At SC11 I ran into Henry Newman, CEO of HPC consulting firm Instrumental Inc. After exchanging the usual pleasantries and deeply offensive personal insults, we got to talking about some of the recently released benchmark results – and how irrelevant most of them are to the real world. In the course of the conversation, Henry told me that he was once a “slimy benchmarker.”

He told me about some of the tricks that vendor benchmarkers use to make sure their systems shine brightly in customer bake-offs. Basically, if a particular trick (special tuning, hardware, etc.) “isn’t clearly and explicitly banned by the customer, then it’s fair game.” I get the feeling that back in the day, Henry was a particularly savage and merciless benchmarker with a pretty good win/loss record.

But now that he’s on the customer side of the industry, Henry has a different point of view, and wants to become part of the solution to the benchmarking problem (which he helped promulgate, damn it.) Toward that end, he’s spent a lot of time working on a project with DARPA to develop a set of benchmarks that get to the heart of what customers really need to know: how well systems scale from small to large.

In the webcast we talk about all of the above topics, drill down into specific scalability benchmarks, and give Henry a chance to confess his benchmarking sins. Give it a listen here…

Share this post

 

A quick meeting with HP at SC11 confirmed that the company is feeling good about their HPC achievements and prospects for the future. HP is the second biggest HPC vendor on the most recent Top 500 list with 141 systems (28%). However, they’re still behind market leader IBM, who has a 44% share with 223 total systems.

The HP picture is worse when you look at a comparison of system size and performance. IBM has 26 of the top 100 systems, while HP has only seven. Of course, one of those seven includes the 1.19 petaflop NEC/HP TSUBAME 2.0 system that’s #5 on the list, which isn’t too shabby.

So why is HP smiling about their HPC chances? First, according to the company, they don’t measure their HPC success by appearances on the Top500 list. They (correctly) assert that there’s plenty of profitable HPC business to be had at smaller customers with smaller-than-Top500 systems. (Read more below...)

Share this post

 

As I trudged toward a swanky hotel for a meeting with Dell, the Seattle sky was spitting cold rain like an old man realizing the soup in his mouth is way too hot. (Adding more drama to these intros, nice, right?)

I expected two things that day: “It’s Seattle in November; it’s going to rain,” and “It’s Dell at Supercomputing; they’re going to talk about hardware.” Only one of those assumptions was correct.

Instead of talking hardware, the meeting was all about Dell’s HPC strategy and how they’re going to engage the market. It wasn’t a typical Dell-like meeting, where they’d reel off server names and configs and I’d nod appreciatively, “Hmm… so you’re going to put newer/faster processors in that one? Way to go, nice job...” (Read more below...)

Share this post

 

One of the most interesting things I saw at SC11 was a joint Mellanox and University of Valencia demonstration of rCUDA over Infiniband. With rCUDA, applications can access a GPU (or multiple GPUs) on any other node in the cluster. It makes GPUs a sharable resource and is a big step toward making them as virtualizable (I don’t think that’s a word, but I’m going with it anyway) as any other compute resource.

There aren’t a lot of details out there yet beyond this press release from Mellanox and Valencia and this explanation of the rCUDA project.

This is a big deal. To me, the future of computing will be much more heterogeneous and hybrid than homogeneous and, well, some other word that means ‘common’ and begins with an ‘h’. We’re adopting the mindset of designing systems to handle particular workloads, rather than modifying workloads to run sorta well on whatever systems are cheapest per pound or flop. (Read more below...)

Share this post

 

One of the presentations I caught at SC11 was by GPU computing pioneer Ian Buck. (Which is a good name for a pioneer, I think, although just ‘Buck’ might be better. I’ll go with that for the rest of this article.)

Buck’s Stanford Ph.D. thesis, “Stream Computing on Graphics Hardware,” capped his research into using GPUs as computing resources and his work to develop ‘Brook,’ one of the earliest programming languages aimed at GPUs.

This work, of course, caught NVIDIA’s attention. They brought Buck aboard six years ago; he founded NVIDIA’s CUDA team, and the rest is sort of history. History that he laid out in his talk at SC11 (video here). He takes us from the earliest days (2002-03) of using GPUs as accelerators to where it is today. And, by the numbers, they’ve come a long way. (Read more below...)

Share this post