Parallel Computing Platform: Overview and Future Directions

Posted By: Charles | Oct 13th @ 8:00 AM

The Parallel Computing Platform team's David Callahan (Distinguished Engineer) and Keith Yedlin (Product Unit Manager) sit down with me to discuss Microsoft's vision for general purpose Parallel Computing. Keith demos the power of the Microsoft Parallel Computing Platform on a 26 processor machine. Even cooler than the substantial performance improvements is the small change to a for loop (adding .NET Extensions' new Parallel.For) that results in incredible performance gains. See for yourself.

Of course, this is Channel 9 (and I am without Marketing bones...) so we talk about a lot of things related to concurrency and parallel computing. Do we have all the answers? Of course not. Are we making significant strides in our quest to bring more super computing capability to the general purpose software developer. David is a seasoned veteran in the parallel computing space (he was an engineer at Cray...) and he owns the Microsoft "Quest" for parallel computing, generally. Keith has to make sure that we come up with is truly useful to customers and not just simplified rocket science.

Tune in!

Tags: Parallel Computing, Parallel Computing Platform, Parallel Extensions, PLINQ

Rating:

Views (44,289)

Comments (8)

Downloads

figuerres
???

26 procs... Wow, what kind of MoBo and PSU for that.... I Bet it's a freaky big case and heat sink setup Smiley

kettch

In reply to figuerres

#Oct 13th @ 2:59 PM

Most likely some kind of blade system? Maybe?

Since we keep hearing about how the teams working on parallelism are testing things out on massive many-core machines, can we have another video about the testing labs? There was a cool one back during the Vista beta cycle, and it would be neat to talk about what kind of massive hardware these frameworks are being tested against. It's nothing important, just a "cool factor" thing since a 26 core machine is way beyond the budget of most of us...combined. Smiley

Judah

Charles, you said in the interview that parallelism is not a key concern for C# developers.

Bah!

Parallelism is huge, huge, huge for us. We've fought so many times with locks and deadlock problems and throughput in our code; we want better parallelism abstractions. I'd say it's #1 or #2 on my wishlist for C# vNext.

Thanks for the interesting video!

Charles
Welcome Change

In reply to Judah

#Oct 13th @ 10:32 PM

I did? Certainly not what I meant to say given the importance of parallelism for all developers going forward, especially main stream (general purpose) developers... The whole point of Parallel Extensions for .NET is to bring parallel computing to the masses.

aL_
mm.. linq.. *arrggha*

so what happend to the acelerator team? i heard them beeing mentioned in another video but finding anything out about them is really hard :/ gp-gpu computing is huuuuge and if you want many core... the ati 4700 has 800 (eight hundred) processors.. thats manycore right there Tongue Out

and th X2 has 1600 processors..

the power of gpu:s is staggering, and microsoft has this secret awsome .net project that could harness that Smiley

aL_
mm.. linq.. *arrggha*

yay, mention of gp-gpu computin around 23:30 Smiley

but the accelerator guys did some gpu computing using dx9.. that stuff goes a long way Smiley

anyhow, i've said it before but i'll say it again, pleease make some interviews with the accelerator team (or what ever they are called now) Smiley

and th X2 has 1600 processors..

the power of gpu:s is staggering, and microsoft has this secret awsome .net project that could harness that Smiley

aL_
mm.. linq.. *arrggha*

sorry for double posting (and posting weirdly) but the edit post function doesnt seem to work.. (tried ie8 and ff3 :/ )

Buanderie
Mr Connard

Well... I can say that the Accelerator team is still alive. I frequently send them bug reports Tongue Out

Truth is, Accelerator by now is far from being a potential concurrent for other GPGPU libraries such as CUDA.
Accelerator now just provides access to the ParallelArrays lib which just provide basic operations on arrays, and delayed computation. For basic operations it's great. But when you want to work on advanced GPGPU projects, it's not enough... You don't have enough control on what you're actually programming. While using Accelerator you actually have no idea of what's going on the GPU. I hope the next version will provide more low-level control.