Prime95 30.8 (beta, big P-1 changes, see post #551)

[axn]

What will be the trigger to go from pre-beta moniker to beta (or even RC)?

[Zhangrc]

Maybe ... 30.8 is pre-beta, and 30.9 is beta?

[kruoli]

Originally Posted by Zhangrc

I'm pretty sure that if I closed hyperthreading, the problem would resolve itself. However I didn't see such option in my BIOS.

It would be a lot easier to use the Affinity= option as explained in undoc.txt where you can also fix the work to one of two hyperthreads.

[ET_]

Originally Posted by Prime95

30.8 build 9

Only new feature is P-1 now works for non-Mersennes.

Is it safe for small Fermats? I have someone at FermatSearch who is asking me to work on F12 and F13...

Luigi

[Prime95]

Originally Posted by ET_

Is it safe for small Fermats? I have someone at FermatSearch who is asking me to work on F12 and F13...

It is safe. I'll bet GMP-ECM has P-1'ed those numbers extensively.

[Prime95]

Originally Posted by kruoli

It would be a lot easier to use the Affinity= option as explained in undoc.txt where you can also fix the work to one of two hyperthreads.

Try it. The code might still work.

In an ideal world, one would not have to mess with affinity settings at all. Prime95 would just do the right thing.

[Prime95]

Originally Posted by axn

What will be the trigger to go from pre-beta moniker to beta (or even RC)?

The big question is "Do I add faster P+1/ECM to 30.8 or postpone that to 30.9"?

Even if I postpone, build 9 is not yet feature complete. So, RC is quite a ways off.

[axn]

Originally Posted by Prime95

The big question is "Do I add faster P+1/ECM to 30.8 or postpone that to 30.9"?

Even if I postpone, build 9 is not yet feature complete. So, RC is quite a ways off.

My view is that, once we push this as an official build and goes to front page, it will be used for official primenet work. The two work types that I can think of that are relevant would be leading edge P-1 and small ECM. Therefore, ECM stage 2 is probably needed.

At this point, primenet-based P-1 wouldn't really be able to take advantage of 30.8 capabilities, as the number of machines with sufficient memory would not be large enough to make a difference.

Curious: what is pending for 30.8 to make it "feature complete"?

[kruoli]

Originally Posted by Prime95

The code might still work.

It works for me. I am using it to assign Prime95 to non-default cores (starting the threads on the second NUMA node of a Threadripper CPU e.g.).

I am also using it on another machine where I have eight Prime95 instances with ECM running. I get much better stage 2 times that way. Somehow there might be a synchronisation issue.

[Zhangrc]

Originally Posted by Prime95

The big question is "Do I add faster P+1/ECM to 30.8 or postpone that to 30.9"?

Even if I postpone, build 9 is not yet feature complete. So, RC is quite a ways off.

Maybe 31.0 will be ready for those

[ATH]

Originally Posted by Prime95

The big question is "Do I add faster P+1/ECM to 30.8 or postpone that to 30.9"?

I do not care what the version number is called, whatever option releases it the fastest