NVIDIA’s Tesla T10P Blurs Some Lines

and also this

["SAN JOSE, Calif. ? Apple Inc. has submitted the Open Compute Language (OpenCL) to an ad hoc industry group that aims to define a programming environment for applications running across both x86 and graphics chips. The move is one of a growing number of efforts to extend the ubiquitous C language for increasing parallelism in multicore processors.]

Next, will we see boards with GPU and FPGA ?

-jg

Speaking of FPGA alternatives, this recently caught my eye. Don't know much about it, but it sure looks cool:

-Jeff

Jeff, Where have I seen that before? Ah yes,

Syms.

And now, for bonus points, put an [X] against any of the following reasons if you think it helps to explain why the Transputer was a market flop...

[ ] it was at least two decades ahead of its time [ ] it was about one decade ahead of the technology needed to make it powerful enough to be competitive [ ] it was based on robust, sound academic theory [ ] it had a clean, effective, readable programming language that was not C [ ] it was a British design [ ] the software community was even more clueless about how to make use of multiple scalable compute resources in the late 1970s than it is today [ ] when presented with a model that permits parallel processing to be spectacularly efficient, the software community retreats into its standard cosy set of assumptions that serial execution is sure to be faster and more efficient, and therefore Concurrent Is Bad, mainly because it isn't C

For maximum credit, write a 10,000 word dissertation explaining why the social dynamics of the software community will ensure the early death of anything that looks like a general-purpose, tightly-coupled network of compute nodes.

Not that I'm bitter, or anything like that :-)

--
Jonathan Bromley, Consultant

DOULOS - Developing Design Know-how
VHDL * Verilog * SystemC * e * Perl * Tcl/Tk * Project Services

Doulos Ltd., 22 Market Place, Ringwood, BH24 1AW, UK
jonathan.bromley@MYCOMPANY.com
http://www.MYCOMPANY.com

The contents of this message may contain personal views which 
are not the views of Doulos Ltd., unless specifically stated.

worse than that: it was a Government funded design under a government that didn't fund technology...

Definitely not this one.

as well as O***m, they came up with Ada, where you can: create a task type; create an array of tasks of that type; loop over the array starting the tasks; loop over the array collecting the results; in pretty much no more code than the above.

Nowadays, C++ has given the SW community ONE positive thing: the old argument that Ada is too large and complex to be usable no longer holds water...

But count the hand-wringing woe-is-me stories about the difficulty of parallelism and the massive new collaborations to solve the problem on one hand, and the mentions of Ada (or O***m) on the other...

The funny thing about the transputer (from the VERY little I did on it at the time; paper only since I couldn't afford one as a hobbyist) is that there was a time window when it appeared to be the fastest single processor available, (ignoring 3-chip solutions) even without its hooks for parallelism... but that got overshadowed by the parallelism.

If it had had an MMU, it could have made a nice Unix workstation (or single-chip Lilith, or...)

- Brian

What is missing is an embedded solution. To my knowledge, there is no compact embedded system using the latest Nvidia GPUs. We can't fit a PC (even a smallish one) inside our instrument. Even if we could, heat dissipation would be a problem. So I guess that FPGAs will still reign for a while in high performance embedded applications.

Patrick

Availability? Price?

Dave

Nvidia. Google.

Yes. Cheap.

MikeWhat?

Really? I got _no_ hits on google for the middle speed-range part number. Where did you find it available from? Price?

Dave

The C870 is $1300, Nvidia direct or second sourced. The C1060 is slated for fall release.

Try snipped-for-privacy@nsc-nvidia.com for more directed info. Would you like the phone number also? Here: For Information on How to Purchase Call 408.392.4120

Second cousin.

(Full initials. Also a pronunciation aid. You get my age and you'll find the dimunitive inappropriate also.)

Ahh. The old switcheroo. Jeff wrote about the Tilera TILE64 processor and I asked about its availability and price. You decided to give availability and price (sort of) on Nvidia's Tesla GPU card.

I thought your second cousin might explain what Nvidia had to do with Tilera. I see another branch of the thread has cleared that up! :-) Cheers, Syms.

Explains why I don't mind top-posting as much as others. :) Didn't notice the change in topic.

I also didn't find info on the Tilera. Seriously thinking about ordering the C870 here, though.

The Nvidia store lists the C870 as "Available for Backorder" so you may have a wait to get one. The D870 is in stock and is only $5K. ;-)

Dave

Symon (symon snipped-for-privacy@hotmail.com) wrote: : Jeff Cunningham wrote: : >

: > Speaking of FPGA alternatives, this recently caught my eye. Don't know : > much about it, but it sure looks cool: : >

: >

: >

: > -Jeff

: Jeff, : Where have I seen that before? : Ah yes,

: Syms.

One of the things that strikes me about the Transputer is that it was parallel hardware designed hand-in-hand with parallel software.

Not C/C++

It seems odd that there is so much convergence happening between a very complex highly sequential CPUs and functionally simple, highly parallel FPGA type devices, with lots of innovative hardware flying about.

All this convergence is happening in hardware, but for it to really work don't the software environments need to do the same...

--
cds

The new XMOS devices (out of the same stable as the transputer) are intended to bridge the gap between FPGAs and processors:

Leon