Thursday, May 14th 2009

Fujitsu Develops the Fastest CPU in the World

10 years after they last held this title, Fujitsu have now snatched the status from previous record holder Intel, by developing a CPU (Central Processing Unit) capable of 128 billion calculations per second. Codenamed Venus, the chip is said to be able to calculate 2.5 times faster than the previous fastest CPU and using miniaturization technology, Fujitsu have managed to double the number of central circuits integrated onto a chip measuring about 2cm from four to eight. Venus is also said to consume only a third of the electricity compared to current levels.
Tens of thousands of these new chips are set to be used in a next-generation supercomputer for an Institute of Physical and Chemical Science which will go into operation at the end of fiscal 2010. It has also been said that should the Venus come to be used in devices such as personal computers and digital electronic appliances, it could lead to the development of equipment such as portable simultaneous interpretation devices and automated driving devices for cars.

Source: Expreview
Add your own comment

29 Comments on Fujitsu Develops the Fastest CPU in the World

#2
Mussels
Moderprator
by: FordGT90Concept
Is it x86?
you forgot "Wprime plz" and "will it run crysis?"
Posted on Reply
#3
AlCabone
Which is the "previous fastest CPU" that they talk about?
Posted on Reply
#4
Mussels
Moderprator
by: AlCabone
Which is the "previous fastest CPU" that they talk about?
i googled it
the prototype with the code name 'Venus' (SPARC64 VIIIfx) has been manufactured using the 45nm technology and it is able to compute at 128 billion times per second (128GFLOPs), which is 2.5 times faster than the current record holder, a CPU made by Intel. The new CPU is also 3 times faster than the company's previous performance CPU, - the SPARC64 VII (quad-core, 65nm) - shipped in 2008.
so... its SPARC.
Posted on Reply
#5
WhiteLotus
I think he is asking what intel chip held the crown before this one came along
Posted on Reply
#6
DrPepper
The Doctor is in the house
I thought IBM held the title of worlds fastest CPU.
Posted on Reply
#7
btarunr
Editor & Senior Moderator
It is based on SPARC, although Fujitsu does hold an x86 license. The company used it rarely for embedded processors.
Posted on Reply
#8
toyo
from wikipedia:

"As of November 2008, only one of the world's top 500 fastest supercomputers is based on SPARC processors:
Rank #221: JAXA Japan, Fujitsu FX1, Quadcore SPARC64 VII 2.52 GHz, Infiniband DDR / 2008 Fujitsu , 20643 GFLOPS

This list compares unfavorably with other processor architectures, which make up a much larger portion of the top 500 list. The SPARC processor family had 88 of the top 500 systems in June 2002, but has since lost popularity to faster chips from IBM, Intel, and AMD."

So the "Crysis playing" CPUs are still top :P
Posted on Reply
#9
FordGT90Concept
"I go fast!1!11!1!"
by: Mussels
so... its SPARC.
That sucks. I figured it wouldn't be x86. x86 is flexible but not necessarily fast. :(
Posted on Reply
#10
lemonadesoda
New SPARC = 128 billion calculations per second
Intel i7 = 50 GFLOPS = 50 billion calculations per second

So indeed about twice as fast

BUT

Larrabee = 2 TFLOPS = 2000 billion calculations per second = at least 10x new SPARC

larrabee >> new SPARC

Very. very interesting link from intel: http://www.intel.com/support/processors/sb/cs-023143.htm
Posted on Reply
#11
MrAlex
by: lemonadesoda
New SPARC = 128 billion calculations per second
Intel i7 = 50 GFLOPS = 50 billion calculations per second

So indeed about twice as fast

BUT

Larrabee = 2 TFLOPS = 2000 billion calculations per second = at least 10x new SPARC

larrabee >> new SPARC

Very. very interesting link from intel: http://www.intel.com/support/processors/sb/cs-023143.htm
Holy Crap! And the Radeon HD 4890 has 1.36 TFLOPS of power!
OH, WAIT, Shame that Larrabee and the 4890 are both GPUs.
Posted on Reply
#12
Mussels
Moderprator
by: MrAlex
Holy Crap! And the Radeon HD 4890 has 1.36 TFLOPS of power!
OH, WAIT, Shame that Larrabee and the 4890 are both GPUs.
larrabee doesnt fit the definition of a GPU. its capable of GPU tasks, but also x86 tasks.
Posted on Reply
#14
Hysteria
Nice and fast but sod Crysis, forget Far Cry.... can it handle Solitaire?

:)
Posted on Reply
#15
mlee49
Great! Another confirmation that we are moving forward. Bye Bye Megahertz, you've had a great run for the last twenty years. Say hello to your replacement, Senior Gigahertz.

Senior Gigahertz is scheduled a 10 year term and will be promptly kicked out by Mrs. Tetra. Shes smoking hot!
Posted on Reply
#17

Here's another article that goes into a fair amount of detail.

The chip is only a prototype and won't ship for at least a year. Also, it will be water cooled. Don't know if that 50gflops figure was stock or oc'd. Therefore it will probably run at 4ghz.

So, it's not exactly vaporware, but it's not bankable yet either. Guess we'll have to see where Intel, IBM and AMD are in a year. :cool:
Posted on Edit | Reply
#18
xenos
by: twilyth
Here's another article that goes into a fair amount of detail.

The chip is only a prototype and won't ship for at least a year. Also, it will be water cooled. Don't know if that 50gflops figure was stock or oc'd. Therefore it will probably run at 4ghz.

So, it's not exactly vaporware, but it's not bankable yet either. Guess we'll have to see where Intel, IBM and AMD are in a year. :cool:
Hopefully in a better position now Intel has had its arse spanked. :rockout:
Posted on Reply
#19
lemonadesoda
Fujitsu spokesperson said made it 2.5 times faster than the best thing that Intel can put into the field
with
according to Fujitsu's presentations from last year, an eight-core Sparc V9-compatible processor with SIMD extensions aimed at boosting performance for parallel supercomputing workloads.
Oh look:

1./ They need 8 cores to do 2.5x the performance Intel Quad/i7. Well, Intel's 8 cores is on 2 chips, rather than one, but cheaply available, a la 2x Nehalem-EP. The performance of SPARC over intel is therefore 1.25x on a comparable 8 core system. Something tells me that on a $ basis, Intel will come out cheaper

2./ SIMD parallel supercomputing workloads; well that is very very similar to Larrabee, so by the end of the year, Intel will be snapping at SPARC as well as CUDA.
IMHO, it's an GPU
Ah, an honest opinion is welcome. But it is trumped by an educated opinion. And that is that we accept CUDA as a math/SIMD programming environment, not a "GPU" as you define. And Larrabee will be much more powerful/flexible than CUDA. So it will be "less" GPU than CUDA is. Background reading will reveal that Larrabee is actually CPU, and GPU functions are in software code, not hardware shaders. So Larrabee is actually CPU.

Don't let marketing and "leaks" fool you. Larrabee is as much supercomputing as it is GPU. But by targeting a consumer application (graphics) it will mean a superscalar CPU can be made at very accessible prices.

Expect Larrabee to kill off what is left of physx and potentially knock over both CUDA and clearspeed.
Posted on Reply
#21
FordGT90Concept
"I go fast!1!11!1!"
by: lemonadesoda
Ah, an honest opinion is welcome. But it is trumped by an educated opinion. And that is that we accept CUDA as a math/SIMD programming environment, not a "GPU" as you define. And Larrabee will be much more powerful/flexible than CUDA. So it will be "less" GPU than CUDA is. Background reading will reveal that Larrabee is actually CPU, and GPU functions are in software code, not hardware shaders. So Larrabee is actually CPU.
I highly doubt a single core processor with a Larrabee chip would cause Windows to run under ACPI multiprocessor rather than ACPI uniprocessor. The main feature of a CPU is it is on the highest tier of computing in a computer. All tasks start and end with it. The CPU divides the tasks among it's subprocessors. Larrabee is a subprocessor--a computer inside of a computer like graphic cards (higher end audio and RAID cards as well) are today. The only thing that makes it unique is that it runs on a trimmed down x86 instruction set rather than solely DirectX/OpenGL + Proprietary (CUDA/Stream). I suspect that Larrabee will receive a pretty significant performance penalty because the driver has to convert DirectX/OpenGL instructions to x86 processes but how much of a penalty that affords remains to be seen.

There are really only two advantages of Larrabee:
1) It is a superscaler design: Intel doesn't have to reinvent it every time they want to release a faster product, they just have to add more cores.
2) It uses a wildly successful instruction set: developers can program specifically for it with relative ease. It should see much better software support than CUDA/Stream which hardly any games use.
Posted on Reply
#22
MrMilli
Larrabee won't be running Windows or Linux for that matter even though it's using a subset of x86. It's basically a co-processor (like any gpu). In essence, Larrabee chips could be added to a Fujitsu SPARC server, so saying x86 servers have an advantage because of Larrabee is wrong. Actually, Japan is building a super computer that mixes Fujitsu SPARC chips with NEC/Hitachi vector processors. Knowing NEC's history in vector processors, you can be sure that these vector chips/systems will kick Larrabee's ass.

More intresting is 'barrier synchronization'. A technology they introduced with the SPARC VII that allows a four-core chip to look like a much faster single-core chip as far as compilers are concerned. Sweet!

Now that Oracle has commited to SPARC, the future looks bright.
Posted on Reply
#23
FordGT90Concept
"I go fast!1!11!1!"
by: MrMilli
More intresting is 'barrier synchronization'. A technology they introduced with the SPARC VII that allows a four-core chip to look like a much faster single-core chip as far as compilers are concerned. Sweet!
Intel is supposed to release a compiler soon that multithreads applications compiled on it. I really don't know how it works but it is certainly something to keep an eye on.
Posted on Reply
#24
Imsochobo
intels I7 isnt very cheap in xeon form.

Higher price doesnt matter when powerconsumtion vs performance is more important, due to cooling actually cost tons of money, and in fact, buying the right thing can make you save alot of money, and make the expensive buy really worth it.

Again, i dont know much about sparc, I dont think larabee gonna kick off straight away, but AMD can make GPGPU's like larabee ;) they got the x86 licence, and ati :)

I wonder when we will see a time when cpu only translate work for the gpu :P
Posted on Reply
#25
MrAlex
by: lemonadesoda
with



Oh look:

1./ They need 8 cores to do 2.5x the performance Intel Quad/i7. Well, Intel's 8 cores is on 2 chips, rather than one, but cheaply available, a la 2x Nehalem-EP. The performance of SPARC over intel is therefore 1.25x on a comparable 8 core system. Something tells me that on a $ basis, Intel will come out cheaper

2./ SIMD parallel supercomputing workloads; well that is very very similar to Larrabee, so by the end of the year, Intel will be snapping at SPARC as well as CUDA.


Ah, an honest opinion is welcome. But it is trumped by an educated opinion. And that is that we accept CUDA as a math/SIMD programming environment, not a "GPU" as you define. And Larrabee will be much more powerful/flexible than CUDA. So it will be "less" GPU than CUDA is. Background reading will reveal that Larrabee is actually CPU, and GPU functions are in software code, not hardware shaders. So Larrabee is actually CPU.

Don't let marketing and "leaks" fool you. Larrabee is as much supercomputing as it is GPU. But by targeting a consumer application (graphics) it will mean a superscalar CPU can be made at very accessible prices.

Expect Larrabee to kill off what is left of physx and potentially knock over both CUDA and clearspeed.
Ok, thanks for explaining to me then.
Posted on Reply
Add your own comment