Author: Vincent Diepeveen
Date: 14:15:30 12/13/02
Go up one level in this thread
On December 13, 2002 at 17:02:20, Michael Vox wrote:
it is complete useless for computerchess.
note that when i checked out homepage i could find not
a single technical detail.
like: what speed processors are clocked at.
how much bandwidth the processors have between each other
how much RAM they carry or gates (whatever)
and what the latency is to communicate.
most important is perhaps facts like atomic synchronization which
is a big must for parallel software.
then we didn't hit the issue of hashtables yet.
i bet they are not going to tell you about how hard it is to use
hashtables with such a design.
without hashtables at say 430 processors you can go home directly.
but perhaps the worst part is that when you start writing a chessprogram
for it, even if on paper it is faster perhaps somehow than a current
processor, by the time you are ready writing something decent for it,
itis way slower than one of the processors that's in the shops then.
if it's useless for even computerchess (and remember some of us could
manage to get chess programs into 2KB washing machine chips) then for
sure it'll be useless for your regular home pc too.
>Does this have any effect on regular home PCs in the near future ?
>
>
>http://www.theregister.co.uk/content/3/28491.html
>
>Extreme CPUs: 430 cores on a die
>By Andrew Orlowski in San Francisco
>Posted: 10/12/2002 at 01:55 GMT
>
>
>British start-up picoChip has announced an extreme processor with 430 16bit
>cores on a die, and is testing the first silicon.
>
>It's a pretty radical and interesting approach, which involves a massively
>parallel array of four different kinds of processors, each of which picoChip
>reckons is equivalent to an ARM 9, linked by a switch. The company boasts a
>former ARM board member and a clutch of experienced executives from TI, Philips,
>Marconi and several of the senior management from Oak.
>
>picoChip reckons that its picoArray approach is cheaper than ASICs, and much
>faster than conventional DSP/FPGA combos. The first 160Mhz part, it claims, is
>19x faster than TI's 600Mhz TMS320C6415.
>
>"The architecture supports dedicated resource allocation, and is software
>programmable," Steve Barraclough, picoChip VP of business development told us.
>But developers can verify at the point of compilation. The picoArray can be
>partitioned on the fly, according to the resources required.
>
>It's aimed at a lucrative niche: 3G base stations. Barraclough says picoChip
>will talk to anyone but thinks the biggest market will be WCDMA.
>
>"Twenty to thirty per cent of the bill of materials is in baseband processing,
>and standards are still not fixed," he said.
>
>Other uses for picoArrays could be 802.11 access points or real time crypto or
>image processing. It also supplies the compilers, tools and software.
>
>The company says NEC demonstrated a 512-core CPU at the recent Microprocessor
>Forum, but that these were 8-bit cores. IBM has hundreds or cores at the heart
>of Deep Blue and ASCI White.
>
>So for now, this appears to be the most powerful processor captivity.®
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.