A Petaflop machine in 20 racks?

Jim Lux james.p.lux at jpl.nasa.gov
Thu Oct 16 19:46:19 EDT 2003


Browsing through ClearSpeed's fairly "content thin" website, one turns up
the following:
http://www.clearspeed.com/downloads/overview_cs301.pdf
The CS302 has an array of 64 processors and 256Kbytes of memory in the array
+ 128 Kbytes SRAM on chip.  That's 4 Kbytes/processor (much like a cache)..
It doesn't say how many bits wide each processor is, though..
51.2 Gbyte/sec bandwidth is quoted.. that's 800 Mbyte/sec per processor,
which is a reasonable sort of rate.

10 microsecond 1K complex FFTs are reasonably fast, but without knowing how
many bits, it's hard to say whether it's outstanding.
It also doesn't say whether the architecture is, for instance, SIMD.  It
could well be a systolic array, which would be very well suited to cranking
out FFTs or other similar things, but probably not so hot for general
purpose crunching.

For all their vaunted patent and IP portfolio, they have only one patent
listed in the USPTO database under their own name, and that's some sort of
DRAM.

----- Original Message -----
From: "Bryce Bockman" <bryce at jfet.net>
To: <beowulf at beowulf.org>
Sent: Thursday, October 16, 2003 1:15 PM
Subject: A Petaflop machine in 20 racks?


> Hi all,
>
> Check out this article over at wired:
>
> http://www.wired.com/news/technology/0,1282,60791,00.html
>
>   It makes all sorts of wild claims, but what do you guys think?
> Obviously, there's memory bandwidth limitations due to PCI.  Does anyone
> know anything else about these guys?
>
> Cheers,
> Bryce
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf

_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list