Fujitsu Claims World's Fastest CPU

[guyladouche]

[citation][nom]eklipz330[/nom]i can't believe no1 asked this, but is that a fucking GOLD IHS?![/citation]

Probably not--probably just a polished copper IHS. Gold has a lower thermal conductivity than copper, so it would make no sense to use gold as the IHS.

It seems like these CPU's will be exhorbitantly expensive. Not because they're in any way "proprietary," but because if you jut look at the size of the CPU, it's unlikely Fujitsu is maximizing their cpu:wafer ratio. But if the goal is to make the fastest overall cpu, then perhaps they don't care about price.

On a separate note, I was wondering when cpu's would be made not to be smaller and maintain/slightly increase performance, and instead when the companies would say, "damn the cpu:wafer ratio, full performance ahead." Since obviously the more transistors you pack in a core, the faster the core will be. But in doing so, you lose volume production to make larger chips.

 

[ThisIsMe]

A lot of us already have a CUDA enabled graphics chip in our desktops that can run CUDA software at or near 1 teraflop.

 

[theangrygimp]

I think most kids here are missing a huge chunk of their education.

Your not kidding. Thanks for writing such a detailed explanation. I hope people will read it before shooting off with bias.

 

[ossie]

CISC is still around to run decades old, closed source applications from the m$-dos x86 era, so windblows maintains binary compatibility with ancient CPU designs. Internally, at the execution level, practically all modern processors are RISC like, x86 code being translated to micro-ops (pioneered by NexGen, and AMD).
One forgotten architecture is Transmeta's Crusoe, which translated x86 to VLIW (used also in the Itanic).
Practically all of the rest of the CPU world, especially the low power one, is RISC.

 

[Guest]

Oh for mother of... the numbers are already in G FL OP (Giga Floating Point Arithmetic Operations) of course this guy is faster than whatever Intel/AMD has. RISC/CISC line is almost non-existent to begin with in this day and age. But being a SPARC chip it is delegated to be a mini-computer server chip by inheritance, does it run Windows with DirectX natively or good emulation handles it if asked to do so? That should take care of "Does it run ___" questions.

 

[xnem3s1sx]

Doesn't the cell @180 GFLOPS destroy that?
Or perhaps that was discounted on the ground that it is not completely a CPU, as it shares far too many characteristics with a GPU.

 

[xnem3s1sx]

As for apple staying with the PPC, that would have been a horrible decision, there wasn't the same type of competitive market on those chips that drove developments like we have scene with intel and amd processors

 

[gomi]

It's a 8 core cpu which is why it is so massive.

 

[mapesdhs]

I'm surprised peoples' ideas on RISC vs. CISC are so out of date.
Modern x86 CPUs have RISC cores and have pinched lots of ideas
from traditional RISC CPUs in th last decade. Meanwhile, traditional
RISC CPUs have become a lot more complicated.

And the idea that RISC = no multimedia is just nonsense. It all
depends on what the chip design includes. Before switching to
IA64, SGI's original plan had been for a chip (SN2) that did
include media extensions (MDMX), of an order far superior to
the MMX in Intel chips of the day. SN2 was also going to include
vector extensions and other things.

Point is, there's nothing about a RISC design which makes it bad
for multimedia. If anything, the nature of media processing lends
itself rather well to RISC methodologies.

Starting with Pentium Pro and the Althon no Intel or AMD chip has
executed X86 instructions natively. They are "decoded" and then
fed to the instruction units. The decoder also fixes up alignment
issues etc. The decoded instructions are "RISC" instructions. The
reason why they don't openly publish them or allow you to compile
to them natively is they want to be able to change them in order
to increase performance.

The Pentium 4 used a different "internal" instruction set. With
the Core and Core2, they went back to the Pentium III since the
instructions were a lot cleaner.

So in the RISC vs. CISC wars, RISC really won, the victory was
just hidden. Neither Intel nor AMD were able to build a faster
native X86 chip.

Ian.

 

[guyladouche]

[citation][nom]thisisme[/nom]A lot of us already have a CUDA enabled graphics chip in our desktops that can run CUDA software at or near 1 teraflop.[/citation]
Except there's a big difference between what a tflop cuda-enabled gpu can do and a tflop cpu. For starters, the cpu can actually run and support an OS environment, which the gpu can't do at the moment. Plus, what you can do with the gpu is at the mercy of whether a program supports cuda-acceleration--support which is sad at best right now. So it's apples to oranges, I'm afraid.

 

[downix]

[citation][nom]xNEM3S1Sx[/nom]Doesn't the cell @180 GFLOPS destroy that?Or perhaps that was discounted on the ground that it is not completely a CPU, as it shares far too many characteristics with a GPU.[/citation]
According to IBM the fastest Cell for floating point (PowerXCell 8i varient) can reach a maximum of 102.4 GFLOPS. http://www.ppcnux.com/?q=node/7144

 

[sonofliberty08]

that means x86 are out of date , they should just eliminate it ... bye bye intel licensing

 

[downix]

[citation][nom]sonofliberty08[/nom]that means x86 are out of date , they should just eliminate it ... bye bye intel licensing[/citation]
You prepared to move past Windows? Windows and Intel are joined at the hip. Eliminate one, you eliminate the other. MSFT has tied their product too closely to Intel to migrate without severe pain, Intel cannot consolidate enough around any OS other than MSFT to keep it's own architecture monopoly intact. These two companies have painted themselves into the same corner, dependent on each other to survive.

 

[Jien Malti]

[citation][nom]TheFace[/nom]If that is an actual picture of the CPU, that thing is massive.Also, if anyone remarks "but can it run.." they should be banhammered.[/citation]

That's not the question to ask. I say... will it blend?