- Oct 18, 2009
- 11,684
- 5,225
- 136
Quoted from Semi-Accurate: (Consider the source.) http://www.semiaccurate.com/2009/12/16/oak-ridge-cans-nvidia-based-fermi-supercomputer/
Last edited:
Oak Ridge cans Nvidia based Fermi supercomputer; Fermi power use the culprit
- Thread starter Meghan54
- Start date
- Status
Well...if it performs then us enthusiasts will probably be all over it regardless of power consumption.
Seems odd to me that the power-consumption envelope for Fermi was not well comprehended in the specifications as created by Oak Ridge.
Cancelling a supercomputer project just doesn't happen, scale and scope change to accommodate reality but to just up and cancel it seems odd. What hardware do they intend on replacing the Fermi-spec'ed machine with?
If it was relying on double-precision then there is nothing that delivers that kind of GFLOP/Watt performance anyways. If it was relying on single-precision then it made sense to go with Cypress-based products anyways.
Cancelling a supercomputer project just doesn't happen, scale and scope change to accommodate reality but to just up and cancel it seems odd. What hardware do they intend on replacing the Fermi-spec'ed machine with?
If it was relying on double-precision then there is nothing that delivers that kind of GFLOP/Watt performance anyways. If it was relying on single-precision then it made sense to go with Cypress-based products anyways.
I can understand them scrapping it if they based it on certain specifications that didn't hold true and what they allocated for it won't work, but then the implication would be a respecification of the project.
I would assume they established thermal and power limits for the system/location, and if Fermi doesn't hold to the promised spec they could go out in terms of thermal and power budget, but then you would expect that they might just scale down the project, or scrap it with an intent to re-do it some other way.
Just scrapping it when, as IDC says, there's nothing likely to come close (at least in double precision), seems pretty insane.
I would assume they established thermal and power limits for the system/location, and if Fermi doesn't hold to the promised spec they could go out in terms of thermal and power budget, but then you would expect that they might just scale down the project, or scrap it with an intent to re-do it some other way.
Just scrapping it when, as IDC says, there's nothing likely to come close (at least in double precision), seems pretty insane.
Maybe this it the semi part of accurate. Doesn't add up in my mind, but was quite funny. The hot rocks, etc leading to a can of whoop ass was well done.
Maybe they're waiting until the next gen of nvidia gpus. Perhaps fermi is going to be a short-lived product and nvidia's got something waiting in the wings, a version 2.0 if you will.
it seems NV doesn't like this website much say they decided to no longer respond to them any more says it all. just from the site name I'd say it's just a rumor may have some truth to it but probably mostly rumor.
It's entirely possible that Nvidia gave Oak Ridge power consumption numbers that they anticipated they would achieve with a respin. But upon testing the new silicon, they couldn't hit close enough to their guesstimate to satisfy Oak Ridge.
Zap
Elite Member
- Oct 13, 1999
- 22,377
- 2
- 81
Point is that unless they were off by a factor of 4 in the power-consumption estimates versus reality the Oak Ridge guys still aren't going to get any better flops/watt by going with an alternative.
Supercomputers are usually budgeted to accomplish some specific computing task, they aren't built for the sake of being built so when someone decides to not build one the decision isn't to save power the decision is really to not proceed with performing whatever calculations were previously believed to have been so badly needed that they warranted a $20+ million expenditure from the DOE.
Cancelling the contract with nvidia is unusual regardless the reasoning. Normally they would iterate on the specifications until a compromise could be attained. That the reason would be power-consumption for something that has never been a limiter in the past is double weird.
It would be like NASA deciding to cancel the Orion project (and the plans to go to the moon period) to send people back to the moon because the mpg fuel efficiency just didn't suit their fancy on second thought.
This guy should check his facts, if he read Charlie's 100% accurate about Fermi site he would know that nVidia doesn't even know what power consumption is going to be, see they don't have final silicon and won't for another couple weeks at the absolute earliest.
akugami
Diamond Member
- Feb 14, 2005
- 6,114
- 2,485
- 136
If the cancellation is true, there likely is another reason for it as most stated. Charlie will almost always spin some nVidia news negatively. Still, if this is true this is not good news for nVidia regardless of the reason for the cancellation.
akugami
Diamond Member
- Feb 14, 2005
- 6,114
- 2,485
- 136
And yet he's been correct enough times to warrant a look at the news coming from his web site. You also see fit to post/push rumors from other web sites that are really cut from the same cloth as Charlie's and no different from Charlie's web site.
Anyone can see that what you are doing is shooting the messenger. If you have a reason to believe what he posts is wrong then do so. We'd love to hear it.
Wreckage
Banned
- Jul 1, 2005
- 5,529
- 0
- 0
No he has not. He runs a satirical\humor site. It's like getting your news from Saturday Night Live.
Schmide
Diamond Member
- Mar 7, 2002
- 5,604
- 765
- 126
Or maybe more like the Daily Show? I'd put John Stewart above many in the main stream news business, cough cough small red furry K9. Considered satire would you also disparage him?
In the end you're just clouding the message by attacking the credibility of the messenger. Stick to the subject and prove or disprove it based on available information please.
True, but he also said TSMC cancelled their 32nm node and there hasn't been anything about that other than the speculation generated by Charlie's own article on the subject.
I'm not putting any stock in the Oak Ridge article until I see/read either Oak Ridge or Nvidia confirming it.
- Status
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 23K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 18K
-
-
-
Facebook
Twitter