Cell to be used in world's fastest supercomputer at Los Alamos (Opteron-Cell hybrid)

[nAo]

I know that, but like you said, it's not very fast. That's the point of approximating with FP numbers, so you can boost the speed rather significantly.

What does make you believe that a sw emulation may be (significantly?) faster than the actual not-so-fast hw implementation?

 

[3dilettante]

... Ok so now that you have all wiped your desks, what is this kind of computing power going to be used on, apart from protein stuff?

Won't they run Katamari Damaci on it, not even once? What a waste. I mean, it will probably run at 1,000,000,000 frames per second, the game will be over before you even realise that you've turned the thing on, but still...

Interesting thing to note is that if the entire system were tasked to run the simple Katamari Damacy, it would probably run at tens of frames per second.

Supercomputers are meant to tackle huge problems in a reasonable amount of time, not small problems very quickly.

If they partitioned off one chip to run it, they'd probably be great. Or they could run tens of thousands of instances of the game, all at moderate frame rates.

 

[Fox5]

Maybe the bulk of the DP is being done by Clearspeed processors? 2x 25 DP-GFLOPs per board. 16,000 boards makes 800 TFLOPs.

Curious why Clearspeed isn't mentioned as being part of the project. Maybe it's just about marketing Cell

Jawed

Clearspeed is in there too? What's the purpose of having Cell and Clearspeed? Are they just making an extremely general purpose super computer to dominate any benchmark?

If they partitioned off one chip to run it, they'd probably be great. Or they could run tens of thousands of instances of the game, all at moderate frame rates.

Raytrace it and run it at the tens of frames where any normal computer wouldn't even get 10ths of frames.

 

[SPM]

Clearspeed is in there too? What's the purpose of having Cell and Clearspeed? Are they just making an extremely general purpose super computer to dominate any benchmark?

Clearspeed lost to Cell no?

Cell will comprehensively beat the crap out of Opterons or any other conventional CPU for supercomputing applications - the numbers indicate that quite clearly - 16000 Cell processors + 16000 Opterons processors gives four times the speed of blue gene the previous fastest supercomputer with 130,000 conventional processors. Cell = 256 G Flops Opteron =?? no contest!

There are two possible reasons for the Opterons. The first is compatibility with existing C and Fortran code, and mature compilers and development tools. Using Opteron will allow existing applications like finite element analysis and matrix libraries to be recompiled quickly for the new supercomputer, with only the critical sections requiring complete rewriting for Cell. The second reason possible reason, is that the Opterons are there to boost i/o performance. What the SPEs are particularly bad at is i/o performance which requires buffering and caching of large amounts of data between RAM and disk/network. This requires large stacks, pointer addressed data structures in RAM, something which the SPE architecture really isn't built for handling. The PPEs are probably fully occupied feeding the SPEs, in which case it would make sense to add another processor to handle i/o.

Los Alamos isn't going to pay any attention to published benchmarks - they will benchmark the programs they actually want to run themselves, so artificial benchmarks aren't going to help sell it. The same applies to any other supercomputer customer - they are technically competent enough to program and benchmark the supercomputer themselves.

 

[one]

Maybe the bulk of the DP is being done by Clearspeed processors? 2x 25 DP-GFLOPs per board. 16,000 boards makes 800 TFLOPs.

Curious why Clearspeed isn't mentioned as being part of the project. Maybe it's just about marketing Cell

The double precision performance is due to the double precision version of Cell. ClearSpeed has nothing to do with the IBM project.

http://www.hpcwire.com/hpc/893353.html

HPCwire: What is the significance of the Roadrunner deployment? Is it a one-off system or does it represent the start of a new line of IBM supercomputers?

Turek: The significance of Roadrunner is that this is our preferred architectural design for the deployment of Cell in the HPC application arena. To be clear, we have no plans to build a giant cluster just out of Cell processors. Instead we think the Roadrunner model is the correct model which employs Cell as an accelerator to a conventional microprocessor-based server.

Over the course of time, we expect accelerators to become a key element to our overarching strategy. So the work that we do here is designed, in particular, to be sufficiently general to encompass a variety of models on how accelerators might be deployed.

Our intention with respect of a more broadly propagated version of Roadrunner is an assignment we've given ourselves for the fall to see exactly how far this can be extended and how deeply it can be played in the marketplace. We've got to resolve programming model issues. Secondly, the early Cell deployment is based on single precision floating point; that's going to go to double precision [for the final deployment]. So there's work to be done here to see exactly how this plays out.

...

HPCwire: You said that the final deployment of Roadrunner will incorporate a double precision floating point implementation of the Cell processor. What will you be accomplishing in the early stages of Roadrunner that uses the single precision version of Cell?

Turek: The early deployments of Cell are really meant to help us deploy and debug all the software tools and the programming model. All that gets preserved regardless of whether you're single or double precision. And then as we go down the path of producing the double precision Cell B.E., that will be more a matter of deployment and scaling issues than it will be to the specification programming models, software tools and things of that sort.

 

[Arwin]

So there we have it - it will be the DP Cell ... thanks again, one, you are always a true fountain of useful information.

 

[Jawed]

Nice to see some actual answers on this topic.

Jawed