pickyantivirus
New member
Huh, well the way it was explained in the article and around the internet, and the fact they've named it after their CPUs XEON I assumed it was a chip. My bad. Nice to see Intel confusing the s**t out of me.
Intel super processor Xeon Phil
- Thread starter pickyantivirus
- Start date
KING_OF_SAND
New member
It is a "co processor" just like Nvidia Tesla is.
Xeon Phi is just an Add-in card similar to NVIDIA's tesla range. It is a co-processor and it uses very basic x86 cores. Not the same kind you'll find in a Core i7 or a desktop/server orientated XEON.
The chip being used is basically what used to be referred to as a General-purpose computing on graphics processing units or GPGPU. This was the term used when CUDA, STREAM, OpenCL and Direct Compute first arrived on the scene but is being less used now.
So in short what is this thing good for? It's good for hugely parallel floating point operations. Data crunching on a massive scale. This card is not the sort of thing you stick in your PC (or server) and then your programs just see an extra 50 x86 cores. Your applications have to be built specifically for this card or be optimized to use OpenCL or Direct Compute so that it can use standard libraries in conjunction with Intels drivers. Exactly how the very popular CUDA works currently.
The benefit to Intels direction with using x86 cores is compatibility but don't be fooled it isn't the same x86 as on our desktops and it lacks complex instructions that have become standard over the years. It's just the bare minimum.
I think actually this chip is simply larrabee with Error-correcting code (ECC) included.
Now that I've deflated some of your balloons I'm going to fire one more bullet. The new NVIDIA K20 Kepler based Tesla card will feature 4 Teraflops of Single-Precision Floating-Point and over 1 Teraflop of Double-Precision Floating-Point. And the K20 is also ECC enabled. That is this Xeon Phi's real competition and I think Intel may actually find it difficult to break in to this market considering CUDA's huge lead.
The chip being used is basically what used to be referred to as a General-purpose computing on graphics processing units or GPGPU. This was the term used when CUDA, STREAM, OpenCL and Direct Compute first arrived on the scene but is being less used now.
So in short what is this thing good for? It's good for hugely parallel floating point operations. Data crunching on a massive scale. This card is not the sort of thing you stick in your PC (or server) and then your programs just see an extra 50 x86 cores. Your applications have to be built specifically for this card or be optimized to use OpenCL or Direct Compute so that it can use standard libraries in conjunction with Intels drivers. Exactly how the very popular CUDA works currently.
The benefit to Intels direction with using x86 cores is compatibility but don't be fooled it isn't the same x86 as on our desktops and it lacks complex instructions that have become standard over the years. It's just the bare minimum.
I think actually this chip is simply larrabee with Error-correcting code (ECC) included.
Now that I've deflated some of your balloons I'm going to fire one more bullet. The new NVIDIA K20 Kepler based Tesla card will feature 4 Teraflops of Single-Precision Floating-Point and over 1 Teraflop of Double-Precision Floating-Point. And the K20 is also ECC enabled. That is this Xeon Phi's real competition and I think Intel may actually find it difficult to break in to this market considering CUDA's huge lead.
pickyantivirus
New member
I understand now I've seen the card
Thanks for that though, very insightful.
It's still good to see Intel doing this kind of thing and progressing towards becoming a more complete company. I'm just hoping they don't do an AMD and fail miserably. But then AMD might come back and bite Intel when APUs are more suited to the market (won't be long) as more people need discrete graphics.
At the moment AMD support OpenCL and Direct Compute and they did also support STREAM on their cards for a while which was their own proprietary compute platform. But I don't think their heart is really in it because they are probably afraid that add-in cards will eat up their Opteron market. They don't want a company being able to buy a consumer orientated motherboard and processor for sub $150 and then sticking $3,000 worth of add-in cards in to it. I still think actually that AMD doesn't see their ATi purchase as part of themselves, they see them still somewhat as a competitor and they aren't willing to give their graphics devision all the clout they need to deliver a true enterprise ready GPU like NVIDIA and now Intel have done.
I hope AMD wakes up quickly so they don't miss this opportunity. Their Fire GL cards which compete with NVIDIA's Quadro range have been floundering in my opinion and the only good thing they've done when it comes to their GPU division is open the drivers for open source development. That will help them kick NVIDIA in the balls when it comes to the enterprise but again the hardware needs to meet the software and they as yet have no competitor to the Tesla or Xeon Phi.
I hope AMD wakes up quickly so they don't miss this opportunity. Their Fire GL cards which compete with NVIDIA's Quadro range have been floundering in my opinion and the only good thing they've done when it comes to their GPU division is open the drivers for open source development. That will help them kick NVIDIA in the balls when it comes to the enterprise but again the hardware needs to meet the software and they as yet have no competitor to the Tesla or Xeon Phi.
konfyoused
New member
Not to bring up a thread from the grave, but this really reminds me of the old Amiga arcitecture where you could throw an accelerator card in the system.
Goldskis45
New member
Propose thread merge or link? http://forum.overclock3d.net/showthread.php?t=45523
Bang
New member
Yeah i thought so too when I first read your post. If youre anything like me (and I know you are because you dig computers) I know you would jump on the chance to play some Pong on a 50 core CPU, just so you could say you hit 9,340,607,931,544 FPS 
Your 3dMark Vantage score would be off the scale.
Your 3dMark Vantage score would be off the scale.
SimonB
New member
well yea for sever porpuses this is amazing news especialy if this is 50 cores not 25 cores 50 treads + if this will work in servers capable of holding 2 cpus like this then this will be uber kill =DD 100cores in 1 server anyway for consumers i doubt they gonna use 50 cores even in 10 years ... why ? ... simple even now most of games and programs cant be used by 6 or 8 cores not even speaking about 12cores or 24treads in 2cpu case of intels cpu .... it would be a waste making like 50 cores 2ghz instead of making something like new arc for something like 2-4 cores of 6-8 ghz stable for something like 90-125w =DD anyway i doubt even that will hapen in 10 years but who knews whats hapen tomorow 
Goldskis45
New member
You could put 10 cards on this
http://www.supermicro.com/products/motherboard/Xeon7000/7500/X8OBN-F.cfm
and they would all run at TRUE 8x link...
Got to play with one of these boards about a year ago... was "bitchin"
Last edited:
SimonB
New member
i doubt even enthusiast would need 8 cpus with 50cores 400 cores in server =D not sure if its needed for home server or even small company really doubt that anyway probably someone need it well i don't =D my main rig is all i need atm i watch tv coz my internet sup provides me with tv signal and gives me free program hum sound is 5.1 home cinema system only thing i miss now another 2 monitors i sold =D but anyway i dont go deep into servers all i have ever done was linegae2 server =DD
Goldskis45
New member
Obviously nobody on here would have a rational reason for getting one. Just pointing out what the possibilities are.
Princeatom
New member
Until you show me proof I'll not believe a word you say.
Goldskis45
New member
(The supermicro) not the Xeon...
