A bit of traffic … for Pegasus!

someone was apparently looking at our Pegasus GPU+Cell box spec’s online … and told their friends.
Most of the comments were ok, though someone thinks this is not a deskside/desktop box.
They wrote:

Great, another low production piece of hardware from a small company, that is not designed primarily to be a desktop computer.

Heh … Won’t dispute the low production piece … we are not making millions of them. As for a desktop computer?
That is most assuredly what this is.
See for yourself
Now imagine throwing even more cores and GPUs at it. This is what Pegasus does. It is not designed to compete with Dell units, or other high volume units, as they are targeted for a decidedly different set of markets. Pegasus can be a deskside super, or, if you indulge yourself like this, a super incredible game machine (up to 7x Quadro FX3800’s in a single box … not sure if the driver is tested for that many), or a tremendous horsepower visualization and visual processing system.
They make a great tie in to clusters of all sizes/shapes.

3 thoughts on “A bit of traffic … for Pegasus!”

  1. Hmm, GPU + Cell. Seems interesting. Do you have any interesting use case for it? Or is it just to “be different”?
    On a related but different note, honestly. why should anyone, I mean _anyone_ bother with writing code for Cell? It’s future sounds to me to be a deadend. When it launched, IBM said that they were gonna come out with an enhanced Cell in ~2010-11 timeframe. But right now, it seems all but abandoned. If I am not sure if it’s offspring will be born, why waste effort to port code and then port it back?

  2. @John
    The latter. That is a live desktop environment that I was playing many videos (streaming from Delta-V) while spinning molecules, etc. This was a slow version of our Pegasus-GPU unit.
    The fast one is … well … very fast.
    GPU power available to Cuda-izable apps and graphics apps. Cell power for existing development in SAR, imaging, etc. The “be different” part is actually the underlying computing substrate … the up to 16 cores, and up to 256 GB ram, and lots of IO horsepower.
    Cell has its place … and it had a good edge early, if in nothing more than hype. TerraSoft, before they were bought by FixStars was the tireless promoter of Cell … almost more-so than IBM, for HPC and other tasks.
    Cell’s issue prior to the GA-180, was no low cost development platform. GA-180 is a reasonably priced card. Coupled with YDL (Kai’s work at TerraSoft translated to Fixstars), you have something that you can actually work on. Better than the PS3 cluster I believe. And a full Linux toolchain via YDL.
    The GA-180’s are PowerXCell 8i’s as I remember. Similar to the Roadrunner bits.
    I expect that GA-180’s and Cell in general will continue to be of interest to image processing/analysis folks. GPUs I expect to continue/grow to be used in many places. If the tool chains were sufficiently similar, and the languages sufficiently abstract, I bet you could substitute the most appropriate one at the right time for the best bang per buck/watt.
    Curiously enough, this is something we had written into our accelerator business plan several years ago. Never was able to get funding for it.

Comments are closed.