mersenneforum.org  

Go Back   mersenneforum.org > Search Forums

Showing results 1 to 25 of 1000
Search took 0.12 seconds.
Search: Posts Made By: preda
Forum: mersenne.ca 2021-03-02, 19:04
Replies: 540
Sticky: mersenne.ca
Views: 58,378
Posted By preda
The most recent GpuOwl's P-1 calculator is here: ...

The most recent GpuOwl's P-1 calculator is here:

https://github.com/preda/gpuowl/blob/master/pm1/pm1.cpp

I dropped the equivalent python version because it was becoming laborious to maintain...
Forum: Software 2021-02-20, 08:48
Replies: 39
Views: 5,434
Posted By preda
3xSP sum()

Unfortunately the sum() I have up to now is a beast: 54 ADDs.


This seems a rather very expensive sum()..

To see some corner-cases that sum() must handle, here is one example: given "x", we'd...
Forum: GPU Computing 2021-02-20, 08:42
Replies: 7
Views: 508
Posted By preda
How user-unfriendly is that! having the guts to...

How user-unfriendly is that! having the guts to say it out loud: we (Nvidia) get to decide what you use your GPU for. I must ask, what about watching porn on Nvidia GPUs, is that allowed by the...
Forum: Software 2021-02-12, 21:02
Replies: 39
Views: 5,434
Posted By preda
Figure 10 seems to indicate: c0,e0 =...

Figure 10 seems to indicate:

c0,e0 = twoSum(a0, b0)

d1,e11 = twoSum(a1, b1)
c1,e12 = twoSum(d1, e0)

c2 = a2 + b2 + e11 + e12

which looks pretty good (i.e. simpler than I was expecting)
Forum: Software 2021-02-12, 19:45
Replies: 39
Views: 5,434
Posted By preda
Funnily, I'm struggling even with implementing a...

Funnily, I'm struggling even with implementing a 3xSP ADD :).

Here is a good paper with the solution for quad-double:
https://web.mit.edu/tabbott/Public/quaddouble-debian/qd-2.3.4-old/docs/qd.pdf...
Forum: Software 2021-02-12, 08:42
Replies: 39
Views: 5,434
Posted By preda
SP plan

I've been thinking some more about a practical SP FFT implementation on GPUs, and here are some problems/ideas:

1. FFT twiddles, i.e. the trigonometric constants (sin+cos) used in the FFT.
...
Forum: GpuOwl 2021-02-06, 20:04
Replies: 48
Views: 5,568
Posted By preda
For what it's worth, on R7 I'm personally running...

For what it's worth, on R7 I'm personally running with B1=9M, B2=180M for 102M-103M exponents (factored to 76bits).
Forum: GpuOwl 2021-02-04, 06:16
Replies: 48
Views: 5,568
Posted By preda
The program also allows to specify a fixed B1 or...

The program also allows to specify a fixed B1 or B2, and in that situation displays options for the other bound. Examples below with fixed B1=1M, or fixed B2=50M (note, the values below are good with...
Forum: GpuOwl 2021-02-03, 21:40
Replies: 48
Views: 5,568
Posted By preda
GpuOwl updated P-1 calculator

Hi, recently I revisited the P-1 calculator that's included with GpuOwl's source code https://github.com/preda/gpuowl/blob/master/pm1/pm1.cpp

The calculator is a small stanalone C++ program; to...
Forum: GPU Computing 2021-01-17, 16:42
Replies: 21
Views: 1,255
Posted By preda
I would recommend to get a 850W or at least 750W...

I would recommend to get a 850W or at least 750W PSU, Gold 80+, and modular or semi-modular. Maybe read some reviews of the model before buying. The reason is: you have some power headroom (to 850W),...
Forum: Hardware 2021-01-05, 10:12
Replies: 128
Views: 10,338
Posted By preda
The cache (L1/L2/L3) is used transparently for...

The cache (L1/L2/L3) is used transparently for the *global* memory operations. It is managed automatically by the cache control (probably a variant of LRU), not explicitly by the software. So yes,...
Forum: GpuOwl 2020-12-13, 04:31
Replies: 197
Views: 15,328
Posted By preda
I'm personally not using 6.x myself, thus I don't...

I'm personally not using 6.x myself, thus I don't have a lot of motivation to improve it. From my POV, 7.x is now better in a couple of ways than 6.x, and I prefer to focus my (limited) resources on...
Forum: GpuOwl 2020-12-06, 22:58
Replies: 2,691
Views: 222,420
Posted By preda
GpuOwl uses very very little PCIe bandwidth. I...

GpuOwl uses very very little PCIe bandwidth. I regularly run it over PCIe x1 Gen1 without significant slowdown.
Forum: Hardware 2020-12-06, 22:54
Replies: 12
Views: 1,225
Posted By preda
Hardware donation and possible meet-ups

I may have some extra hardware located in Sydney (Australia), I'm considering donating it based on GIMPS participation. Is somebody [else] from the forum living in Sydney?

Now that I think of it,...
Forum: GpuOwl 2020-12-01, 23:42
Replies: 2,691
Views: 222,420
Posted By preda
Nice, thank you! I hope others will find the...

Nice, thank you! I hope others will find the package useful.
Forum: GpuOwl 2020-11-24, 21:38
Replies: 197
Views: 15,328
Posted By preda
"has slowed down" -- relative to what? did you...

"has slowed down" -- relative to what? did you compare two versions, to see what is the difference between them? Let's call the two versions you compare "before" and "after", or "good" and "bad"....
Forum: GpuOwl 2020-11-23, 22:23
Replies: 197
Views: 15,328
Posted By preda
No that is not possible with the current...

No that is not possible with the current savefiles, they are most probably different between GpuOwl and mprime.

OTOH mprime may be offering the merged PRP+P1 at some point in the future.
Forum: GpuOwl 2020-11-23, 09:53
Replies: 197
Views: 15,328
Posted By preda
Interesting. I haven't tried FFT 6M myself yet...

Interesting. I haven't tried FFT 6M myself yet (I'm still on 5.5M), I probably should.
- time helps a bit with timing the kernels. Sometimes running with -time old/new and comparing may provide a...
Forum: GpuOwl 2020-11-23, 02:58
Replies: 197
Views: 15,328
Posted By preda
Is your GPU actually a 4GB RAM card?

Is your GPU actually a 4GB RAM card?
Forum: GpuOwl 2020-11-13, 22:57
Replies: 2,691
Views: 222,420
Posted By preda
Not ATM. Based on the information available, it...

Not ATM. Based on the information available, it is more efficient to run with a higher B2 than to use BS (for the same amount of compute in P2).
Forum: GpuOwl 2020-11-13, 20:10
Replies: 2,691
Views: 222,420
Posted By preda
It is simply misreporting from GpuOwl: when it...

It is simply misreporting from GpuOwl: when it finds a P-1 factor, the part that writes the result doesn't know anymore whether it was found in stage1 or in stage2. It could write the result with...
Forum: GpuOwl 2020-11-12, 21:54
Replies: 197
Views: 15,328
Posted By preda
I think ROCm 3.3 is good. Also any after 3.5...

I think ROCm 3.3 is good. Also any after 3.5 should work fine, but maybe slower than 3.3. You can try, now it's easier to install multiple versions of ROCm OpenCL in parallel (at the same time) and...
Forum: Math 2020-11-11, 00:06
Replies: 14
Views: 933
Posted By preda
In pari/gp: 1 / mathilbert(n)

In pari/gp:

1 / mathilbert(n)
Forum: Math 2020-11-10, 23:33
Replies: 14
Views: 933
Posted By preda
Thanks! I thought that must be some well-known...

Thanks! I thought that must be some well-known matrix, but I didn't know its name :)
Forum: GpuOwl 2020-11-10, 22:48
Replies: 2,691
Views: 222,420
Posted By preda
Nice detective work! (yes, that sounds like a...

Nice detective work! (yes, that sounds like a reasonable explanation)
Showing results 1 to 25 of 1000

 
All times are UTC. The time now is 16:28.

Mon Mar 8 16:28:05 UTC 2021 up 95 days, 12:39, 1 user, load averages: 1.46, 1.81, 1.75

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, 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.