HECToR, the UK’s latest national supercomputing service, is the subject of an article in the Guardian newspaper today. The 60TF Cray XT4 is described as:
To the untrained eye, it looks like a sports hall half filled with neatly arranged burgundy Ikea storage units.
The key for the story is the news that the service will be officially opened by the UK’s Chancellor Alistair darling later this month. In a neat twist, Darling is both the local MP for the region hosting the supercomputer, and also ultimately responsible for the government’s £113M bill for the system.
The article also notes that although the Cray supercomputer itself cost £5.6m less than the £65m capital budget, the HECToR annual running costs have jumped from the £5.4m budget to £8.2m. This is attributed to the doubling of electricity prices in the UK since the planning stages.
That’s my new baby. It’s nice to see it be recognized.
A quick note, Not only has the price of electricity caused the operating cost to increase, but also the fact that AMD didn’t give good numbers for the power consumption of their chips. That has REALLY bitten us on the butt.
When you’re trying to do power consumption estimates for a 10,000 core system and your numbers are flawed from the beginning, well it can get ugly.
Hope everyone has had a great New Year.
Rich Hickey
Rich – that sort of thing is exactly what I was writing about at InfoWorld (story here). The business model between HPC vendors and chip manufacturers is fundamentally broken, and needs to be replaced with one that ties the chip makers to the performance promises they make for their goods.
I didn’t realize that you wrote for InfoWorld also. That was news to me. pun intended. Congrats, by the way.
I just read the article and you are very correct. From someone who works for, and supports, a supercomputing company, (plug for Cray here), we’ve had a very rough time lately with the issues of AMD’s production delays.
I have no idea of how much money those delays have cost Cray both monetarily and also in labor. We’ve shipped several high cabinet count systems that should have gone out the door as Quad-Core, and instead we had to ship as dual core with the intent of upgrading them shortly out in the field. Anyone who has done anything similar knows the nightmare of this.
Hmmm… I think “Your” system is being shipped this way if rumors are correct.
I like the way you thought this out and phrased it.
But for those of us building parallel computers these errors are a big deal. An 11 watt per socket error in a 1,000 or 10,000 socket machine adds up to real money, if the customer can even support the increased power requirement.
Rich
hmmm… I forgot that your system uses html translation. My “snips” before and after your quote went the way of the dinosaur. Sorry about that. I did try to give you credit for that last quote.
Rich
Rich – no worries at all. The HTML snipping is an attempt to control spam, of which we (inexplicably) get an unbelievable amount.
My efforts at InfoWorld have been start and stop. They have a new editor in charge of herding us bloggers over there, and we’re resolved to making another run at an HPC-themed blog of more interest to IW’s non-HPC readership. My last run at it wasn’t not all I thought it could have been.
I appreciate your insights from inside a company affected by AMD’s errors. It was exactly Cray I had in mind when I wrote that, motivated by my very first hand experience with “my” system. You’re exactly right about the dual/quad thing. My dually is supposed to be quad enabled in the coming weeks.