[gmx-users] ClearSpeed Performance and availability

[Erik Lindahl]

Hi Florian,

On Nov 15, 2005, at 6:22 PM, Haberl Florian wrote:

> Hi,
>
> a german IT news site ( http://www.heise.de/newsticker/meldung/66214 )
> announce the availability of clearspeed CXS600 coprozessors.
>
> optimized fftw libaries are also available
> http://www.clearspeed.com/downloads/Architecture%20Whitepaper.pdf
> http://www.clearspeed.com/downloads/AdvanceAcceleratorBoard.pdf

> there is also gromacs mentioned.
>
> I  found only this link in google cache:
> http://66.249.93.104/search?q=cache:FZnb6apxi_EJ:www.clearspeed.com/ 
> bio/+csx600+gromacs&hl=de
>
> Main question is: How expensive are such boards, anyone knows and  
> are this
> claimed performance rates for real?
>

Yes, it's for real (I've worked with them on it), but there are some  
caveats:

1. It is not yet a general speedup - right now we only have some  
cutoff and reaction-field interactions.
2. It is definitely a best-case speedup.
3. It's not yet ready for general consumption, but the idea is that  
it will be when we're done.
4. I don't have any recent or accurate price estimate, but think in  
the ballpark of a couple of thousand $ for a dual-processor plugin  
board (I could be off here, though), depending on the market  
situation for the new chip.

Cheers,

Erik
-----------------------------------------------------------
Erik Lindahl  <lindahl at sbc.su.se>
Assistant Professor, Stockholm Bioinformatics Center
Stockholm University, SE 106 91 Stockholm
Phone: +46 8 5537 8564     Fax: +46 8 5537 8214