mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Software (https://www.mersenneforum.org/forumdisplay.php?f=10)
-   -   Early Beta of version 24.11 (https://www.mersenneforum.org/showthread.php?t=3934)

 ET_ 2005-04-12 11:57

Benchmark timings

George, I have a dumb question about v24.11

Looking at the factoring benchmark, I noticed that the best time results are printed in ms.

Assuming that there are about 1,029,407,319 k from 57 to 58 bits (exponent 35000011), and that about 95% of those k are sieved out, we have about 51,470,366 k to check.

Prime95 runs the factoring work between 57 and 58 bits in about 15,450 ms and 58-59 in 31,500 ms (time of Prime95); the benchmark shows a timing of about 7 ms for bit depth comprised between 58 and 61 bits, so I assume the same for 57 bits.

The "one iteration time" in factoring should be 15,450/51,470,366 = 30*10^-4 ms; 7 ms are worth about 23,320 iterations.

As a demonstration, if we divide k by 23,320 and multiply the result by 7, we again have 15,450 ms

May I ask you why did you choose 23,320 iterations for the benchmark? :rolleyes:

Luigi

 Prime95 2005-04-12 13:36

You lost me.

The 32-bit factoring assembly code processes a 16KB sieve. This is "one iteration" in the benchmark. When doing trial factoring you control how often prime95 writes to the screen in Options / Preferences.

The 64-bit factoring assembly code processes a 48KB sieve.
This is "one iteration". The benchmarking code divides this time by 3 so you can properly compare it to 32-bit machines.

 Frodo42 2005-04-12 22:13

I seeem to be having a problem with sprime when I try to factor a number that requires FFT length of 1024K

I get the following message:
[code]
Starting P-1 factoring with B1=40000, B2=420000
Chance of finding a factor is an estimated 0.92%
P-1 on 55459*2^9011278+1 with B1=40000, B2=420000
Using FFT length 1024K
SUMOUT error occurred.
[/code]

There doesn't seem to be any problems P-1 factoring numbers with lower FFT length (even 1024 zero-padded)
I am testing this on another machine just now ... I haven't had any hardware problems with this computer so far.

Anyhow just thought I would report this ... it may just be a hardware error on my side.

 ET_ 2005-04-12 22:26

[QUOTE=Prime95]You lost me.

The 32-bit factoring assembly code processes a 16KB sieve. This is "one iteration" in the benchmark. When doing trial factoring you control how often prime95 writes to the screen in Options / Preferences.

The 64-bit factoring assembly code processes a 48KB sieve.
This is "one iteration". The benchmarking code divides this time by 3 so you can properly compare it to 32-bit machines.[/QUOTE]

I'm lost too :surrender

I'd better look into Prime95 source code before asking dumb questions.

Luigi

 Frodo42 2005-04-12 22:40

[QUOTE=Frodo42]I am testing this on another machine just now ... [/QUOTE]

It reproduced the error message, so I guess it's not a hardware failure.

 Prime95 2005-04-13 00:43

[QUOTE=Frodo42]It reproduced the error message, so I guess it's not a hardware failure.[/QUOTE]

Is this version 24.11 that is doing this? When I try that number on a P4 I get a zero-padded 1024K FFT and no errors.

 Frodo42 2005-04-13 04:36

[QUOTE=Prime95]Is this version 24.11 that is doing this? When I try that number on a P4 I get a zero-padded 1024K FFT and no errors.[/QUOTE]

Yes it's from sprime2411.tar.gz

The line I have in worktodo.ini is:
[code]
Pfactor=55459,2,9011278,1,49,1.5
[/code]
My local.ini:
[code]
OldCpuType=12
OldCpuSpeed=3017
DayMemory=768
NightMemory=768
CPUHours=24
DayStartTime=480
DayEndTime=1320
Pid=0
RollingAverage=1000
RollingStartTime=0
[/code]

 Frodo42 2005-04-13 08:07

[QUOTE=Prime95]Is this version 24.11 that is doing this? [/QUOTE]

No I am wrong ... it's not version 2.4.11 even though I thought it was. It was an older version.
Sorry about all this fuss :redface:
It works fine when i actually use ver. 2.4.11

The problem arose because I wrongly assumed that the static binary was called sprime, while it is actually mprime ... I had an sprime binary in the directory from an old package that contained both the sprime and mprime.

 Prime95 2005-04-18 23:18

I've just uploaded my third attempt at 24.11.

The only new feature is I've spent time optimizing the SSE2 macros further. P4s will see about a 1% boost and AMD64s will see about a 2.5% boost.

 PhilF 2005-04-18 23:40

Earlier in this thread you mentioned only AMD64 machines should run this version. Is it now OK for P4 machines to run this version on exponents currently being tested by version 23.8?

 Prime95 2005-04-19 00:03

[QUOTE=PhilF]Earlier in this thread you mentioned only AMD64 machines should run this version. Is it now OK for P4 machines to run this version on exponents currently being tested by version 23.8?[/QUOTE]

It is still beta software. You should be OK, but no guarantees. You must ask yourself if it is worth the small risk for the small speed gain.

All times are UTC. The time now is 08:46.