I just installed another program - "PRST" on SR5.

PRST is a new prime testing program developed by Pavel Atnashev. Its main advantage is a new certification method developed by Darren Li (first introduced in Genefer 2.0) which work better for tests where b <> 2. In short, new method eliminates 28-35% speed penalty appeared in LLR2 with a cost of double certification time (3% for SR5). Combined, this gives a very good speedup.

New application is available for Win32, Win64, Linux64 and Mac64 platforms.

Although new program received some testing on dev server and successfully competed few test tasks, its behavior differs from LLR2 signiticantly. Please report here If something will not work.

Note that server still have a queue of old tasks which, highly likely, will be sent first until new tasks became available. So it may take some time until new tasks will appear on your computers. Similarly, old resent task may appear for a while.

SR5 still receives increased credit comparing to other projects (we added 35% boost after switch to LLR2 to compensate LLR2 slowdowns) but it may be changed in the future.

My first (and only) 9.50 task was a proof task that errored out without doing any computation on a stable Xeon v4 system. It passed on a Phenom II. Fluke or ?

https://www.primegrid.com/result.php?resultid=1544680374
____________
Eating more cheese on Thursdays.

Haven't had any more errors on my systems, yet. (Keeping an eye on it).

The profile page task completion list however isn't showing the number of threads used.
____________
Eating more cheese on Thursdays.

My first (and only) 9.50 task was a proof task that errored out without doing any computation on a stable Xeon v4 system. It passed on a Phenom II. Fluke or ?

https://www.primegrid.com/result.php?resultid=1544680374

My first (and only) 9.50 task was a proof task that errored out without doing any computation on a stable Xeon v4 system. It passed on a Phenom II. Fluke or ?

https://www.primegrid.com/result.php?resultid=1544680374

This bug was fixed few days ago, as soon as first wrong result was returned. Old DC LLR task was erroneously assigned to PRST and, of course, failed, because command line and everything else is absolutely different.

Of course those who already received bad tasks will see them failed. It's safe to ignore this "Too many exists" error.

Glad it is being introduced on PrimeGrid. Can't wait for it to move over to GCW.
____________

OK, performed some tests on my AMD5950X and Win11.
1, 2, 4 and 8 threads, CPU afifinty, no other things running...you know the drill.

It seems that PRST is considerably (~2x) slower with small candidates/primes like 157672*5^333603-1 (80K FFT).
No big deal, we are way past this range.
btw, PRST isn't doing Lucas sequence like LLR2 to verify it's a primes so there is not time penalty in that rare case.

Then I picked up on recent candidates like 60394*5^4384584+1 (1MB FFT)
Those are similar in speed with LLR2 doing a bit better.

I'm reading speed from time per bits since task manager in CPU load/times are way off on Windows 11.
It may be Win11 specific, it may be AMD specific so I may do extra tests on different CPUs/OSes.
____________
My stats

Now AMD 7800X3D DDR5 2x32GB, Win 2019.
It's intentionally without affinity since it acts funny otherwise.

PRST looks marginally faster there, with pretty good thread number scaling.
____________
My stats

Now AMD 7800X3D DDR5 2x32GB, Win 2019.
It's intentionally without affinity since it acts funny otherwise.

PRST looks marginally faster there, with pretty good thread number scaling.

You MUST use -oGerbicz=1 for LLR2. Did you? It's not enabled by default when b=5

I'm getting 20-30% increased daily output based on task time averages on Broadwell-EP and Skylake-SP hardware. Good stuff! Boosted the daily credit for my 3 active boxes right over 500k.
____________
Eating more cheese on Thursdays.

Will the 5GB swap still be needed for efficient GCW testing once that testing is moved to the PRST application?

Thanks for explanation, great answer.
____________
My stats

I'm getting 20-30% increased daily output based on task time averages on Broadwell-EP and Skylake-SP hardware. Good stuff! Boosted the daily credit for my 3 active boxes right over 500k.

They tell us now you found a PRST prime: 177742*5^4386703 - 1

Congratulations! Enjoy you K badge once it is validated.

Also congratulations to PRST and its creator Pavel Atnashev.

/JeppeSN

BTW, I see that the T5K entry for this prime is an "L" proof code, indicating the final proof was done with LLR, and has PRST credited in a secondary role. Since PRST is capable of fully proving primality by itself (although I guess it would have to be as a separate run without -fermat, either manually or initiated by the BOINC wrapper after seeing the Fermat PRP result), is this a mistake in the prime submission, or was the PRP in fact proven prime with LLR? (Or LLR2, although - as a separate known issue - the T5K site doesn't presently support selecting LLR2 as a final proof program.)

For CPU cache requirements is it still FFT*8 like with other LLR/LLR2 tasks or is it FFT*16 like with Primorial/Factorial?
____________
My Primes
Badge Score: 4*2 + 6*2 + 7*1 + 8*10 + 9*2 + 11*2 + 12*2 = 171