WHEN IT RAINS, it pours for Intel, it seems, with Chipzilla receiving bad news not only on the legal front but also on the tech front as well this week, as Fujitsu has just unveiled a new CPU thought to be the world's fastest yet.

Image courtesy of PC Watch
The Japanese firm revealed its latest supercomputer CPU offering at "Fujitsu Forum 2009." It claimed the chip is capable of crunching through an astounding 128 Billion computations per second, 2.5 times faster than Intel's current best performer. This makes Fujitsu the first Japanese firm in a decade to snatch the lead in raw CPU performance.
The little beauty has aptly been dubbed “Venus,” a much sexier name than its rather more nerdy technical moniker “SPARC64 VIIIfx."
Fujitsu says its prototype still in development is fabbed on 45nm process technology and has eight cores plus an integrated memory controller spread across just two square centimetres.
Fujitsu, not content with blowing past Intel speed wise, also says its new chip boasts a power consumption advantage, consuming just a third of the juice Chipzilla's prize CPU guzzles.
Apparently it will take several years for Fujitsu to come up with practical applications for Venus, but Fujitsu says the CPU shows promise in the fields of pharmaceutical research, astronomy, weather prediction, scientific research and more.
The Japanese electronics giant also says it hasn't ruled out adapting its CPU from one used only by the supercomputer industry to one more suited to mainstream computing applications available to mass market consumers. µ
L'Inq
PC Watch (Japanese)
As any computer science student will tell you, speed parallelism.
You need to adapt existing algorithms, for parallel processing. You don't need to do anything with a faster CPU.
Speed wins out in most cases.
Example:
Double core, 3.0Ghz CPU isn't twice as fast as a single core 3.0Ghz (except in very rare cases).
6.0Ghz CPU _IS_ twice as fast as 3.0Ghz (all the time).
Now imagine they have x86 license!
This really doesn't mean much. Raw processing computer is nice but doesn't automatically translate into real performance.
Besides, the chip is still a prototype, so it's hardly the fastest anything.
Remember: Larrabee is expected have a raw processing power well above 1 TFLOPS, if not 2 TFLOPS.
When Rich Wargo Sees word: Fujitsu, Rich Has memories of Being Happy Camper.
On Sarralee side, heres art: http://www.dailytech.com/Intel+Visual+Computing+Institute+Opens+Will+Spur+Larrabee+Development/article15115.htm
Notice even if larrrabee works properly, its X86 chip Only. Take Money & Run. Larrabee is So Old, 50 Years Old, its steep in corruption & false promises.
As World Turns, So Do, Days Of Our Lives. Intel: Turn Larrabee Around & Get Going. Fugitsu, Making Product For only Themselves.
"All the time"? So a 3ghz P4 is faster than a 2ghz Core 2?... All the time? Just because it's MHZ? I think you need to go back to school.
I think you (purposely) missed the intent of Ashkante's statement...as a 6.0 GHz P4 IS always 2x as fast as a 3.0 GHz P4 whereas 2 P4s running in parallel at 3.0 GHz is in almost all cases NOT 2x as fast as a single 3.0 GHz P4. That was obviously the point he was making...
Is that an American billion (10^9) or a British billion (10^12)?
yeah, but can it play Crysis?
6GHz is not twice as fast as 3GHz for the same CPU.
There is more then just a CPU to make up for the speed.
However, if you take 2 (or more) AMD's Opterons, they scale very well compared the speed of a single CPU with double frequency.
It just depends how you scale.
The P4 or I7 are poor examples of parallelism, as Intel never scales well.
Please do some reading on the matter, as they are talking server stuff, not the silly desktop TOY-CPU's.
Intel is a joke in the (mid & high) server market :-)
Will it play Fallout 3?
yeah exactly. find a use for it. its not double precision X86, CELL has superior floating point performance and GPGPU is off the charts for parallel ops. ever heard of intel polaris? well go look that shit up lol.
Crysis? Way too short. Fallout 3? utter crap (so disappointed).
Any word on what frequency said 128GFLOPS are accomplished at?
I see we have another Battle of the JayWalk All Stars here. Read this link from Los Alamos National Laboratory on high performance computing.
http://www.lanl.gov/news/index.php/fuseaction/1663.article/d/200805/id/13277
This is from a group of people who each one individually knows more than any other commentator here will ever know. I quote:
"Transistors are still shrinking, and the number per microprocessor is still growing, but performance measured in arithmetic operations per second has flattened out since 2002.
"The biggest reason for the leveling off is the heat dissipation problem,” says Ken Koch, one of the leaders of the Roadrunner project. “With transistors at 65-nanometer sizes, the heating rate would increase 8 times whenever the clock speed was doubled, outstripping our ability to carry the heat away by standard air cooling.” It would be like running a car at high speed with no water in the radiator" . There are a lot of other factors that enter into how fast a cpu can perform non trivial operations 8k page files. Since we are at a point where the laws of thermodynamics limit how fast a core can operate other issues such as bandwidth and latency times account for almost all speed improvements since 2002. http://gcn.com/articles/2006/03/03/when-are-supercomputers-really-super.aspx
There is no better example of slow parallel processors being faster than ultrafast processors than the IBM Blue Gene series. Number 1 in linpack scores for 3 1/2 years. The pentium pro 233 was the last Intel chip to top Linpack. L series processors are 700mhz yet two or three of them routinely trounce any Intel chip including Hehalam in any benchmark that does more than reflect cpu speed. They also accomplish this with about 1/8-1/16th the power of the Nehalam. The Fujitsu chip is the production version of the SPARC64 chip developed by Sun as part of the Superpetaflop competition. It is about as main stream as the IBM P7 cpus.
I think what's really needed, and I'm specifically talking about gamers needs, would be a 65nm architecture with 32nm transistors. That's an over-simplification, but basically if you give 32nm transistors the same amount of wiggle-room as their 65nm counterparts, than they have more of an opportunity to cool down. It would be as if you had 64 fat people in a room on a huge chessboard and then switched them out with 64 skinny people on the same squares, which room would be warmer after a few hours? The skinny people room or the fat people version. Honestly, I don't know why AMD or Intel haven't built chips based around this basic concept.
Everybody knows how to build the fastest CPU on the planet.
Just ask the folks from the former the Alpha group.
It's not (just) a matter of how to do it, it's a matter whether there is a market that accepts the price point.
And whether you have the financial clout to survive to invest billions with a ROI in 5-7 years. Its extremely costly to creating such a beast and then market und support it successfully.
Nor Intel or AMD strive the build the fastest CPU they can,
simply because it does not make economical sense to them.
Intel and AMD build the fastest CPU for a market that they think is viable.
That's about it.
The money is not in the highest performance CPUs, its in the highest volume CPUs ...
The Carnot cycle is an idealization, since no real processors are reversible and all real physical processors involve some increase in entropy. In computer programming, the "maximal munch" principle is the rule that as much of the input as possible should be processed when creating some construct. Unique “stream computing” software that enables massive amounts of data to be analyzed in real time, delivering extremely fast, accurate insights to enable smarter business decision-making; transforming the average "duck" into a perpetual analytic.
Parser lexical analysis, preprocessing, semitic analysis, code generation and optimization; to these we must add parallel processing.
This will have to wait until later, since we are so invested in our present research on laying pipe at the University of Minnesota; and you may quote me on this, when I say: "I Carnot do everything."
Ok, tehy built a chip with no practical applications, so why compare it to Intel's 'offerings' and not to Intel's 80-core monster that also has no practical applications?
preproduction parts without actual application benchmarks are pretty meaningless.
Try figuring out how to compare the raw specs on an ati and nvidia graphics card? It's pretty tough because they are completely different designs, just like ps3 vs xbox 360, and powerpc vs x86 etc.
Show us a working operating system running a useful application on this system and an intel based one, then do benchmarks etc.
I'm not saying it won't be great, I'm saying everybody says their great :-). And x86, for all it's age and quirks, still beat powerpc and itanium in almost all markets.
"It claimed the chip is capable of crunching through an astounding 128 Billion computations per second"
Unfortunately, all of those operations happened to be NOP.
Not so, von Drashek. It's thoughts of DEC, PDP-11 and VAX-11 which makes me pine for "the good old days."
On the topic, this is a Sparc64 right? So it should be compatible with other processors in the Sparc64 family, so it should be easy enough for Sun.... oops, never mind. Given the takeover by Oracle, it's tough to predict what will happen to Solaris and any Sun-developed application development tools. I know Oracle has said it will keep all that going, but I've heard stuff from Oracle before that hasn't come to pass.
Poor Fujitsu, developing a processor for an architecture that is slowly dying or turning into a database engine, same difference.
After all Intel lost the memory business to Japan long ago, why should they not loose the processor business too.
Well, it would not surprise me at all.
It's weird that Fujitsu claims that it will take years to find an application for it. Solaris runs on Sparc and Linux probably, too. They are not the first Sparcs, so they could replace/upgrade/extend existing installations.
I would take one if they would sell them to consumers and are cheap enough. But that is probably a problem. And then there is still the T2 with 64 threads (not good with floating point, but nice for a lot of things).
by a factor of 20 ...
http://www.youtube.com/watch?v=We_PRtRfiNs
2 Tera to beat, 128Giga is funny ;-)
Intel's Polaris prototype can do 1 TFLOP at around 3 GHz (iirc). Raw compute power doesn't matter much. Anyone can do that.
@Bas
They're talking about RAW THEORETICAL compute power here. It DOES double with frequency. It may or may not double with the number of cores, depending on how parallelizable the code is. YOU should do some reading and thinking.
128Billion NOP/sec doesn't really count.
You need a CISC CPU to do something useful on each of those cycles..
history as already shown us that....
LOL
Good job Fujistu....you just made a processor that is still half as fast as Ibm, Sony, and Toshiba's Cell Processor that has been out for a few years now.
Good job.
As Everyone Who Know About Computers Deeply They Already Said That 3Ghz Single Core Pro. Are Always Better Than 2Ghz Dual Core
But About Fujitsu?
ill always say Wee Dont Need It.....
As Per combination of Prize & Perfomance Mine One Is Ultimate
Core 2 Quad 2.4Ghz Q6600 8MB L2 Catche
I've Buyed This Pro. At Just 10,000 INR(250$)
& I can Run Every Damn Thing On It's Like
Butter On Hot Pan.......
About Games! I've Played Every Game Like Crysis Warhead, Prototype, Far Cry2, Call Of Duty 5, Assasin Creed2, Gear Of War & All Of Them Runs Great On Highest Graphic Settings
Not A Single Application Ever Make My Processor Cross 55% of CPU Usage
So U Think We Need Any More Power???
I Think Its Much Affordable Power Processor Ever!!!!!