Prime95 version 29.6/29.7/29.8

[Falkentyne]

Originally Posted by Prime95

I need to know your CPU architecture as reported in the main window at startup. I'll need caches sizes as reported by CPU dialog box. How many torture test threads were you running? What were the FFT sizes shown in the greyed area of the torture test dialog box?

I can confirm this bug.
i9 9900K.

Bug:
Clicking "Disable AVX" but NOT disable AVX2 on large FFT and on Blend, causes "No FFT's in range" error.

Clicking "disable AVX2 *AND* Disable AVX does not cause the error.
I can only assume (Guessing) that doing this with custom ranges may fail also.

[Prime95]

Originally Posted by Falkentyne

Clicking "Disable AVX" but NOT disable AVX2 on large FFT and on Blend, causes "No FFT's in range" error.

Grrr, I had not contemplated turning off AVX but not AVX2. In fact, using the Linux menus this is impossible to do.

[ric]

Two minor points with respect to 29.6.2 (mprime, 64bits):

• PRP-CF results are written to json file, but no longer appear in results.txt (see attach 1 & 2) - in full disclosure, I've a different name for results.txt file, via a specific prime.txt entry
• PRP-CF results are parsed by the server (so probably this observation should go to another thread) as if the result were "regular" PRP tests (see M20585021, where I tested M20585021/factor, as per json file)

[GP2]

Originally Posted by ric

PRP-CF results are written to json file, but no longer appear in results.txt

This is desirable behavior, it's redundant otherwise.

PRP-CF results have been output exclusively in JSON format since version 29.4 build 2 in October 2017, so it's not a matter of writing new parsing code, you just need to look for them in a different file.

Factoring results do get written to both files (in different formats). That makes sense, since people looking for non-Mersenne factors that don't get reported to Primenet might have their own code that parses the old format.

PRP-CF results are parsed by the server (so probably this observation should go to another thread) as if the result were "regular" PRP tests (see M20585021, where I tested M20585021/factor, as per json file)

It does get parsed correctly. The log entry in History no longer displays the list of factors, but the result goes into the PRP Cofactor section.

[GP2]

For the very smallest exponents, there is way too much output:

I am using ScaleOutputFrequency=1 in prime.txt but it doesn't help.

In a matter of minutes, the log file grew to 800 MB.

ECM2=1,2,1087,1,43000000,43000000,8700,"3,121398914459,144288053938906556201,59015046142680340633"

[Main thread Feb 22 18:58] Starting worker.
[Work thread Feb 22 18:58] Worker starting
[Work thread Feb 22 18:58] Using all-complex FMA3 FFT length 64
[Work thread Feb 22 18:58] ECM on 2^1087+1: curve #1 with s=2147430289016201, B1=43000000, B2=43000000
[Work thread Feb 22 18:58] 2^1087+1 curve 1 stage 1 at prime 3 [0.00%]. Time: 0.000 sec.
[Work thread Feb 22 18:58] 2^1087+1 curve 1 stage 1 at prime 5 [0.00%]. Time: 0.000 sec.
[Work thread Feb 22 18:58] 2^1087+1 curve 1 stage 1 at prime 7 [0.00%]. Time: 0.000 sec.
[Work thread Feb 22 18:58] 2^1087+1 curve 1 stage 1 at prime 11 [0.00%]. Time: 0.000 sec.
[Work thread Feb 22 18:58] 2^1087+1 curve 1 stage 1 at prime 13 [0.00%]. Time: 0.000 sec.
[Work thread Feb 22 18:58] 2^1087+1 curve 1 stage 1 at prime 17 [0.00%]. Time: 0.000 sec.
[Work thread Feb 22 18:58] 2^1087+1 curve 1 stage 1 at prime 19 [0.00%]. Time: 0.000 sec.
[Work thread Feb 22 18:58] 2^1087+1 curve 1 stage 1 at prime 23 [0.00%]. Time: 0.000 sec.
[Work thread Feb 22 18:58] 2^1087+1 curve 1 stage 1 at prime 29 [0.00%]. Time: 0.000 sec.
[Work thread Feb 22 18:58] 2^1087+1 curve 1 stage 1 at prime 31 [0.00%]. Time: 0.000 sec.
[Work thread Feb 22 18:58] 2^1087+1 curve 1 stage 1 at prime 37 [0.00%]. Time: 0.000 sec.
...

By contrast, for slightly larger exponents and identical prime.txt, I get:

ECM2=1,2,1201,1,43000000,43000000,8700,"3,107117096323,12227544337,1924003,2983620281"

[Main thread Feb 22 19:15] Starting worker.
[Work thread Feb 22 19:15] Worker starting
[Work thread Feb 22 19:15] Using all-complex FMA3 FFT length 64
[Work thread Feb 22 19:15] ECM on 2^1201+1: curve #1 with s=1747100447905230, B1=43000000, B2=43000000
[Work thread Feb 22 19:16] 2^1201+1 curve 1 stage 1 at prime 3832079 [8.91%]. Time: 12.277 sec.
[Work thread Feb 22 19:16] 2^1201+1 curve 1 stage 1 at prime 7658591 [17.81%]. Time: 12.285 sec.
[Work thread Feb 22 19:16] 2^1201+1 curve 1 stage 1 at prime 11480467 [26.69%]. Time: 12.281 sec.
[Work thread Feb 22 19:16] 2^1201+1 curve 1 stage 1 at prime 15299671 [35.58%]. Time: 12.279 sec.
[Work thread Feb 22 19:17] 2^1201+1 curve 1 stage 1 at prime 19118647 [44.46%]. Time: 12.263 sec.
[Work thread Feb 22 19:17] 2^1201+1 curve 1 stage 1 at prime 22936033 [53.33%]. Time: 12.273 sec.
[Work thread Feb 22 19:17] 2^1201+1 curve 1 stage 1 at prime 26750747 [62.21%]. Time: 12.263 sec.
[Work thread Feb 22 19:17] 2^1201+1 curve 1 stage 1 at prime 30561731 [71.07%]. Time: 12.265 sec.
[Work thread Feb 22 19:17] 2^1201+1 curve 1 stage 1 at prime 34377377 [79.94%]. Time: 12.265 sec.
[Work thread Feb 22 19:18] 2^1201+1 curve 1 stage 1 at prime 38188457 [88.81%]. Time: 12.261 sec.
[Work thread Feb 22 19:18] 2^1201+1 curve 1 stage 1 at prime 41999567 [97.67%]. Time: 12.263 sec.
[Work thread Feb 22 19:18] Stage 1 complete. 1126236177 transforms, 1 modular inverses. Time: 3.220 sec.
[Work thread Feb 22 19:18] Stage 1 GCD complete. Time: 0.000 sec.
[Work thread Feb 22 19:18] ECM on 2^1201+1: curve #2 with s=5405473801738053, B1=43000000, B2=43000000
[Work thread Feb 22 19:18] 2^1201+1 curve 2 stage 1 at prime 3832067 [8.91%]. Time: 12.276 sec.

and also (note different B1, B2):

ECM2=1,2,2011,1,11000000,11000000,5200,"3,328434864016035321284153784239230639028291"

[Main thread Feb 22 18:22] Starting worker.
[Work thread Feb 22 18:22] Worker starting
[Work thread Feb 22 18:22] Using all-complex FMA3 FFT length 96
[Work thread Feb 22 18:22] ECM on 2^2011+1: curve #1 with s=491274290942190, B1=11000000, B2=11000000
[Work thread Feb 22 18:23] 2^2011+1 curve 1 stage 1 at prime 3835597 [34.86%]. Time: 16.495 sec.
[Work thread Feb 22 18:23] 2^2011+1 curve 1 stage 1 at prime 7661827 [69.65%]. Time: 16.523 sec.
[Work thread Feb 22 18:23] Stage 1 complete. 287339925 transforms, 1 modular inverses. Time: 14.418 sec.
[Work thread Feb 22 18:23] Stage 1 GCD complete. Time: 0.000 sec.
[Work thread Feb 22 18:23] ECM on 2^2011+1: curve #2 with s=3310755350124869, B1=11000000, B2=11000000
[Work thread Feb 22 18:24] 2^2011+1 curve 2 stage 1 at prime 3835577 [34.86%]. Time: 16.588 sec.
[Work thread Feb 22 18:24] 2^2011+1 curve 2 stage 1 at prime 7661821 [69.65%]. Time: 16.609 sec.
[Work thread Feb 22 18:24] Stage 1 complete. 287339925 transforms, 1 modular inverses. Time: 14.493 sec.
[Work thread Feb 22 18:24] Stage 1 GCD complete. Time: 0.000 sec.

[Prime95]

Originally Posted by ric

[*]PRP-CF results are written to json file, but no longer appear in results.txt

See undoc.txt:

Setting	OutputPrimes=1 will cause prime95 to output a non-JSON message for probable primes to results.txt.
Setting	OutputComposites=1 will cause prime95 to output a non-JSON message for composite PRP tests to results.txt.
Setting	OutputJSON=0 will cause prime95 to not output the JSON messages to results.json.txt.

[Mark Rose]

How hard would it be to include a benchmark mode that is run from the cli without any interaction? I think mprime would make a nice addition to the Phoronix Test Suite. It would give us benchmarks are much more hardware.

We'd probably want benchmarks for single core and worker-per-numa-node. We'd probably want to test a single FFT for time, but also pick an FFT that would be applicable for years. Maybe 8192k?

It would also be nice if there were a parameter to automatically submit the results to PrimeNet, too.

[Madpoo]

Originally Posted by Prime95

Can you give Aaron an exponent to look at? I'm guessing he is parsing the new JSON format to create the web page.

Found it... snippet of the JSON looks like:

"b1":330000, "b2":8745000, "brent-suyama":12

I just need to snag that "brent-suyama" value if present and display it as the E.

[Madpoo]

Originally Posted by Madpoo

Found it... snippet of the JSON looks like:

"b1":330000, "b2":8745000, "brent-suyama":12

I just need to snag that "brent-suyama" value if present and display it as the E.

Done... see this for an example:
M5337373

[harlee]

Found an issue with the Windows 32-bit version. Noticed that 29.6b1 is running much slower then 29.4b7. Yes, I know I should stop using this system - I'm using it to do P1 work on smaller exponents with low B1 & B2 bounds..

29.4b7

[Feb 23 12:22] Using Pentium4 FFT length 288K, Pass1=384, Pass2=768, clm=2

29.6b1

[Feb 23 20:17] Using Pentium4 FFT length 288K, Pass1=384, Pass2=768, clm=4

Noticed that 29.6b1 selected a different "clm".

This morning 29.6b1 ran another benchmark. Notice the benchmark was for FFTlen=288K while the P1 was using a FFTlen=320K.

[Feb 24 05:19] Optimal P-1 factoring of M5339599 using up to 2048MB of memory.
[Feb 24 05:19] Assuming no factors below 2^62 and 10 primality tests saved if a factor is found.
[Feb 24 05:19] Optimal bounds are B1=320000, B2=9520000
[Feb 24 05:19] Chance of finding a factor is an estimated 7.47%
[Feb 24 05:19] Using Pentium4 type-0 FFT length 320K, Pass1=320, Pass2=1K, clm=4

<SNIP>

[Feb 24 05:54] M5339599 stage 1 is 43.31% complete. Time: 105.074 sec.
[Feb 24 05:56] Worker stopped while running needed benchmarks.
[Feb 24 05:58] Benchmarks complete, restarting worker.
[Feb 24 05:58] Optimal P-1 factoring of M5339599 using up to 2048MB of memory.
[Feb 24 05:58] Assuming no factors below 2^62 and 10 primality tests saved if a factor is found.
[Feb 24 05:58] Optimal bounds are B1=320000, B2=9520000
[Feb 24 05:58] Chance of finding a factor is an estimated 7.47%
[Feb 24 05:58] Using Pentium4 type-0 FFT length 320K, Pass1=320, Pass2=1K, clm=4

[Sun Feb 24 05:56:40 2019]
FFTlen=288K, Type=3, Arch=3, Pass1=384, Pass2=768, clm=4 (1 core, 1 worker): 11.31 ms.  Throughput: 88.45 iter/sec.
FFTlen=288K, Type=3, Arch=2, Pass1=384, Pass2=768, clm=4 (1 core, 1 worker): 11.27 ms.  Throughput: 88.71 iter/sec.
FFTlen=288K, Type=3, Arch=1, Pass1=384, Pass2=768, clm=2 (1 core, 1 worker): 10.07 ms.  Throughput: 99.32 iter/sec.
FFTlen=288K, Type=3, Arch=2, Pass1=384, Pass2=768, clm=2 (1 core, 1 worker): 11.37 ms.  Throughput: 87.93 iter/sec.
FFTlen=288K, Type=2, Arch=3, Pass1=384, Pass2=768, clm=4 (1 core, 1 worker): 11.39 ms.  Throughput: 87.77 iter/sec.
FFTlen=288K, Type=2, Arch=2, Pass1=384, Pass2=768, clm=4 (1 core, 1 worker): 11.37 ms.  Throughput: 87.96 iter/sec.
FFTlen=288K, Type=2, Arch=1, Pass1=384, Pass2=768, clm=2 (1 core, 1 worker): 10.09 ms.  Throughput: 99.07 iter/sec.
FFTlen=288K, Type=2, Arch=2, Pass1=384, Pass2=768, clm=2 (1 core, 1 worker): 11.43 ms.  Throughput: 87.48 iter/sec.

[harlee]

Originally Posted by Madpoo

Done... see this for an example:
M5337373

Madpoo, Please look at M18521411 as 29.6b1 found and reported a a factor but the P1 bounds are not being shown.

[Sat Feb 23 14:42:47 2019]
{"status":"F", "exponent":18521411, "worktype":"P-1", "factors":"165718901219828172167153", "b1":330000, "b2":8745000, "brent-suyama":12, "fft-length":983040, "security-code":"DFB5B559", "program":{"name":"Prime95", "version":"29.6", "build":1, "port":10}, "timestamp":"2019-02-23 19:42:47", "user":"harlee", "computer":"i5-5250U_1600", "aid":"<deleted>"}