mersenneforum.org  

Go Back   mersenneforum.org > Search Forums

Showing results 1 to 25 of 1000
Search took 0.18 seconds.
Search: Posts Made By: preda
Forum: GpuOwl 2020-10-31, 06:27
Replies: 43
Views: 985
Posted By preda
I don't know, probably not trivial due to the...

I don't know, probably not trivial due to the need to do the whole dance: write to a new temporary file, rename worktodo to .bak, rename to temp to worktodo. Plus all the traps along the way due to...
Forum: GpuOwl 2020-10-31, 04:32
Replies: 43
Views: 985
Posted By preda
Well I did explain the reason for that...

Well I did explain the reason for that requirement: new work lines are added by *appending* them to the worktodo.txt file. If the worktodo.txt is not terminated with a newline, the append produces...
Forum: Software 2020-10-29, 05:16
Replies: 35
Views: 1,993
Posted By preda
Yes I think we're the state of the art in what...

Yes I think we're the state of the art in what concerns carry propagation :)
Forum: Hardware 2020-10-29, 01:43
Replies: 8
Views: 299
Posted By preda
Yep, unfortunately: the hardware is done, time to...

Yep, unfortunately: the hardware is done, time to start work on the software side..
(and by that I don't mean *our* software side, I mean AMD's software side)
Forum: Software 2020-10-29, 01:36
Replies: 35
Views: 1,993
Posted By preda
Thank you. While the thesys contains a wide...

Thank you. While the thesys contains a wide overview, it's does not go too deep into multi-precision FP (at least based on my cursory inspection).
Forum: Software 2020-10-28, 22:06
Replies: 35
Views: 1,993
Posted By preda
It seems the wavefront (100M+) could be handled...

It seems the wavefront (100M+) could be handled with 2xSP at 6.5M FFT or *maybe* 6M FFT pushing it a bit. Sounds efficient enough to be worth a try.
Forum: GpuOwl 2020-10-28, 19:50
Replies: 152
Views: 4,960
Posted By preda
No, 2xSP is an experiment, can't be used for...

No, 2xSP is an experiment, can't be used for anything yet. Still a long way to go. (I was just measuring the precission that can be achieved *if* it was implemented)
Forum: GpuOwl 2020-10-28, 19:49
Replies: 152
Views: 4,960
Posted By preda
P2 needs at least 24 buffers. As the exponent...

P2 needs at least 24 buffers. As the exponent grows, the buffer size grows, and this minimum required may not be met dependning on the -maxAlloc allowed. I do not plan to fix this ATM, let's simply...
Forum: Software 2020-10-28, 12:18
Replies: 35
Views: 1,993
Posted By preda
After a few accuracy fixes, the 2xSP experiment...

After a few accuracy fixes, the 2xSP experiment can do 17bits/word, which is exactly where I was expecting it.

OTOH the multiprecision ADD uses 20 SP ADDs!!
Given that in the FFT we do lots of...
Forum: Software 2020-10-28, 10:49
Replies: 35
Views: 1,993
Posted By preda
2xSP initial experiments

The 1xSP 2M convolution code I mentioned previously is here:

https://github.com/preda/gpuowl/tree/7547cff0540d8932b5a33756b0e812b32bd4bd0c/SP...
Forum: GpuOwl 2020-10-28, 07:52
Replies: 2,556
Views: 159,650
Posted By preda
Well in fact not really. The behavior of the -log...

Well in fact not really. The behavior of the -log right now is to control how often to do the GEC check + save (by default 200k). That 10k display does not do much anything other than a bit of...
Forum: Software 2020-10-28, 00:42
Replies: 35
Views: 1,993
Posted By preda
I did some experiments with single SP FFT (the...

I did some experiments with single SP FFT (the simplest case), to establish the baseline.

A 2M convolution (1024x1024 SP pairs), not-weighted, done in the best possible accuracy (perfect SP...
Forum: GpuOwl 2020-10-25, 07:04
Replies: 152
Views: 4,960
Posted By preda
Proof validation

For those who can, it may be a good idea to use -proof 9, which enables validation of the proof. The cost of the validation is 0.2% which is small enough (on the order of 2-3 minutes on R7), but it...
Forum: GpuOwl 2020-10-25, 06:34
Replies: 152
Views: 4,960
Posted By preda
Thanks, I'll need to look into why STATS fails...

Thanks, I'll need to look into why STATS fails for large exponents.
Forum: GpuOwl 2020-10-25, 06:25
Replies: 43
Views: 985
Posted By preda
There is a simple pragmatic reason for requiring...

There is a simple pragmatic reason for requiring the line-ending, and that is composability of files.

In file A.txt we have:
AAA
BBB
In file B.txt we have:
CCC
DDD

Then one would naturally...
Forum: GpuOwl 2020-10-24, 11:54
Replies: 152
Views: 4,960
Posted By preda
Yep. I forgot that the driver on Windows does not...

Yep. I forgot that the driver on Windows does not support ASM.
Forum: GpuOwl 2020-10-24, 11:37
Replies: 152
Views: 4,960
Posted By preda
Yes I agree. I wanted to say that the fix to the...

Yes I agree. I wanted to say that the fix to the loop you reported was high-priority because otherwise it represented such a waste.
Forum: GpuOwl 2020-10-24, 11:32
Replies: 152
Views: 4,960
Posted By preda
OK I understand, I'll consider implementing this....

OK I understand, I'll consider implementing this. I still consider "stop the GPU" a safe bail-out, while "keep running 100% doing nothing" a waste.
Forum: GpuOwl 2020-10-24, 11:29
Replies: 152
Views: 4,960
Posted By preda
NO_ASM on R7 -- some like it slow?

NO_ASM on R7 -- some like it slow?
Forum: GpuOwl 2020-10-24, 11:24
Replies: 43
Views: 985
Posted By preda
Well my point was that I'm not rewriting the...

Well my point was that I'm not rewriting the file. As you can imagine, re-writing the file has its own drawbacks, involving files being renamed and deleted and whatnot, so if all I want to do is to...
Forum: GpuOwl 2020-10-24, 11:13
Replies: 43
Views: 985
Posted By preda
It's not clear what you mean by "will stop". ...

It's not clear what you mean by "will stop".

If it throws an exception telling you to add a newline, that's fine.

OTOH what you seem to have reported earlier was an exponent that was ran twice....
Forum: GpuOwl 2020-10-24, 10:08
Replies: 152
Views: 4,960
Posted By preda
Ken I merged an attempted fix, maybe you could...

Ken I merged an attempted fix, maybe you could try it on the looping exponent and check the behavior.
Forum: GpuOwl 2020-10-24, 09:42
Replies: 152
Views: 4,960
Posted By preda
Were you using -log 100000 by any chance in your...

Were you using -log 100000 by any chance in your config?
Forum: GpuOwl 2020-10-24, 08:56
Replies: 152
Views: 4,960
Posted By preda
Yes pretty serious. I'll try to address this (I...

Yes pretty serious. I'll try to address this (I didn't hit this myself yet, thus didn't realize the regression).

You should be able to continue the exponent, just bump up the FFT-size or some...
Forum: GpuOwl 2020-10-24, 05:56
Replies: 43
Views: 985
Posted By preda
If I only add in memory, it is not possible to...

If I only add in memory, it is not possible to append new lines to the file by simple concatenation.

If the file ends with "AAA" without newline, and I append "BBB\n", now the file contains...
Showing results 1 to 25 of 1000

 
All times are UTC. The time now is 20:49.

Sat Oct 31 20:49:08 UTC 2020 up 51 days, 18 hrs, 2 users, load averages: 1.85, 1.88, 1.99

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.