Showing posts with label supercomputing. Show all posts
Showing posts with label supercomputing. Show all posts

Monday, October 15, 2007

Sun's not very contrarian thinking

Since Sun seems to be treating the launch of their new T5x20 servers as the second coming (and it is, of Niagara, that is), why haven't they done the logical thing and proposed a Niagara/ClearSpeed combo?

Various reasons indicating such a combination would be in demand include:
  1. Power usage
    Since the ClearSpeed boards have approximate power consumption of 25-33 Watts (depending on model), adding two of the PCIe-8x boards to a a T5220 would only increase power usage by 66W with an aggregate 133 GFLOPS increase in computing power. Sun obviously believes stuffing ClearSpeed boards into their servers are a valid play: TIT's Tokyo Tech's TSUBAME cluster achieved a 24% speedup for a 1% increase in power usage.

  2. Floating Point performance
    Real World Tech's excellent article posits that the floating-point ratio in the instruction mix on the first Niagara iteration could exceed no more than 1%-3%. Clearly, the demand for floating point was sufficient to increase the resources dedicated to it from 1 FP unit per chip, up to 1 FP unit per core. Presumably further FP acceleration would be even more beneficial.
Of course, Sun might have more than a little trouble with the integration. Perhaps they might have more luck hooking up some ClearSpeed X620s into a Thumper...

Tuesday, October 02, 2007

Silicon Graphics on the comeback trail?

Thank you insideHPC.com
SGI: “We will be back in the visual supercomputing business”
I’m at the second day of IDC’s HPC User Forum today. Bo Ewald, SGI’s president, just said something I think is interesting
We will be back in the visual supercomputing business.
He bookended that by saying he wasn’t going to say anything else today, but that there would be more in the future.
Later in his brief he followed up that statement with this quotable quote
It was really stupid for the company to stop doing visualization types of things.


Heck yeah!

Friday, September 21, 2007

Folding@Home > 1 PetaFLOPS

Checking the Folding@Home stats you can see that this distributed computing effort now tops out at over 1 PFLOPS, currently sitting at ~1.1 PFLOPS.

Mmmm, and costing a lot less than the NCSA's new toy.

Monday, August 06, 2007

Aug. 6, 1945: 'I Am Become Death, Destroyer of Worlds'

Saw this in the morning feed:

Aug. 6, 1945: 'I Am Become Death, Destroyer of Worlds'

and realised I'd forgotten how long it had been since this actually happened - over 60 years! And they didn't have shiny new [$200m super]computers to do the calculations.

Quite interesting that the NCSA - one of the US groups that actually have the word in their name is finally coming back to prominence after what seemed to be a period of being eclipsed by the National Labs.