http://www.brightsideofnews.com/news/2012/6/18/intel-announces-xeon-phi2c-50-core-22nm-coprocessor.aspx
 

Today at the International Supercomputing Conference (ISC) 2012 in Hamburg, Germany, Intel announced their Knights Corner coprocessor, the first member of their XEON Phi family. Knights Corner is the next generation of their multi-core, parallel processing coprocessor for the HPC (High Performance Computing) race to be a contender in the TOP500 Supercomputing Sites.
 

don't let linuxrouter see this ... they want to make a folding farm with Z77 M-ITX Boards ... and if this comes in a decent price the M-ITX Form Factor with this in the PCI-E Slot with make some amazing folding/crunching numbers in a very small ammount space

Still slower than the latest AMD and Nvidia GPGPUs..
 
Intel should really leave the whole making GPU business to the pros.

or how bout this is their first real attempt at it while the other two have had decades to work on things? give them a chance 

emepror
or how bout this is their first real attempt at it while the other two have had decades to work on things? give them a chance

This isn't their first real attempt.
 
Sony's Playstation 4 was going to be full Intel.
Intel CPU + Intel Larrabee.  I sure you heard the rumors about it being full AMD enough said.
 
If you want a working GPGPU with 1.5-2 TFlops of DP you will go with AMD or Nvidia.  Instead of using a GPGPU that was supposed to launch all the way back in 2009.

If Intel did their development right, Im hoping to see 3D (CGI) Apps and render apps take full advantage of these, right out the gate?
 
 

Wouldn't the up coming NVidia K20 Tesla card using the GK110 chip also fit in an M-ITX form factor? It would also be faster, is the specs are to be believed.
 
If I won the lotery and I rather dump a few grand into Nvidia's cards over intel's for the reason above, Nvidia has more experience with stuff like this

Similar to how AMD doesn't need to have faster x86 CPU's to compete with Intel, Intel doesn't need to have equal performing GPGPU's with AMD or nVidia to compete.  The only requirement is to exist as an alternative, and thus stealing sales from either AMD or nVidia.  Intel primary advantage is their share of the market with particular focus on their x86 infrastructure.
 
Says so in the article:
 

Target market for XEON Phi family

Those folks have old X86 apps and they were interested in XEON Phi and Knights Corner because they won't have to re-write their old stuff into GPGPU (general purpose Graphic Processor Unit) languages like Nvidia's CUDA or open-source OpenGL.

 
In otherwords, let's say Xeon PHI is 5x slower than an nVidia or AMD competitor, it doesn't matter; Intel will win the market by sheer volume.

Nvidia's CUDA uses x86
AMD's OpenCL uses x86
 
So, I don't get your point Lehpron, HPC is price/performance/power
Those three align only with AMD and Nvida.

Power? What's this intel chip rated for compared to similar offerings from nvidia and amd?

seronx


Nvidia's CUDA uses x86
AMD's OpenCL uses x86

So, I don't get your point Lehpron, HPC is price/performance/power
Those three align only with AMD and Nvida.

CUDA and OpenCL are higher level programming languages that use x86 instruction sets. Which means the software must be written to use CUDA or OpenCL which the Intel coprocessor works natively in x86 so software programs just see it as another x86 processor like a CPU.

The GK110 should have 1728 GigaFLOPS.
The Tahiti XT (FirePro W9000) should have 1024 GigaFLOPS.
The Xeon Phi will have over 1 TeraFLOPS.

So the Xeon Phi will most likely beat what AMD offers. It also has a chance of beating what nVidia offers, purely on FP64 performance alone. I don't know how it performs in FP32, although FP32 is not where the big scientific developments are going to happen. Thus, Tesla and GeForce perform the same in FP32. Although, assuming the Xeon Phi just pushed 1.5 TeraFLOPS, making it less powerful than the GK110, the simplicity of programming offered by the Xeon Phi will alone make it a tempting buy. Not saying Intel is going to win this round. Just saying all three companies will have their strong points, meaning 3-way competition, meaning price drops and aggressive developments. Price drops won't affect gamers, since the GeForce line uses a different pricing system, but the aggressive developments will affect us.

Aggressor Prime
The Tahiti XT (FirePro W9000) should have 1024 GigaFLOPS.

That isn't the card to compare it to if you are doing Double Precision workloads.
 
Tahiti XT can have 1:2, 1:4, 1:8, 1:16 configurations for DP:SP.
925 MHz(XT1): 1,894.4 GFlops/ 1.89 TFlops
1000 MHz(XT2, Enterprise): 2048 GFlops/ 2.05 TFlops
1175 MHz(XT2, Consumer, Uncrippled): 2,406.4 GFlops/ 2.41 TFlops
 
GK110 will most likely have a turbo boost that disables the SP units if the workload is purely in need for DP power.
I'm going to assume it will basically operate like the 670/680.
GK110 ~2.0-2.5 Teraflops
 
Then add that these processors are around ~600-700mm² while the AMD solution 352 mm² with the 450-520 mm² biggy coming next year.  Nivida K20/GK110 520-550mm² so not only is Xeon Phi more expensive to make it is on a process made for small processors not big.

It will be interesting to see how Intel prices these and whether they can drum up enough developer support. If the price is right and it works with vray or Maxwell, I might just have to give one a try.

Can someone say -bigbeta in an mitx form factor? :-)

Well, a lot of supercomputer companies are jumping over to Xeon Phi for the supposed easier to program and get all your TeraFLOPs out of the processor approach.

http://www.extremetech.co...away-from-nvidia-tesla

"According to VR-Zone, Xeon Phi is apparently so desirable that it has replaced Tesla as the coprocessor of choice in many upcoming 100-petaflops supercomputers, due for completion in the next few years. The same sources told VR-Zone that porting code to Intel’s MIC architecture took days; while Nvidia’s CUDA took months."

So while nVidia may have faster cards, it takes smarter engineers to find their full potential; and smarter engineers cost money too.

Aggressor Prime
So while nVidia may have faster cards, it takes smarter engineers to find their full potential; and smarter engineers cost money too.

False, Kepler is easier to program for...
 
No one knows how to really program for the modified i86/i96 instruction set in Xeon Phi.  Xeon Phi isn't x86.

Do I see Big Adv for folding?

^
 
This is what I'd be more interested in. Tesla or Xeon Phi for F@H farms. It would be hard to say until the K20 Tesla card is released as well as the equivalent Intel Xeon Phi card to compare their PPD scores.
 
Anyone got the spare cash here to buy both and test them? Not me XD

The thing with Big Adv is they are requiring machines witrh 16 cores to process them.  If these are really cards with 50+ X86 cores I would think these would be Big Adv [link=mailto:F@H]F@H[/link] monsters, even if they are 1ghz cores just the number of them would smash some points I would think.  Some one would have to weigh the price differnce compared to a 16 core/thread system (SR-2, or other 5520 and newer 2011 socket and AMD systems) and what this card will pice for.  I will say you would be better off with a full system if this card comes out to be $4k+ and you are just wanting to get into the Big Adv folding.

Btw, what a weird number 50 is for computer tech. Not 48, not 64, but 50? Curious how it turned out to be so...

theGryphon

Btw, what a weird number 50 is for computer tech. Not 48, not 64, but 50? Curious how it turned out to be so...

The rules have changed; companies make up whatever number they want that nets the most gain with the least risk, they aren't going to keep doubling for the sake of binary math.
 
For example, Haswell-EP is projected to support up to 14-cores.  Brilliant, eh?