[QUOTE=kladner;489168]Oops. Those [U]are[/U] collected in a sub-thread of Hardware:
[URL]http://www.mersenneforum.org/forumdisplay.php?f=134[/URL][/QUOTE]

I need to update those. They still tell people to use c4 instances, but now there's c5.

Also I never did get around to writing the one for Google Cloud. It was less cost-effective than AWS, since Google's Skylake chips are slower than AWS's, and spot prices for c5 were low for a long time, but now they're higher so maybe it needs another look.

gpuowl performance testing
 

Currently testing version 2.2 or github master branch. On certain exponents it is being slow as hell, like 32-34 ms/it, but the exponents in question all of the form 85226xxx.

[QUOTE=SELROC;489419]Currently testing version 2.2 or github master branch. On certain exponents it is being slow as hell, like 32-34 ms/it, but the exponents in question all of the form 85226xxx.[/QUOTE]
It's still work-in-progress. What GPU?
It prints the FFT size at startup. Likely it uses 8M FFT (2048x2048x2) which is better for 150M exponents instead of 85M. Anyway 30ms/it is very slow. As a comparison, on 390x which is pretty old I get about 11ms/it for a 16M FFT (300M exponent). So I wonder what card you have. Or something is wrong.

[QUOTE=preda;489422]It's still work-in-progress. What GPU?
It prints the FFT size at startup. Likely it uses 8M FFT (2048x2048x2) which is better for 150M exponents instead of 85M. Anyway 30ms/it is very slow. As a comparison, on 390x which is pretty old I get about 11ms/it for a 16M FFT (300M exponent). So I wonder what card you have. Or something is wrong.[/QUOTE]

The cards are Asus Radeon RX 580 (Ellesmere). On other exponents (851xxxxx) I have 6-7 ms/it which is good. On the exponents 85226xxx I have 32-34 ms/it. But I have to investigate what happens if I rotate exponents on cards.

[QUOTE=SELROC;489424]The cards are Asus Radeon RX 580 (Ellesmere). On other exponents (851xxxxx) I have 6-7 ms/it which is good. On the exponents 85226xxx I have 32-34 ms/it. But I have to investigate what happens if I rotate exponents on cards.[/QUOTE]
Probably it's not because of different exponents, but because of different program version, which uses larger FFT and long-carry. You can look at this information printed on start-up. In that case, it'd probably be better to remain on the old version, at least until the on-going exponents are completed.

[QUOTE=preda;489472]Probably it's not because of different exponents, but because of different program version, which uses larger FFT and long-carry. You can look at this information printed on start-up. In that case, it'd probably be better to remain on the old version, at least until the on-going exponents are completed.[/QUOTE]

The problem here is not different program versions but an experimental system with experimental kernel and firmware, it is absolutely unstable and upgrades may break the system. At this time it was necessary (about 6 months ago) because the Drivers required some features of the latest kernel. So, I am going to refrain from upgrading this system further, gpuowl included, until those exponents are fully computed.

I am working to build a stable production system, this requires that the drivers work with the stable version of the kernel, the firmware also needs work. We can only wait until a new stable Debian release that works with latest drivers is out.

version and feature announcements
 

I've put together a set of links to gpuowl version announcements.
It's at [URL="http://www.mersenneforum.org/showpost.php?p=489083&postcount=7"]http://www.mersenneforum.org/showpos...83&postcount=7[/URL]

Please provide urls of any I missed.

Silly question. I know gpuOwl switched from LL to PRP tests some time ago. Can the user still use it to run LL tests, or has that functionality been removed?

[QUOTE=ixfd64;489736]Silly question. I know gpuOwl switched from LL to PRP tests some time ago. Can the user still use it to run LL tests, or has that functionality been removed?[/QUOTE]
Answer is at the link I posted just above your question, or very near it in the same thread there.

It looks like the LL code has been removed:

[QUOTE]gpuOwl does not implement LL (Lucas-Lehmer primality test) anymore. Instead, it does the "PRP-3" probable prime test, which is:[/QUOTE]

I'd have preferred to retain the old functionality and maybe print a warning saying that LL tests are no longer supported. But that's just me.

When time cames, I'll think about adding LL back. In the meantime I have CUDA-port at the top of the stack.