Prime95 v25.3 - Page 2

BranMuffin · 2007-07-10, 21:41

Installed on a machine with dual 3.0 Ghz Irwindales (single core HT Xeons). Automatic worker thread settings had it only running on 2 processors, but didn't appear to be using the *virtual* HT processors. According to the CPU Usage view in the task manager, only processors 0 and 2 were being utilized (100%). Had to fiddle with the settings to get it to use all processors.

Prime95 · 2007-07-10, 22:53

Quote:

Originally Posted by BranMuffin

Installed on a machine with dual 3.0 Ghz Irwindales (single core HT Xeons). Automatic worker thread settings had it only running on 2 processors, but didn't appear to be using the *virtual* HT processors. According to the CPU Usage view in the task manager, only processors 0 and 2 were being utilized (100%). Had to fiddle with the settings to get it to use all processors.

Isn't Irwindale hyperthreaded? If so, the default is to run only one worker thread.

BranMuffin · 2007-07-10, 23:52

I'm not familiar enough with the worker thread defaults. I noted on a Prescott system v25.3 installed and wanted to use 1 worker thread on 2 CPUs, whereas a Presler system wanted to use 2 worker threads on 1 CPU. I guess by that logic, the Irwindale had defaulted properly, but I did note that after install on both the Prescott and the Presler system, the CPU utilization was 100% on both "CPUs" (main and virtual on the Prescott, both cores on the Presler). On the Irwindale, it was only running 100% on CPUs 0 and 2.

Am I the confused one, or am I just trying to confuse everyone else?

drew · 2007-07-11, 05:16

I just tried this on a test that was formerly being run in v. 25.1. I got the following:

Error reading intermediate file: p4B02589
Error reading intermediate file: pY868357
Renaming intermediate file q4B02589 to p4B02589.
Error reading intermediate file: p4B02589
Renaming intermediate file qY868357 to pY868357.
Error reading intermediate file: pY868357

Version 25.1 still works with these files.

Drew

Prime95 · 2007-07-11, 13:12

Quote:

Originally Posted by drew

I just tried this on a test that was formerly being run in v. 25.1. I got the following:
Error reading intermediate file: p4B02589
Version 25.1 still works with these files.

Hmm. 25.3 should be able to read version 24 save files, but I did not put any effort into reading 25.1 save files as very few people are running that version. The only solution is to complete the assignments in 25.1.

drew · 2007-07-11, 14:21

Quote:

Originally Posted by Prime95

Hmm. 25.3 should be able to read version 24 save files, but I did not put any effort into reading 25.1 save files as very few people are running that version. The only solution is to complete the assignments in 25.1.

No problem. Those assignments will be done in a week, anyway. Thanks for the response.

Drew

monst · 2007-07-11, 14:35

Regarding v25.1 -- I've been running in on a couple of machines. My question is for George regarding factoring types of work. 25.1 had two selections: "Trial factoring to low limits" and "Trial factoring". Although that selection didn't seem to always work properly in 25.1, I believe the idea is to either get a candidate and factor it 1 bit at a time or get a candidate and factor it to the bit level determined before the candidate should switch over to LL testing.

What is you plan regarding this for version 25? I see 25.3 does not have the two selections, but each candidate is reporting back to the server after each bit is completed -- I like this idea. Also, is 25.3 giving any factoring credit?

Prime95 · 2007-07-11, 16:51

Quote:

Originally Posted by monst

What is you plan regarding this for version 25? I see 25.3 does not have the two selections, but each candidate is reporting back to the server after each bit is completed -- I like this idea. Also, is 25.3 giving any factoring credit?

Version 25.3, when talking to the new server, will have two factoring types. Factoring small exponents (right now in the 40-45 million range) and factoring large exponents (79.3 million and up). Factoring will be done one bit level at a time.

The reason for 2 trial factoring work types is that slow computers can work on large exponents to low bit levels and return a result in a reasonable time frame. See this forum's LMH sub-forum.

If you use version 25 to talk to the old server you should get factoring credit.

S00113 · 2007-07-11, 17:02

A segmentation fault which may be connected to the end of P--1 stage 2:

Code:

[Worker #3 Jun 28 18:23] Resuming.
[Worker #3 Jun 28 18:23] No work to do at the present time.  Waiting.
[Worker #4 Jun 28 18:26] Available memory is 800MB.
[Worker #4 Jun 28 18:26] Using 217MB of memory.  Processing 4 relative primes (476 of 480 already processed).
[Worker #2 Jun 28 18:31] Iteration: 4450000 / 19582247 [22.72%].  Per iteration time: 0.041 sec.
[Worker #4 Jun 28 18:37] M39041309 stage 2 is 100.00% complete. Time: 1548.279 sec.
[Worker #1 Jun 28 18:44] Iteration: 4450000 / 19582231 [22.72%].  Per iteration time: 0.045 sec.
[Worker #4 Jun 28 19:03] M39041309 stage 2 is 100.00% complete. Time: 1552.869 sec.
[Worker #2 Jun 28 19:05] Iteration: 4500000 / 19582247 [22.97%].  Per iteration time: 0.039 sec.
[Worker #4 Jun 28 19:20] M39041309 stage 2 complete. 1861912 transforms. Time: 133754.016 sec.
[Worker #4 Jun 28 19:20] Starting stage 2 GCD - please be patient.
[Worker #1 Jun 28 19:20] Iteration: 4500000 / 19582231 [22.98%].  Per iteration time: 0.041 sec.
Segmentation fault

The error did not reproduce when resuming from a save file. I'm using the 64bit version on RedHat RHEL 4. Why only 217 MB in stage 2, btw? The machine has 6 GB of RAM, and I have configured mprime to use 2048/4096 day/night.

Bent · 2007-07-17, 17:42

I finally decided to give this alpha a shot on some of my multi cpu and hyperthreaded machines.

First thing I noticed is that the program has a hard time picking up all current assignments if you were running multiple instances of version 24. The first worker thread gets the assignment just fine and resumes, but the other worker threads get new assignments without me having a chance to stop them :) Any assignments in work0001.ini, etc, are not carried over.

Other than that, no problems so far, will start testing some of the things in the first post later this week. I did notice my hyperthreaded machine dropped its iteration time down to about 65 ms from about 72 ms, not bad :) All my cores and hyperthreading were correctly detected. Love the new interface, having all my tests in a single window is also great. One thing though, I used to be able to stop a single instance of version 24 if I wanted to, I used to do this to make an assignment finish up slightly faster, or for testing reasons. Will there be a way to stop an individual worker thread in v25? I think the current wording is fine, but I also like the "Job" suggestion.

Looking forward to the final, are you planning to have a few more alpha/beta versions first or is it getting close to finished?

rayrob · 2007-07-17, 21:50

Any chance of a Windows x64 build to try?
I'm just finishing my last exponent before I try out 25.3 (will be done tomorrow). I'm running Vista64 on a dual core.

I'll use the 32 bit version now but would be keen to see a x64 build.

2007-07-10, 21:41	#12
BranMuffin Dec 2002 12₈ Posts	Installed on a machine with dual 3.0 Ghz Irwindales (single core HT Xeons). Automatic worker thread settings had it only running on 2 processors, but didn't appear to be using the virtual HT processors. According to the CPU Usage view in the task manager, only processors 0 and 2 were being utilized (100%). Had to fiddle with the settings to get it to use all processors. Last fiddled with by BranMuffin on 2007-07-10 at 21:48

2007-07-10, 23:52	#14
BranMuffin Dec 2002 2·5 Posts	I'm not familiar enough with the worker thread defaults. I noted on a Prescott system v25.3 installed and wanted to use 1 worker thread on 2 CPUs, whereas a Presler system wanted to use 2 worker threads on 1 CPU. I guess by that logic, the Irwindale had defaulted properly, but I did note that after install on both the Prescott and the Presler system, the CPU utilization was 100% on both "CPUs" (main and virtual on the Prescott, both cores on the Presler). On the Irwindale, it was only running 100% on CPUs 0 and 2. Am I the confused one, or am I just trying to confuse everyone else?

2007-07-11, 05:16	#15
drew Jun 2005 2×191 Posts	I just tried this on a test that was formerly being run in v. 25.1. I got the following: Error reading intermediate file: p4B02589 Error reading intermediate file: pY868357 Renaming intermediate file q4B02589 to p4B02589. Error reading intermediate file: p4B02589 Renaming intermediate file qY868357 to pY868357. Error reading intermediate file: pY868357 Version 25.1 still works with these files. Drew

2007-07-11, 14:35	#18
monst Mar 2007 179 Posts	Regarding v25.1 -- I've been running in on a couple of machines. My question is for George regarding factoring types of work. 25.1 had two selections: "Trial factoring to low limits" and "Trial factoring". Although that selection didn't seem to always work properly in 25.1, I believe the idea is to either get a candidate and factor it 1 bit at a time or get a candidate and factor it to the bit level determined before the candidate should switch over to LL testing. What is you plan regarding this for version 25? I see 25.3 does not have the two selections, but each candidate is reporting back to the server after each bit is completed -- I like this idea. Also, is 25.3 giving any factoring credit?

2007-07-17, 17:42	#21
Bent Jul 2007 2·7 Posts	I finally decided to give this alpha a shot on some of my multi cpu and hyperthreaded machines. First thing I noticed is that the program has a hard time picking up all current assignments if you were running multiple instances of version 24. The first worker thread gets the assignment just fine and resumes, but the other worker threads get new assignments without me having a chance to stop them :) Any assignments in work0001.ini, etc, are not carried over. Other than that, no problems so far, will start testing some of the things in the first post later this week. I did notice my hyperthreaded machine dropped its iteration time down to about 65 ms from about 72 ms, not bad :) All my cores and hyperthreading were correctly detected. Love the new interface, having all my tests in a single window is also great. One thing though, I used to be able to stop a single instance of version 24 if I wanted to, I used to do this to make an assignment finish up slightly faster, or for testing reasons. Will there be a way to stop an individual worker thread in v25? I think the current wording is fine, but I also like the "Job" suggestion. Looking forward to the final, are you planning to have a few more alpha/beta versions first or is it getting close to finished?

2007-07-17, 21:50	#22
rayrob Mar 2007 UK 2 Posts	Any chance of a Windows x64 build to try? I'm just finishing my last exponent before I try out 25.3 (will be done tomorrow). I'm running Vista64 on a dual core. I'll use the 32 bit version now but would be keen to see a x64 build.