sweet i n eed that for wcg NOW
Printable View
sweet i n eed that for wcg NOW
Quote:
Originally Posted by nn_step
"Intel's prototype uses 80 floating-point cores, each running at 3.16GHz, Justin Rattner, Intel's chief technology officer, said in a speech following Otellini's address. In order to move data in between individual cores and into memory, the company plans to use an on-chip interconnect fabric and stacked SRAM (static RAM) chips attached directly to the bottom of the chip, he said."
http://news.zdnet.com/2100-9584_22-6119618.html
Ok considering you need 512Mb of ram per core for distributed computing, you would only need 32768Mb for that 64core ^ ^
But it looks damn sexy. Too bad that a quadcore is already overkill for most apps, because of bad multicore support :/
For distributed computing, such a beast would be definatly a nice addition.
Edit: According to that screenshot, that thing has 134gb of ram, so it must be a cluster of some sort, right? Dont think that much Ram would fit into 16 dimms.
and 64 CPUs networked together has to do with 1 80core CPU how?Quote:
Originally Posted by sxs112
LOL at that pic
Quote:
Originally Posted by Fr3ak
So even if these things somehow work out to sell for say under $3000, were still gonna be looking at around 5k for the ram *eye falls out*
Unless.....by the time these things come around to the consumer market, well all probably be using solid state hard drives as ram :p: :rolleyes:
interesting CPU although its not an X86 arch.
might prove useful for research though, as Itanium :rolleyes:
I think the real issue here, as NapalmV5 pointed out, is: How many threads are we talking about here?
I wonder how much ram you would need to run an 80 core setup ? 40GB ? :eek:
omg 128 gb of ram :DQuote:
Originally Posted by sxs112
80 cores for 1 teraflop...
i thought g80 or r600 can do 500gflop already?
The R580 does 360 http://www.peakstreaminc.com/referen...m_technote.pdfQuote:
Originally Posted by vitaminc
well the Xeon 5160 does 24 and Cell does 256.
and I couldn't find any numbers on the 8800GTX
But these 80 cores occuppies a smaller die size than the g80, so GFLOP/die size is better on the intel chip. Of course when it comes to practical use the G80 is far better, given theres probably very little code that the intel chip could run, let alone run optimally.
very true, it is far easier to optimize for a GPU than it is for many many CPUsQuote:
Originally Posted by onewingedangel
I don't see why that would nescecarrily be true, modern graphics card pretty much are like a massively multicore processor, and the code they run is very easily parallelizable. I see absolutely no reason why an 80 core CPU couldn't have every core working jsut fine when running graphics code like pixel shaders, or raytracing so long as they all have acess to that top level cache and huge memmory bandwidth in order to keep up with all the cores. This particular processor wouldn't work since it doesn't have high enough precision ALUs, but its just a proof of concept anyways, so thats to be expected.Quote:
Originally Posted by nn_step
It is similar to Itanium (Also uses VLIW (Very Low Instruction Word)) in concept. They seem to be simple, small and efficient Cores, but they kick a whole bunch of complexity of it to the compiler and programmer that must do all the code optimization job if they expect to get any amount of decent performance.
Just think on most lazy programmers that are developing commercial Software (Usually games) that becomes bloatware pretty fast demanding an awful amount of resources due to the lack of optimizations for rushing it to market. Definately not for general purpose.
8800GTX should have more, but ya, not sure how much more.Quote:
Originally Posted by nn_step
Not sure if die size are that much different?Quote:
Originally Posted by onewingedangel
80 "cores" or 128 "pixel shaders". whats the difference? they are just naming simple processing units differently, except NVIDIA/ATI has devoted a lot of embedded software support while Intel is "lagging" behind somewhat.
VLIW mean Very LONG Instruction Word. And "simple, small and efficient" describes everything Itanium is NOT.Quote:
Originally Posted by zir_blazer
I duno what to say. Great there really pushing. God know what they have done exactly, let alone how. seam fitting quad onto the LGA775 is pretty hard currently and die srink is needed so unless this ''cpu'' is 80 cores only and the catch is the ram or soemthing whom knows but scray the fact each core wants to know what the other is doing...
To make better to anyone to understand how they achieved that look in this video below.:)
Intel 80 cores Video Demonstration :slobber:
:rofl: :lmao: :lol: it isn't even close to VLIWQuote:
Originally Posted by zir_blazer
The Reasons why that is funny (all about Itanium)
1) it uses 592,000,000 transistors which is more than kentsfield
2) it is in the 130 watts evelope, once again worse than any x86 proc
3) it originally only produced 3.2 GFLOP of work and now it is only at 6.4 GFLOP, FX-53 cranked out 4.8 @ 2.4Ghz single core
4) VLIW is the hardest thing possible to program for, Even if you spend thousands of hours making 100% usage of the hardware; the next version of it wouldn't be able to run that same code & if it could, there is no way in hell that it would use the potential of the hardware.
5) VLIW is like a boxer with a Glass jaw, even though it looks good on paper; it goes down in the first round every time
That's my typo. Itanium logical Core (Ignore Cache) is smaller than that of an x86 Core. As it kicks instruction schedule responsability to the compiler, the logical Core itself is smaller and less complex than a similar x86 Core. It is efficient if you have a good team of programmers optimizing by hand an application code, that doesn't means that it is practical for everything though.Quote:
Originally Posted by BrownTown
If it is not VLIW, what Dailytech is saying that it is?Quote:
Originally Posted by nn_step
Again, when doing Transistors-based comparision, remove the Cache, as that 600M Transistors (Montecito I suppose?) monster got obviouly tons of Cache L3 (It will cripple its performance but that is not what I am comparing). Anandtech did an article with a in-depth architecture comparision of IA64 vs x86, but I wasn't able to find it after 30 minutes searching.Quote:
A major limitation of the current 80-core chip is that it is not based on the X86 architecture. Instead, it uses a 96-bit Very Long Instruction Word (VLIW) architecture, another architecture currently used in the Itanium server processors. A major hurdle that Intel hinted at will be moving from VLIW to X86 on its 80-core chip.
Intel did the same with this prototype, it is similar to Itanium as its trades lot of Core logic for simplicity of each them, but goes for massive thread parallelization instead of massive instruction parallelization like Itanium did.
is there a way to download it?Quote:
Originally Posted by Metroid
right click->save asQuote:
Originally Posted by -Sweeper_
tks :p:Quote:
Originally Posted by BrownTown