Monday 8 August 2011

Summer season big changes - football or supercomputing?

The world of supercomputing has gone mad.

So it seems as I catch up on the news around the HPC community after a week's vacation. Just today the news of IBM walking away from half a decade's work on Blue Waters and the story of an unknown organisation [now revealed to be NVidia] tempting Steve Scott to leave his Cray CTO role have been huge news but thinking back over the summer months there has been more.

The immediate comparison to me is that of the European football summer season (soccer for my American readers). Key players are signed by new clubs, managers leave for pastures new (or are pushed), and ownership takeover bids succeed or fail. It feeds a few months of media speculation, social gossip, with occasional breaking news (i.e. actual facts) and several major moves (mostly big surprises, but some pre-hyped for long before). But clubs emerge from the summer with new teams, new ambitions, and new odds of achieving success.

The world of HPC has such a summer I think.

There are many places to start, but I'll choose Intel swooping to capture Mark Seager from LLNL. (Probably the result of some earlier winter transfer activity at Intel, with some key retirements etc.) Good for Mark, good for Intel, bad for IBM (LLNL supercomputing, led by Mark, was a good supporter of IBM's supercomputing efforts, especially in the Blue Gene line - from BG/L to BG/Q and the 20 Petaflops Sequoia).

Then soon after, Alan Gara, IBM's chief architect of the Blue Gene systems left IBM, also to join Intel. Again, good for Intel, bad for IBM. An opportunity for other supercomputing leaders at IBM (such as Don Grice). And maybe IBM can find an opportunity from the forced team change, depending on the vision of whoever their next supercomputing technology leader is.

Then today, the news broke that Steve Scott is stepping down as CTO of Cray, having been tempted by a role that is just as "fun and rewarding" at a technology partner of Cray. As I write this, there is no word on the identity of the lucky recipient of Steve's services, but good for them and I hope good for Steve. Bad for Cray, but I think Cray are among the companies that should be able to entice a high quality new CTO to join. Which raises the question as to where Steve is going that is better. Speculation so far includes AMD (CEO?), NVidia (HPC CTO??) [Update: this one! see below], Intel (buying up all the talent - but too many stars in one team?), US Government (DARPA, or other agencies?).

Not to mention the slightly unpredicted winner of the summer Top500 cup - the K computer.

And then, possibly the biggest news story of the summer so far - IBM and NCSA announce that IBM has withdrawn from its contract for Blue Waters - the NSF's Track 1 sustained Petaflops national academic leadership supercomputer - half a decade into the project. Bad for NCSA. Bad for NSF. Bad for IBM. Bad for the expectant academic community waiting for a supercomputer to do research with.

The reasons cited by IBM - "required significantly increased financial and technical support by IBM beyond its original expectations" - simply incites more pressing questions. How bad is the increased expectation of financial and technical support that it is worth the bad PR hit and the refund of ~$30m from IBM to NCSA? Must be a substantial increase in costs and technical support. Having said that, history eventually seems to treated SGI OK after they pulled out of the PSC Track 2 deal.

Is there even a process for NSF/NCSA to replace the technology provider for the Blue Waters goal? [Update: Yes, NSF seem to be behind NCSA - see below] How long will this take, given the developments to date - datacentre, software planning, etc - have been focused on the IBM technology?

[Update:  comment - NCSA is big enough, and full of enough talent, to pull this back on track. This might even be an opportunity for a solution to emerge that is better in some ways.]

I think there is more to come on all of these stories - Intel's talent collection, Steve's mysterious new employer, and IBM's dramatic pull out.

It will be interesting to see in a few weeks/months, when all has settled, what the shape of the new teams in the supercomputing world will be - their new star players, the new goals and ambitions, and who has emerged as winners.

PS - I'm writing this a little hurried, so I'm sure I've missed a key transfer or two. Or have garbled some detail. Or have got some facts wrong. None of those seem to stop the mainstream press when writing about football (or about anything else for that matter) - but please do use the comments below to correct/update me!

Update 1 (Blue Waters): NCSA spokeswoman Trish Barker says NCSA hope to deliver sustained Petaflops with an alternative technology, still on the same 2012 completion date.

Update 2 (Blue Waters): NSF statement - "... NSF has been working with UIUC to identify a solution whereby UIUC can deliver the desired capability in a timely fashion. That plan will require review by an outside panel convened by NSF's Office of Cyberinfrastructure as early as mid-September ..."

Update 3 (Steve Scott): NVidia announces Steve Scott as CTO Tesla - "... 'There are few people on the planet that have Steve's deep system level understanding of high performance computing,' said Bill Dally, NVIDIA's chief scientist."

To follow my football analogy - I think this might turn out to be a key move that changes the balance between the teams vying for leadership of HPC ...

No comments: