Intel's Knights Corner: 50+ Core 22nm Co-processor

[eddieroolz]

I think the consumer processors are in the GFLOPS range, correct? If so, this is leaps and bounds ahead...

 

[sinfulpotato]

[citation][nom]eddieroolz[/nom]I think the consumer processors are in the GFLOPS range, correct? If so, this is leaps and bounds ahead...[/citation]

Consumer processors are not used in the same fashion. This is aimed to compete with CUDA and STREAM. Workstation applications. This won't make crysis run any faster.

 

[nottheking]

[citation][nom]g00ey[/nom]I take it that you in some way or other are working for AMD or even representing AMD.[/citation]
No, I am in no way associated with AMD. If I were working for them, I'd be an engineer and not a marketer, and be restricted by a bunch of NDAs so I wouldn't be saying anything here.

[citation][nom]soccerdocks[/nom]"the central, innermost, or most essential part of anything". In the Bulldozer architecture the most central item is the module.[/citation]
Actually, if we think back to pre-486 CPUs, we'd consider a "Core" to be an individual ALU unit. My main point is that the FPU cluster of a Bulldozer module can actually run two pipelines at once. Hence, I draw the line there considering that akin to much of the other similar route of merging or un-coring various parts when scaling up; L2/L3 cache was one of the first, along with the memory controller and I/O. The only real strong argument for considering a module a single core is that you can't cleanly disable half of it.

[citation][nom]fazers_on_stun[/nom]NV's top GPU does 515 peak GFLOPs: http://www.nvidia.com/object/perso [...] uting.htmlIIRC, sustained throughput is about 2/3rds of peak. Since the Knight's Corner does over 1 TFLOP sustained (according to the slide anyway), that would be 3X NV's best GPU on double precision..[/citation]
The 676 gigaFLOPS number was the peak for the 6970; AMD's GPUs have tended to be slighly better in theoretical peak math power... Though at least from what I've seen, the real-world maximum performance seen by GPGPUs in LINPACK was closer to 1/3 their theoretical peaks, not 2/3; desktop CPUs have ranged from around 70-90%, with non-x86s ones on the low end, and Intel's Xeons on the top.

[citation][nom]Th-z[/nom]All these numbers are posted by NVIDIA, we don't know if they are theoretical peak or sustained. If Intel's number is true, 1 TFLOPS in DP sustained is surely very impressive.[/citation]
For one, the numbers from nVidia are peak, not sustained. As for Intel's numbers, it LIKEWISE is the theoretical peak; it's not anything but until we see some benchmarks.

 

[g00ey]

[citation][nom]nottheking[/nom]No, I am in no way associated with AMD. If I were working for them, I'd be an engineer and not a marketer, and be restricted by a bunch of NDAs so I wouldn't be saying anything here.[/citation]
It's what you do that defines you, not who you say you are. I don't think any stealth marketer or astroturfer would admit on any forums that s/he is one.

 

[g00ey]

[citation][nom]nottheking[/nom]Actually, if we think back to pre-486 CPUs, we'd consider a "Core" to be an individual ALU unit.[/citation]
Now you're talking through your hat, there were no multi-core chips in the pre-486 era. There were MIMD systems but that's a horse of a different color.

 

[Guest]

Wow 5Tflops single precision (1 Tflop double precision) general purpose cpu I thought intel will take a long time to outpower the amd video cards that goes up to 4Tflops single precision ( 500Gflops Double Precision).

Just wow..

 

[g00ey]

Pardon my French but I think AMD core count scheme is f*ing FRAUD and someone's got to go to jail for this!!!

I think this is a part of a not so clever scheme to increase license income. A lot of software licenses in the server market are charged on a per-core basis so by doubling the core count they'll get twice as much.

 

[MasterCATZ]

my 2x ?

1) Price
2) can i use it for Trans-coding my Video Clips ...

 

[bloc97]

After that Intel created the "Megahertz Myth", they are now trying to make everyone think that it is better to have that much cores?

So that means if I have 5000000 Cores with 1 transistor on each core, it would be faster than 1 Core with 5000000 transistors?
Im going to buy a lot of transistors tomorrow... Uber multi-threading... 😀

 

[zankuto]

A computer within a computer? Inception! Hey dawg we heard you like to compute so we put a computer in your computer so you can compute while you compute.

 

[zankuto]

So its faster than over 9000 Pentium ii Xeons?

 

[lamorpa]

[citation][nom]saturnus[/nom]This is what is left of Intel's bid to compete in the GPU market, Larabee. A co-processor with very lack luster compute performance compared to todays GPU cards but really it can neither do GPU work not CPU work, so one has to wonder why Intel even bothers. Seems like a desperate, an failed, last attempt at trying to keep up with ARM.[/citation]
You would have better understanding of why Intel did this if you read an article about it, like the one 8" above your comment. All of your misunderstandings are covered there.

 

[berock212]

I've heard a lot of people comparing this to amd chips and graphics processors but I don't think we should be comparing this to anything other than coprocessors. This isn't meant to be used for running crysis and it's not the same as a desktop processor, so stop comparing it to gpu's and amd's 16 core cpu. This is for complex simulations and stuff like that.