[Beowulf] Vector coprocessors

Joe Landman landman at scalableinformatics.com
Thu Mar 16 12:03:07 PST 2006

Jim Lux wrote:
> At 09:54 AM 3/16/2006, Joe Landman wrote:


>> I think (WAG here) that Clearspeed wants people to design a socket 
>> onto motherboards for them.  Lowers the costs all around.
> But that greatly increases the cost of the motherboards, and mobo 
> manufacturing is a very price sensitive business.  every square cm of 
> board costs a bunch, all the way through the whole supply chain (mfr, 
> storage, shipping, etc.)

Yes it does.  You would not expect this in the high volume MB business 
which is extremely price sensitive.  You might expect this on some of 
the "specialty" server boards.  Not unlike the HTX connector.  BTW:  any 
other HTX cards out there apart from Infinipath?  I had heard about one 
or two others (RAID/10GBe) but haven't seen too many.

> Interesting that there was HUGE economic pressure to integrate the FPU 
> onto the chip after the 386/387 era. And the whole reason the x87 wasn't 
> integrated wasn't because of marketing, but because of die size 
> limitations.  

Yup.  Remember that.  I remember sticking the x87 in the socket and 
adjusting toggle switches.  Then coding to it in assembler, and setting 
up a TSR to hook int 21.

> With the 486, the dies got big enough to integrate, and 
> they've never looked back.  I haven't even seen any mass-market 
> coprocessors of that type (i.e. closely coupled to the CPU) since then.  
> The coprocessors have tended towards performing some specific function 
> (e.g. graphics, or PCI bridge with integrated peripherals) and have been 
> fairly loosely coupled.

This is the coprocessor model that generally works.  Server systems have 
a minimalistic video chip on the MBs typically.  The cost is very low to 
do this.  An PCIe accelerator board, be it a graphics, IO raid, 
networking, or computational accelerator is a reasonable way to add the 
offload processing functionality in if you need it.

>>>> I would say that there is more potential for a clever soul to 
>>>> reprogram the guts of Matlab, etc., to transparently share the work 
>>>> across multiple machines.  I think that's in the back of the mind of 
>>>> MS, as they move toward a services environment and .NET
>>> Lots of people have thought about that for a long time, including
>>> Cleve Moeller.   The potential clever soul should be well above
>>> average, and considering MS products, well above MS average programmer.
> At least there's lots of MS programmers to draw from, so on a 
> statistical basis, you'd expect one to pop up.
> jim

Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 734 786 8452
cell : +1 734 612 4615

More information about the Beowulf mailing list