Better than Intel: Fujitsu develops world’s fastest processor

[akhilles]

by Serkan Toto on May 14, 2009

Bad news for Intel for two days in a row. Fujitsu yesterday took the wraps off a new CPU made for supercomputers that can perform 128 billion computations per second, which is 2.5 times faster than the super processor of the current record holder Intel.

The prototype is fabricated using a 45-nanometer process and consists of a total of eight cores. Fujitsu says the new CPU, dubbed Venus, also consumes just one third of the power the Intel chip uses. The company expects the processor not only to be sold to makers of super computers but also to the industry. One possible area of usage is product development, i.e. in research for new drugs.

Fujitsu is now the first Japanese company in about 10 years to hold the record. Practical application of Venus is expected to take place “within several years”.

Better than Intel: Fujitsu develops world’s fastest processor

[brentpresley]

Show me benches !

[DaSickNinja]

This isn't another Cell type chip where they hype SIMDs as cores?

[Simul4ti0n]

It will be a niche product since they have no x86 license. Still cool tho!

[djgandy]

Look at the wafer. If that is an actual wafer and 300mm it is HUGE. I estimate at over 500mm2 per die. Looks like they are getting about 100 usable die minus defects.

[Šg‘å‰æ‘œ]�yPC Watch�z •xŽm’Ê�A�¢ŠE�Å‘¬‚ÌSPARC64 CPU�uSPARC64 VIIIfx�v



The package in the guys hand is huge too.


EDIT: If you look in the background too there is an actual wafer and it says 300mm2. I bet most the power is from a SIMD unit anyway. Once x86 goes AVX with 8 cores processors like this will look archaic.

[CapnBFG]

Show me folding benches !

Fixed

[orangegator]

Fujitsu is now the first Japanese company in about 10 years to hold the record. Practical application of Venus is expected to take place “within several years”.

Within only one year this thing would be obsolete, let alone several. And this is only a laboratory sample, not ready for production. I'm sure both Intel and AMD have new chips in the pipeline that will spank this thing. And they'll likely be in production sooner. Articles like these are just pure hype.

[arisythila]

They will probably be showing at Computex, We will check them out when we get there.

~Michael

[albundy2]

actually, i would be happy if this took off... three chip makers in the ring fighting for our money? i like that idea!

[beerandcandy]

its like 2 inches accross
show the bench marks prove it is better

[Cain]

Fastest at what exactly?

[tool_462]

Fastest at what exactly?

Oracle apps.

They says "fastest in the world" as in it can do the most calculations per second.

[Cain]

I get that but its not exactly a useful measurement. I can build a CPU specialise in anything and say its better then a more generalised CPU like intel/amd's are.

[tool_462]

I get that but its not exactly a useful measurement. I can build a CPU specialise in anything and say its better then a more generalised CPU like intel/amd's are.

Yeah...

There is a thing called marketing that isn't always providing facts at face value to the average consumer/news reporter

I'm sure this CPU is powerful, it just lacks x86 application support to make it usable

[djgandy]

It's a SPARC64 so it is not insignificant. However it measures about 22mm x 22mm, so an 8 core i7 will pretty much keep up with this. Also with respect the peak performance, they do not state whether the processor will do Fused Multiply Add (MUL+ADD in a single clock). If it does this doubles peak performance. Once x86 CPU's have fused multiply add their peak performance will double too. This puts a quad core i7 at approx 100 GFLOPS.

A paper I read about using GPU's to solve Linear Algebra by some Berkeley students already puts i7 and Core2 at ~100 GFLOPS. I'm not sure how they acheived that though

This CPU is really not that great considering it is not x86 and is in "other markets"

128 GFLOPS / 8 cores = 16 GFLOPS per core.
Running at 2 GHz and with a 256bit SIMD (8-wide) that is easily obtainable. As a processor geared towards computational computing you'd expect that. They don't need to focus so much on optimising single threaded performance.