Mar 8 21:59] Beginning a continuous self-test to check your computer.
[Mar 8 21:59] Please read stress.txt.  Choose Test/Stop to end this test.
[Mar 8 21:59] No FFT lengths available in the range specified.
[Mar 8 21:59] Worker stopped.

Prime95 · 2012-03-09, 15:56

Quote:

Originally Posted by flashjh

Tried with Opteron 6272:

Sigh. What OS are you using? Can I email or sendspace versions for you to debug/test?

flashjh · 2012-03-09, 17:29

Quote:

Originally Posted by Prime95

Sigh. What OS are you using? Can I email or sendspace versions for you to debug/test?

No problem. I use Win7 64. I'll PM my info.

Jerry

Prime95 · 2012-03-09, 20:09

Quote:

Originally Posted by flashjh

Tried with Opteron 6272:

Code:

AMD Bulldozer, L2 cache size: 2 MB, L3 cache size: 12 MB

The code for determining L3 cache size looks OK. I don't suppose you could write a little program that outputs the EDX register after a CPUID (0x80000006) instruction.

Note: Prime95 doesn't use this information, so it is hardly a critical bug.

2012-03-09, 02:57	#1
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 1110101101110₂ Posts	Prime95 version 27.4 Prime95 beta version 27.4 is available. This version fixes a number of bugs in 27.3 - mostly related to LLR and PFGW issues. There are no speed improvements over 27.3. Hopefully, Bulldozer processors now properly run SSE2 FFTs. Unless you have a Sandy Bridge CPU, there is little reason to upgrade to this version. Download links: Windows 64-bit: ftp://mersenne.org/gimps/p64v274.zip Windows 32-bit: ftp://mersenne.org/gimps/p95v274.zip Linux 64-bit: ftp://mersenne.org/gimps/mprime274-linux64.tar.gz Linux 32-bit: ftp://mersenne.org/gimps/mprime274.tar.gz Mac OS X: ftp://mersenne.org/gimps/Prime95-MacOSX-274.zip FreeBSD 64-bit: ftp://mersenne.org/gimps/mprime274-FreeBSD64.tar.gz Source code: ftp://mersenne.org/gimps/source274.zip I have one more significant change planned that should speed up multithreaded FFTs somewhat. Last fiddled with by Prime95 on 2012-03-09 at 19:54

2012-03-09, 02:57	#2
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2·3,767 Posts	Known bugs and fixes: 1. Does not work on AMD Bulldozer processors. Fixed in gwnum 27.5 (there will not be a prime95 27.5). 2. Length 1280K FFT crashes on Pentium 4s with 512K L2 cache or less. Fixed in 27.6. 3. Hyperthread detection is not working properly. Fixed in 27.6. 4. The 6K 2^N-1 and 8K, 10K, 12K, 16K 2^N+1 AVX FFTs are missing. Fixed in 27.6. 5. Some exponents above 536 million generate roundoff errors and incorrect results. Fixed in 27.6. 6. FFT crossover points need retuning. Many can be more aggressive, but a few need to be more conservative. Fixed in 27.6. Last fiddled with by Prime95 on 2012-04-23 at 05:14

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Prime95 version 27.3	Prime95	Software	148	2012-03-18 19:24
Prime95 version 26.3	Prime95	Software	76	2010-12-11 00:11
Prime95 version 25.5	Prime95	PrimeNet	369	2008-02-26 05:21
Prime95 version 25.4	Prime95	PrimeNet	143	2007-09-24 21:01
When the next prime95 version ?	pacionet	Software	74	2006-12-07 20:30

2012-03-09, 03:37	#3
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 16065₈ Posts	No point in going from 27.3-.4 for production GIMPS?

2012-03-09, 04:05	#5
Xyzzy "Mike" Aug 2002 2×23×179 Posts	The mersenneforum mirror has been synced.

2012-03-09, 04:49	#7
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3×29×83 Posts	Were they trying AVX or SSE FFTs?