Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Opteron HyperTransport

Author: Robert Hyatt

Date: 08:49:39 07/08/03

Go up one level in this thread


On July 08, 2003 at 00:33:09, Jeremiah Penery wrote:

>On July 07, 2003 at 23:33:48, Robert Hyatt wrote:
>
>>On July 07, 2003 at 11:30:34, Vincent Diepeveen wrote:
>
>>>The total bandwidth a 256 node machine with bricks of 4 processors will have
>>>then a total bandwidth of somewhere around: 12.4 Terabyte a second.
>>>
>>>Now *that* is very impressive.
>>
>>And imaginary I'll bet.
>
>Cray is supposed to be building a 10k processor Opteron supercomputer for Sandia
>Labs.  Its total aggregate bandwidth should far exceed 12.4TB/sec.


Again, that is a theoretical peak, which assumes _all_ processors are
communicating with each other in such a way there are no conflicts or
hot spots.  The 48 gb/sec for a single T90 CPU doesn't have that kind of
assumption.



>
>>>Especially for the price they will deliver it for.
>>>
>>>A 5000 processor vector machine now costs around 700 million dollar.
>>
>>Where can you find a "5000 processor vector machine?"
>>
>>I know where there is a 32 processor vector machine.  Nothing beyond
>>that that I know of, unless you start counting I860 type boxes.
>
>NEC Earth Simulator has 5120 NEC SX-7(?) vector processors.  Total cost was less
>than $400m.

OK.  That's a cluster.  In that light Cray Research used to have a big cluster
of YMPs at their headquarters.

I have run Cray Blitz on an SX in the past.


>
>Here is a blurb about the chip, from the webpage:
>
>"Each AP consists of a 4-way super-scalar unit (SU), a vector unit (VU), and
>main memory access control unit on a single LSI chip. The AP operates at a clock
>frequency of 500MHz with some circuits operating at 1GHz. Each SU is a
>super-scalar processor with 64KB instruction caches, 64KB data caches, and 128
>general-purpose scalar registers. Branch prediction, data prefetching and
>out-of-order instruction execution are all employed. Each VU has 72 vector
>registers, each of which can has 256 vector elements, along with 8 sets of six
>different types of vector pipelines: addition/shifting, multiplication,
>division, logical operations, masking, and load/store. The same type of vector
>pipelines works together by a single vector instruction and pipelines of
>different types can operate concurrently."
>
>Each chip consumes only about 140W, rather than Vincent's assertion of 150KW.


The 150KW was for a machine like the C90/T90/etc.  125KW might just get them
started.  :)



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.