The prime-crunching on dedicated hardware FAQ - Page 13

Robert Holmes · 2009-09-25, 15:36

You might be interested in this:

http://www.mpi-inf.mpg.de/~emeliyan/poly_mul.pdf

Not sure how this compares to floating point implementations, though.

zarabatana · 2009-09-27, 19:17

Adding information about GPU Double Precision floating point performance:

Tesla C1060: 78 GFLOPS - $ 1.299,99 (tigerdirect)
Tesla S1070: 311 to 345 GFlops (S1070 use 4 C1060) $ no ideia
AMD FireStream 9270: 240 GFLOPS -$ 1.204,99 (newegg)

Thank you for your time.

joblack · 2009-09-30, 13:12

The

Radeon HD 5870

has dobule precision and is fast as hell. Now it should be possible to optimize Prime95 on it ;) ...

jasonp · 2009-09-30, 13:49

Quote:

Originally Posted by zarabatana

Tesla S1070: 311 to 345 GFlops (S1070 use 4 C1060) $ no ideia

Not for casual use

fivemack · 2009-09-30, 14:44

Quote:

Originally Posted by joblack

The

Radeon HD 5870

has double precision and is fast as hell. Now it should be possible to optimize Prime95 on it ;) ...

Only when AMD get round to releasing the specifications for its assembly language, or at very least a compiler capable of compiling for it. At the moment you have to code in VLIW assembly language to get any kind of double-precision performance out of the RV770.

Robert Holmes · 2009-09-30, 18:02

Rumor says the upcoming GT300 cards will have stronger DP support:

http://www.brightsideofnews.com/news...6gb-gddr5.aspx

At 2 Ghz, this means 1 TFlop/s DP (256 * 2 * (2*10^9)).

In other news, NVIDIA seems to have released its OpenCL SDK to the public:

http://developer.nvidia.com/object/get-opencl.html

Robert Holmes · 2009-09-30, 20:36

Official GT300 blurb:

http://www.nvidia.com/content/PDF/fe...Whitepaper.pdf

Quote:

Originally Posted by Robert Holmes

Rumor says the upcoming GT300 cards will have stronger DP support:

http://www.brightsideofnews.com/news...6gb-gddr5.aspx

At 2 Ghz, this means 1 TFlop/s DP (256 * 2 * (2*10^9)).

In other news, NVIDIA seems to have released its OpenCL SDK to the public:

http://developer.nvidia.com/object/get-opencl.html

fivemack · 2009-09-30, 22:02

I think I want one of those.

ldesnogu · 2009-09-30, 22:44

Some more info about GT300 here: http://www.realworldtech.com/page.cf...3009110932&p=1

frmky · 2009-11-12, 05:23

Quote:

Originally Posted by jasonp

Q: Can I use my graphics card to increase my contribution to projects like GIMPS, LLR, Riesel, etc?
Q: My Playstation 3 has wondrous crunch power. Can I use it to increase my contribution to projects like GIMPS, LLR, Riesel, etc?

A: No.

This has very recently changed to yes, at least for GIMPS on nVidia cards. Thanks to the inclusion of double-precision FFTs in the CUDA 2.3 cufft library (an FFTW-like library for CUDA), msft has successfully ported MacLucasFFTW to CUDA. It works with nVidia GTX 260 and higher cards; earlier cards do not have the required double precision units. While not blazingly fast, the performance is on par with Prime95. The program is distributed as source only and requires manual reservation of exponents.

jasonp · 2009-11-15, 23:20

Part 2 of the FAQ is now posted here; please direct followups to that thread instead.

(The original FAQ exceeded the 10k post size limit, which was imposed after it was written, and editing it directly is not allowed unless it is split into multiple parts)

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
New PC dedicated to Mersenne Prime Search	Taiy	Hardware	12	2018-01-02 15:54
The prime-crunching on dedicated hardware FAQ (II)	jasonp	Hardware	46	2016-07-18 16:41
How would you design a CPU/GPU for prime number crunching?	emily	Hardware	4	2012-02-20 18:46
DSP hardware for number crunching?	ixfd64	Hardware	15	2011-08-09 01:11
Optimal Hardware for Dedicated Crunching Computer	Angular	Hardware	5	2004-01-16 12:37

2009-09-25, 15:36	#133
Robert Holmes Oct 2007 2×53 Posts	You might be interested in this: http://www.mpi-inf.mpg.de/~emeliyan/poly_mul.pdf Not sure how this compares to floating point implementations, though.

2009-09-27, 19:17	#134
zarabatana Sep 2009 2³·3 Posts	Adding information about GPU Double Precision floating point performance: Tesla C1060: 78 GFLOPS - $ 1.299,99 (tigerdirect) Tesla S1070: 311 to 345 GFlops (S1070 use 4 C1060) $ no ideia AMD FireStream 9270: 240 GFLOPS -$ 1.204,99 (newegg) Thank you for your time.

2009-09-30, 13:12	#135
joblack Oct 2008 n00bville 2³·7·13 Posts	The Radeon HD 5870 has dobule precision and is fast as hell. Now it should be possible to optimize Prime95 on it ;) ...

2009-09-30, 18:02	#138
Robert Holmes Oct 2007 2·53 Posts	Rumor says the upcoming GT300 cards will have stronger DP support: http://www.brightsideofnews.com/news...6gb-gddr5.aspx At 2 Ghz, this means 1 TFlop/s DP (256 * 2 * (2*10^9)). In other news, NVIDIA seems to have released its OpenCL SDK to the public: http://developer.nvidia.com/object/get-opencl.html

2009-09-30, 22:02	#140
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 7²×131 Posts	I think I want one of those.

2009-09-30, 22:44	#141
ldesnogu Jan 2008 France 2×5²×11 Posts	Some more info about GT300 here: http://www.realworldtech.com/page.cf...3009110932&p=1

2009-11-15, 23:20	#143
jasonp Tribal Bullet Oct 2004 3,541 Posts	Part 2 of the FAQ is now posted here; please direct followups to that thread instead. (The original FAQ exceeded the 10k post size limit, which was imposed after it was written, and editing it directly is not allowed unless it is split into multiple parts)