50 additional gigaflops in my PC with this card?

guyloginin
guyloginin
Joined: 3 Mar 06
Posts: 3
Credit: 53860
RAC: 0
Topic 191036

Greetings.

I stumbled upon a product called the ClearSpeed Accelerator card. The card uses dual CSX600 processors and interfaces using an PCIE bus. More than one of these cards can be installed in a single system.

Code porting is apparently not required as the card " accelerates standard math libraries".

Would this product be useful with the Einstein@Home BOINC client?

Can anyone get one donated to phys.uwm for evaluation?

The company has not shared cost info with me but I imagine the cost is beyond practical for a general home PC user.

I didn't post the company's url here but you should be able to guess what it is.

Cheers.

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4527270
RAC: 0

50 additional gigaflops in my PC with this card?

Quote:
I stumbled upon a product called the ClearSpeed Accelerator card. The card uses dual CSX600 processors and interfaces using an PCIE bus. More than one of these cards can be installed in a single system.

I read more about this card about a year ago. It wasn't too cheap.
Like as a good car... or two...
And the usability of this car is limited.
It doesn't know trigonometrcial functions.

edit:
AMD Duron 1800 -> 1.8GFlops (~40USD)
ClearSpeed CSX600 -> 50 GFlops (~10.000USD)
(just one car)

archae86
archae86
Joined: 6 Dec 05
Posts: 3145
Credit: 7023754931
RAC: 1805738

Old story. I actually

Old story.

I actually visited a house in Silicon Valley about 1985 with a group of potential angel investors to look at an attempted startup trying to make and sell an 8087-replacement board for PC's. Claim was several times faster than 8087 for many core function (maybe 10x), 8087-equivalent for the neglected math functions.

I told my fellow angels not to invest. Same problem as here. Speeding up a portion of the computation for lots of money leaves you with the rest of the computation unspeeded up, and lots of money gone. Always a bad deal unless the portion speeded up is overwhelmingly high, which was not true in that case, and seems unlikely in this save for special cases.

The startup disappeared without a trace-so perhaps I was right.

Buy it if it fits your case. Avoid like the plague for general use.

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4527270
RAC: 0

RE: It doesn't know

Message 27428 in response to message 27426

Quote:
It doesn't know trigonometrcial functions.

CSX600 doesn't know them, but it supports (emulation).

The ClearSpeed Advance board supports FFTW library.
(Would be good to SETI users.)

tullio
tullio
Joined: 22 Jan 05
Posts: 2118
Credit: 61407735
RAC: 0

RE: The ClearSpeed Advance

Message 27429 in response to message 27428

Quote:
The ClearSpeed Advance board supports FFTW library.
(Would be good to SETI users.)


This is my SETI app running on Linux, 400 MHz Pentium II (Deschutes):
(Optimized SETI client V4.07.3a for i686 with FFTW3 by Ned Slider)
Tullio

guyloginin
guyloginin
Joined: 3 Mar 06
Posts: 3
Credit: 53860
RAC: 0

The ClearSpeed folks Did want

The ClearSpeed folks Did want to know if the Einstein BOINC application was DGEMM and FFT intensive. All I know about that though is on the Wikipedia encyclopedia.

[boinc.at] Walter
[boinc.at] Walter
Joined: 6 Jun 05
Posts: 2
Credit: 321953
RAC: 0

For home-users this

For home-users this addon-card would not be useful. Proprietary code has to be added to the applications via the compiler that comes with the csx600. It is just a mathematical coprocessor, nothing else.

Director of Architecture at ClearSpeed is Simon McIntosh-Smith - a good name behind this product, hm?

Wouldnt it be interesting if AMD or Intel could implement this technology in their processors? Intel uses already AMD's 64 bit extensions, so why not use MCIntosh-technology?

Reuben Gathright
Reuben Gathright
Joined: 15 Feb 06
Posts: 23
Credit: 6851215
RAC: 0

Just like Akosf stated in

Just like Akosf stated in another thread the problem would be the PCI bus, the amount of memory transfers a 50gflop card needs would keep the central CPU so busy you could not use the computer for other daily functions.

However... using specific models of video cards could be a great source of additional processing power. Just think, a quad SLI system that was able to use all 4 video cards like a extra cpu's. Here is some more info on the subject: http://news.povray.org/povray.beta-test/thread/%3C4nvlm1dqoh3hb6i6k0e58s1adgmdgcu4u5@4ax.com%3E/

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4527270
RAC: 0

Coprocessor for

Coprocessor for Opteron

This technology offers about 10-100 times preformance improvements.

1 x Opteron -> 3.2 - 4.5 GFlops
+ 1 x DRC -> 32 - 450 GFlops

RenaudKener
RenaudKener
Joined: 11 Jun 06
Posts: 91
Credit: 5614714
RAC: 0

RE: Greetings. I stumbled

Quote:

Greetings.

I stumbled upon a product called the ClearSpeed Accelerator card. The card uses dual CSX600 processors and interfaces using an PCIE bus. More than one of these cards can be installed in a single system.

Code porting is apparently not required as the card " accelerates standard math libraries".

Would this product be useful with the Einstein@Home BOINC client?

Can anyone get one donated to phys.uwm for evaluation?

The company has not shared cost info with me but I imagine the cost is beyond practical for a general home PC user.

I didn't post the company's url here but you should be able to guess what it is.

Cheers.

I haven't found any hints about Clearspeed CSX600'price, however here's a link you may find usefull
http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=49400990

here's the extract :
" ClearSpeed described the CSX600 as its "first commercial microprocessor" although the CSX600 was said to have doubled the performance of ClearSpeed's CS301, which was described as having a unit volume price of $975 in November 2003.

"

"Entia non sunt multiplicandam praeter necessitatem"
(OKHAM)

ExtraTerrestrial Apes
ExtraTerrestria...
Joined: 10 Nov 04
Posts: 770
Credit: 536594329
RAC: 187645

I read about the details of

I read about the details of the card some time ago. The concept is nice and may actually look like something that intel or AMD may implement in a heteroegeous many-core processor in 10 years or so. But as a damn expensive add-on card it's not very useful. I can't remember the exact price, but it was >10,000$.

MrS

Scanning for our furry friends since Jan 2002

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.