Prime95 version 29.4

[ATH]

You can also decrease the time between Jacobi checks:

You can control how often Jacobi error checking is run. Default is 12 hours.
If a Jacobi test takes 30 seconds, then the default represents an overhead of
30 / (12 * 60 * 60) or 0.07% overhead. Each Jacobi test has a 50% chance of
discovering if a hardware error has occured in the last time interval. In prime.txt:
JacobiErrorCheckingInterval=N (default is 12)
where N is in hours.

[kriesel]

When no factor is found, the report includes the E value that was used.
When a factor is found, apparently not. (I checked the worker window, results.txt, and prime.log)

[James Heinrich]

Originally Posted by kriesel

When no factor is found, the report includes the E value that was used. When a factor is found, apparently not

"E" is only relevant in stage 2. If the factor was found in stage1 then E is not applicable and therefore not reported.

P-1 found a factor in stage #1, B1=730000.
M89565797 has a factor: 164493217479527458358561 (P-1, B1=730000)
P-1 found a factor in stage #2, B1=730000, B2=14782500, E=12.
M89565907 has a factor: 16352015139068430008287498903 (P-1, B1=730000, B2=14782500, E=12)

E will be reported for all no-factor results because both stage1+2 were run.

[kriesel]

Originally Posted by James Heinrich

"E" is only relevant in stage 2. If the factor was found in stage1 then E is not applicable and therefore not reported.

P-1 found a factor in stage #1, B1=730000.
M89565797 has a factor: 164493217479527458358561 (P-1, B1=730000)
P-1 found a factor in stage #2, B1=730000, B2=14782500, E=12.
M89565907 has a factor: 16352015139068430008287498903 (P-1, B1=730000, B2=14782500, E=12)

E will be reported for all no-factor results because both stage1+2 were run.

Sorry, I should have specified that I was certain the factor in question without an E value indicated was from completion of stage 2 in prime95 V29.4b8. I've been eagerly watching for its completion, since it's part of a set I'm running to measure run time scaling of P-1 in prime95 for completion through stage 2.

[James Heinrich]

You only have 4GB available which is probably insufficient to run the extension for that exponent.
Using the approximate values returned by on my P-1 Probability page, I'd expect that you'd need to allocate at least 8GB before the extension gets used.

I'm not sure about the intricacies involved in how Prime95 selects the number of relative primes (typically 480, sometimes 960 as per your screenshot, sometimes 192) and how it relates to the choice of whether to use the Brent-Suyama extension or not. Perhaps in this case it was better to use 960 and skip the extension. George would be able to better answer those details.

[kriesel]

Originally Posted by James Heinrich

You only have 4GB available which is probably insufficient to run the extension for that exponent.
Using the approximate values returned by on my P-1 Probability page, I'd expect that you'd need to allocate at least 8GB before the extension gets used.

I'm not sure about the intricacies involved in how Prime95 selects the number of relative primes (typically 480, sometimes 960 as per your screenshot, sometimes 192) and how it relates to the choice of whether to use the Brent-Suyama extension or not. Perhaps in this case it was better to use 960 and skip the extension. George would be able to better answer those details.

Total installed RAM on the system is 8GB. Paging was excessive at 7.2GB or 6GB; it stopped at 4GB. Which has seemed adequate on gpus. It should be in good shape around 100M though, per your calculator.
I guess I'm too accustomed to CUDAPm1 indicating E=12, 6, or 2, even on a 4GB GTX 1050TI up to 383M+ in stage 2, or 0 for a stage 1 result printed to the console, such as for the following samples. Maybe they're trading off bounds and extension differently.

M85320343 Stage 1 found no factor (P-1, B1=735000, B2=17272500, e=0, n=4704K CUDAPm1 v0.20)
M85320353 Stage 1 found no factor (P-1, B1=735000, B2=17272500, e=6, n=4704K CUDAPm1 v0.20)
M85343233 Stage 2 found no factor (P-1, B1=735000, B2=17272500, e=6, n=4704K CUDAPm1 v0.20)
M289999981 Stage 1 found no factor (P-1, B1=2280000, B2=53010000, e=0, n=16384K CUDAPm1 v0.20)
M289999981 Stage 2 found no factor (P-1, B1=2280000, B2=53010000, e=2, n=16384K CUDAPm1 v0.20)
M375000013 Stage 2 found no factor (P-1, B1=3085000, B2=69412500, e=2, n=21168K CUDAPm1 v0.20)
M383000063 Stage 2 found no factor (P-1, B1=2930000, B2=63727500, e=2, n=21504K CUDAPm1 v0.20)
M425000083 Stage 2 found no factor (P-1, B1=2840000, B2=62000000, e=2, n=24192K CUDAPm1 v0.20)
More info on what that GPU could run is shown in one of the attachments at https://www.mersenneforum.org/showpo...73&postcount=9

[James Heinrich]

My above sample results are actually two sequential results I got a few days ago, and used about 38GB to get E=12 on 89M exponents.

[ixfd64]

I have two suggestions for the next version:

1. If the user has a proxy configured, then Prime95 should fall back to a standard connection if the proxy isn't available.

Use case: I have a work laptop that I regularly take home. All network connections must go through the proxy when connected to the corporate network. Because the proxy is publicly inaccessible, I have to change the settings in Prime95 every time I take the laptop home.

Alternative idea: allow users to configure more proxies to fall back to.

2. I believe this is a known issue, but Prime95 will always try to use more than one core per worker window even when "CPU cores to use" is set to 1 per worker. I have to set CoresPerTest=1 in local.txt to solve this problem.

[Prime95]

Originally Posted by ixfd64

I have two suggestions for the next version:

1. If the user has a proxy configured, then Prime95 should fall back to a standard connection if the proxy isn't available.

Use case: I have a work laptop that I regularly take home. All network connections must go through the proxy when connected to the corporate network. Because the proxy is publicly inaccessible, I have to change the settings in Prime95 every time I take the laptop home.

Alternative idea: allow users to configure more proxies to fall back to.

Done.

2. I believe this is a known issue, but Prime95 will always try to use more than one core per worker window even when "CPU cores to use" is set to 1. I have to set CoresPerTest=1 in local.txt to solve this problem.

Please elaborate. Windows/Linux? Brand new install? Your CPU? Steps to reproduce bug?

[ixfd64]

Thanks for the response.

The second issue occurs on Windows at least — I haven't tested any other platforms. To reproduce:

1. Set the number of workers = number of physical cores
2. Set the worker number to "All workers"
3. Set the number of CPU cores to 1 per worker

When the user tries to save the settings, there will be a message saying Prime95 is using more cores than available. The number of cores per thread will revert to a value > 1 afterwards.

[Prime95]

Originally Posted by ixfd64

The second issue occurs on Windows at least — I haven't tested any other platforms. To reproduce:

1. Set the number of workers = number of physical cores
2. Set the worker number to "All workers"
3. Set the number of CPU cores to 1 per worker

When the user tries to save the settings, there will be a message saying Prime95 is using more cores than available. The number of cores per thread will revert to a value > 1 afterwards.

I'm doing something wrong. What were the setting prior to entering Test/Worker Windows?