From: Quadibloc on
The message header says it...

There are a few other items I've heard of. There's the chip from
ClearSpeed with 96 floating-point processors. There are a couple of
GPU-based floating-point accelerator cards; one is forthcoming soon,
one is here now but more expensive.

There's the new version of the Cell microprocessor with better double-
precision support.

And there's the 16-processor Rock from Sun, and the Larabee
architecture from Intel, with one big CPU and perhaps 32 small ones -
which is going to come out as a GPU-substitute (where, I fear, it
indeed may not perform well, as its detractors claim) before being
used as a CPU (where I think its real potential lies).

I suspect, though, that there are other high-performance technologies
in development out there that I just haven't heard of. Even without
exotic technologies, for example, there's the Mac Pro as the most
readily available 8-core system (two Intel quad cores), and a company
called BOXX, I've just found out today, makes workstations with four
or sixteen quad-cores from AMD. And an outfit that makes a small
compute server with many chips advertised in Linux Journal not long
ago (came with Microsoft Windows Compute Server, though, ironically).

John Savard
From: Chris M. Thomasson on

"Quadibloc" <jsavard(a)ecn.ab.ca> wrote in message
news:a6abcde5-f407-444a-9eb7-ecffb3926cbf(a)k7g2000hsd.googlegroups.com...
> The message header says it...
>
> There are a few other items I've heard of. There's the chip from
> ClearSpeed with 96 floating-point processors. There are a couple of
> GPU-based floating-point accelerator cards; one is forthcoming soon,
> one is here now but more expensive.
>
> There's the new version of the Cell microprocessor with better double-
> precision support.
>
> And there's the 16-processor Rock from Sun, and the Larabee
> architecture from Intel, with one big CPU and perhaps 32 small ones -
> which is going to come out as a GPU-substitute (where, I fear, it
> indeed may not perform well, as its detractors claim) before being
> used as a CPU (where I think its real potential lies).
>
> I suspect, though, that there are other high-performance technologies
> in development out there that I just haven't heard of. Even without
> exotic technologies, for example, there's the Mac Pro as the most
> readily available 8-core system (two Intel quad cores), and a company
> called BOXX, I've just found out today, makes workstations with four
> or sixteen quad-cores from AMD. And an outfit that makes a small
> compute server with many chips advertised in Linux Journal not long
> ago (came with Microsoft Windows Compute Server, though, ironically).

AMD is working on scaleable atomic transactions therefore making highly
advanced non-blocking synchronization algorithms come to life on a
mainstream platform...

From: Chris M. Thomasson on
"Quadibloc" <jsavard(a)ecn.ab.ca> wrote in message
news:a6abcde5-f407-444a-9eb7-ecffb3926cbf(a)k7g2000hsd.googlegroups.com...
> The message header says it...
>
> There are a few other items I've heard of. There's the chip from
> ClearSpeed with 96 floating-point processors. There are a couple of
> GPU-based floating-point accelerator cards; one is forthcoming soon,
> one is here now but more expensive.
>
> There's the new version of the Cell microprocessor with better double-
> precision support.
>
> And there's the 16-processor Rock from Sun, and the Larabee
> architecture from Intel, with one big CPU and perhaps 32 small ones -
> which is going to come out as a GPU-substitute (where, I fear, it
> indeed may not perform well, as its detractors claim) before being
> used as a CPU (where I think its real potential lies).
>
> I suspect, though, that there are other high-performance technologies
> in development out there that I just haven't heard of. Even without
> exotic technologies, for example, there's the Mac Pro as the most
> readily available 8-core system (two Intel quad cores), and a company
> called BOXX, I've just found out today, makes workstations with four
> or sixteen quad-cores from AMD. And an outfit that makes a small
> compute server with many chips advertised in Linux Journal not long
> ago (came with Microsoft Windows Compute Server, though, ironically).

General Purpose GUP programming is interesting. I mean why would I want to
use a CPU to crunch a vector when I can transpose it onto a texture and
render every pixel of that sucker! Pixel==slot in array...