Who needs sandy bridge?

[SuicidaL]

anyone fancy running one of these

http://www.bbc.co.uk/news/technology-15758057

now if it doesn't bottleneck my 6870 and comes in a 775 package it could be a winner

[CAT-THE-FIFTH]

Looks like Intel has been canvassing the media sucessfully!!

Its a co-processor like the Tesla cards and their AMD equivalents. It is NOT a CPU. It is based on Larabee IIRC. The chip AFAIK even makes the 40NM Nvidia Fermi based GPUs look small despite being 22NM. On top of this the APIs which this uses will determine its success too and will also determine its real world performance (not theoretical).

The chip is still a prototype and older chips like the Nvidia GF110 GPU(GTX580) used in the Tesla M2050 already can produce 665GFLOPs DP. The RV870 GPU(HD5870) used in the AMD FirePro 3D V9800 already can do 544GLFOPs DP and the card was released last year. These are based on the older TSMC 40NM process too.

The next generation 28NM Nvidia GPUs and even the next AMD GPUs will probably exceed 1TF. These are both based on new architectures which will improve DP performance significantly.

[SuicidaL]

Looks like Intel has been canvassing the media sucessfully!!

Its a co-processor like the Tesla cards and their AMD equivalents. It is NOT a CPU. It is based on Larabee IIRC. The chip AFAIK even makes the 40NM Nvidia Fermi based GPUs look small despite being 22NM. On top of this the APIs which this uses will determine its success too and will also determine its real world performance (not theoretical).

The chip has barely entered production yet and older chips like the Nvidia GPU used in the Tesla M2050 already can produce 665GFLOPs DP. The AMD FirePro 3D V9800 already can do 544GLFOPs DP and was released last year. These are based on the older TSMC 40NM process too.

The next generation 28NM Nvidia GPUs and even the next AMD GPUs will probably exceed 1TF. These are both based on new architectures which will improve DP performance significantly.

i know it's a co-unit, i should had added a sarcasm smiley to my post really

[kalniel]

On top of this the APIs which this uses will determine its success too.

It's x86, it hardly needs fancy gubbins. Throw in Platform Computing support and it's almost plug and play in many datacentres:
http://www.xbitlabs.com/news/other/d...elerators.html

We've got a cluster of terabyte RAM machines at work that I'd love to see paired up with a MIC.

The next generation 28NM Nvidia GPUs and even the next AMD GPUs will probably exceed 1TF. These are both based on new architectures which will improve DP performance significantly.

Whereas these do.

[shaithis]

Return of the x87 chip?!?!?!?

[CAT-THE-FIFTH]

It's x86, it hardly needs fancy gubbins. Throw in Platform Computing support and it's almost plug and play in many datacentres:
http://www.xbitlabs.com/news/other/d...elerators.html


We've got a cluster of terabyte RAM machines at work that I'd love to see paired up with a MIC.

Yawn. Again,if it was so easy Intel would have launched 500GFLOP to 600GFLOP versions last year using 32NM production tech and it would have been competitive. You don't even know if it the 1TF figure is actually a sustainable figure or whether under real-world conditions it will be anywhere near that. Power consumption is another factor too as the prototypes chip from what I gather are huge. Until it has actually hit production you have no clue. Both the AMD and Nvidia are known quantities and the Nvidia units are widely used.

However, I have been following Knights Ferry and Knights Corner for a while. ATM despite Intel's bragging it will be the best thing since slice bread it seems not a huge number of companies are interested in it even now. AFAIK,there is one design win for it(summer 2011) and even then the statement is vague,ie,"it will be added when available".

The machine which will be using it will be operational in 2013:

http://www.physorg.com/news/2011-09-...igital-xd.html

This makes sense if the cards will be launched sometime in 2012. Last year, 2011 was meant to be the actual launch year for Knights Corner. If the cards were already delivered to customers this year it would be a bigger deal but it isn't though.

Whereas these do.

Whereas,the cards I mentioned already do and have been out for a while . There were protoypes of the latest Tesla cards out even in late 2010 and these can hit nearly 700GFLOPs a year ago.

The Intel card mentioned is a PROTOTYPE which has not even entered production.

Big flipping deal. They won't even been out until next year. Wait a second! BOTH Nvidia and AMD are working on newer generation GPUs like Keplar and GCN which will be out early next year in quantity.

But of course since they don't canvass the international media so much as Intel, every time they make a prototype, of course none of them exist!

[Biscuit]

"Traditional supercomputers were built by putting thousands of processors in a room but in the last few years there has been a shift toward graphic processors," said Martin Reynolds, a vice president at research firm Gartner.

"GPUs allow you to get results more quickly but will take longer to program so there is an interesting trade-off," he said.

Surely once the initial coding is done, you are off, so the trade off isn't really much of a trade off its more just an initial milestone which isn't in the way for Intel based HPCs.

Added to which, how many GPUs can you stick in a server compared to how many co-processors can you stick in a server?

[kalniel]

Surely once the initial coding is done, you are off, so the trade off isn't really much of a trade off its more just an initial milestone which isn't in the way for Intel based HPCs.

Depends on the what you're doing. We're always updating programs so creating additional development branches for non-x86 architectures adds way too much cost either for us or our collaborators (or we'd already have done it). But the programs that we do create are already designed to scale, and I think they'd need very little optimisation for this.

Added to which, how many GPUs can you stick in a server compared to how many co-processors can you stick in a server?

We don't know the thermal/power characteristics yet, but theoretically exactly the same if they both use PCI-E card factor. There's talk of KC using QPI as well which creates some further opportunities.

GPGPU is neat for a few tricks, but it's way too inflexible for our uses.

[watercooled]

Basically still Larabee which they're doing their best to make use of so it's not a complete waste of time and effort. Let's face it, you're not going to get 1TF worth of Sandy Bridge cores or anything close on these chips, they are going to be very cut back, simple cores and x86 doesn't tend to scale very well towards that end of things. Also, you're not going to be able to use the same programs you would use on a big chip if you hope to fully utilise what the chip has to offer. It seems it fits somewhere between RISC chips like SPARC Niagara and 'GPUs'. GPUs are so much more than graphics cards with a bodge to make them do GPGPU now, in fact the new architectures are designed pretty much the other way round. I'd be very surprised if these chips give anything like decent FLOPS/watt compared to RISC/GPU...

Edit: Oh and Tilera is another company to look up for some interesting reading IMO...

[bobfoc]

Guess you can only really use it with specialist applications specifically written for it.

Shame... imagine how it could run Battlefield 3

[Biscuit]

probably no better than a i5 2500K as its usually GPU limited

[watercooled]

Made me giggle that. xD

I'm probably more critical than most about this but I'm more of the opinion, if something's worth doing, it's worth doing properly the first time. I.e. I'd rather use a different architecture if it means better performance in the long run over keeping things as they are to make things easier in the short term. It's worked well multiple times in the past e.g. multi-core, unified shaders - even though programmers really disliked programming for them as it took more effort, it really paid off. I think the consoles played a big part in that happening as soon as it did, and it's a shame to think how much longer we'd have been with single-threaded games if that hadn't been the case.