backtop


Print 35 comment(s) - last by ThisSpaceForRe.. on Jan 6 at 4:48 PM

A quick look at the program outline for the International Solid State Circuits Conference reveals Power6 chips faster than 5GHz, Cell chips faster than 6GHz and 80-core Intel processors

This article was first published at HWUpgrade.com.

Many believe that the race for GHz ended with the Pentium 4 architecture, but you wouldn’t be able to tell by looking at the speed of these two upcoming IBM processors. According to the agenda of the International Solid State Circuits conference, IBM’s Power6 Processor will operate at speeds of over 5GHz in high-performance applications. The second-generation of the Cell Broadband Engine processor, which was co-developed by IBM, Sony and Toshiba, will run at 6GHz.  More details of Intel's "teraflop-on-a-chip" network processor will also be revealed.

The ISSCC2007 advanced program (PDF) claims IBM Power6 server processors are expected to begin shipping in 2007. Although it has been widely believed that the processors will run at a clock frequency between 4GHz and 5GHz, information provided by IBM for the ISSCC agenda states otherwise. The processor will run at over 5GHz in “high-performance applications” while also running at under 100 watts in “power sensitive applications.”’

IBM states that the Power6 offers “ultra-high frequency operation, aggressive power reduction, a highly scalable memory subsystem, and mainframe-like reliability, availability, and serviceability.” The dual-core 341 square millimeter processor features 700 million transistors. It is constructed using a 65-nanometer manufacturing process.

The ISSCC program also revealed technical details of the second generation IBM, Sony and Toshiba Cell Broadband Engine.  The first generation of the chip, which is currently used in the Sony PlayStation 3, runs at frequencies up to 3.2GHz. The second generation chip, on the other hand, will receive a frequency boost of nearly 3GHz and have an operating frequency of 6GHz at 1.3V. In addition, it will be constructed using 65nm CMOS SOI technology and will feature a dual power supply helping increase memory performance.

Despite IBM’s evident advancements in clock speed, Intel will still be holding firm to its multi-core approach to performance. According to the program schedule for ISSCC, Intel will reveal more details regarding its 80-core Tera-scale processor which can run 1.28 trillion floating-point operations per second.

Described as a “network-on-chip architecture,” the 225 square millimeter chip has 80 cores, each operating at 4GHz. The die is built using a 65nm process and is able to “achieve a peak performance of 1.0TFLOPS at 1V while dissipating 98W.” Currently, the processor is not able to run conventional applications for Intel chips.

Even though the chip is currently little more than a prototype, Intel CEO Paul Otellini claims it will be available within five years. The processor was first announced last September at the Intel Developer Forum.

The International Solid-State Circuits Conference will be taking place February 11-15 at the San Francisco Marriott Hotel. Other notable appearances will be given by Sun Microsystems, who will be discussing its Niagara2 processors, and AMD, who will be talking about its quad-core Barcelona processors expected to ship mid-2007.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Power6
By darkfoon on 1/4/2007 8:08:27 PM , Rating: 2
I am curious what the floating point performance of the new power6 architecture is like.

I remember thinking when Apple announced their switch to Intel CPUs that it was a bad move because Apples are often used in floating-point intensive applications: video editing/rendering, graphic design, audio, etc.

The power5 has a better floating-point unit than anything Intel has made. SSE has always been a little gimped, like it is an after-thought. Altivec, on the other hand, is still better than SSE3.

Too bad Apple switched, because the power6 sounds like it might have the power characteristics that Steve jobs was looking for. Of course, the ability to run Windows on a Mac has probably converted quite a few people who would otherwise have stayed with their PC, and is probably worth more money to Apple than the better vector performance offered by the Power architecture.




RE: Power6
By Meaker10 on 1/4/2007 8:42:53 PM , Rating: 2
Lol, but did you read that even the chips meant for power critical applications are running "under 100 watts" which means they will be close to that mark, so I dread to think what the high performance one will eat :/


RE: Power6
By Felofasofa on 1/4/2007 8:46:08 PM , Rating: 2
I have always thought it odd how Apple and IBM espoused the virtues of RISC over CISC, then they go and add a whole bunch of new instructions to their chips, seems to negate the whole philosophy? You describe SSE3 as an after-thought and Altivec isn't?

As a 3D artist, if that 80 core chip could run render farms, it would be very useful.


RE: Power6
By psychobriggsy on 1/5/2007 10:16:04 AM , Rating: 2
RISC isn't about reducing the number of instructions. It is about reducing the instruction set complexity. Think of fixed instruction lengths (i.e., 32-bits), only a few instruction formats, orthogonal instructions, and so on. Altivec does fit within this definition within the POWER instruction set.

On the other hand, dedicated function coprocessors (e.g., TCP/IP offloading and so on) aren't so RISCy :)


RE: Power6
By masher2 (blog) on 1/5/2007 10:23:37 AM , Rating: 2
> "RISC isn't about reducing the number of instructions. It is about reducing the instruction set complexity."

Err, this just isn't true at all. There are a few design goals of RISC, but they all center around paring the number of instructions down to the simplest, most commonly used set. This allows (in theory, at least) for designers to increase the speed of those instructions, and to devote more transistors to registers and cache, as opposed to core logic.


RE: Power6
By giantpandaman2 on 1/5/2007 1:27:06 AM , Rating: 3
Apple needed to change to Intel because IBM couldn't create a good mobile chip. Apple caters primarily to the consumer market...which, by far, prefers buying notebooks now.

Hence why Apple switched to Intel.

Pretty simple if you ask me.


RE: Power6
By NullSubroutine on 1/5/2007 7:40:41 AM , Rating: 2
exactly


RE: Power6
By kkevin6154 on 1/5/2007 8:13:53 AM , Rating: 2
The switch to Intel converted me to Apple.


RE: Power6
By crap on 1/5/2007 7:56:47 PM , Rating: 2
quote:
I am curious what the floating point performance of the new power6 architecture is like.

I remember thinking when Apple announced their switch to Intel CPUs that it was a bad move because Apples are often used in floating-point intensive applications: video editing/rendering, graphic design, audio, etc.

The power5 has a better floating-point unit than anything Intel has made. SSE has always been a little gimped, like it is an after-thought. Altivec, on the other hand, is still better than SSE3.


Power5 does not have Altivec! It has good floating point performance because it can issue 2 floating point instructions per cycle, it has floating multiply accumulate, and plenty of bandwidth. So Power5 is able to do 4 floating point operations per cycle.

quote:

Too bad Apple switched, because the power6 sounds like it might have the power characteristics that Steve jobs was looking for. Of course, the ability to run Windows on a Mac has probably converted quite a few people who would otherwise have stayed with their PC, and is probably worth more money to Apple than the better vector performance offered by the Power architecture.


Too bad you're stuck in the past.


c2d
By thepinkpanther on 1/5/2007 7:13:05 AM , Rating: 2
Comdrpopnfresh: have you ever seen a game test where a 3.7 ghz pentium beats a e6800 in oblivion which runs dual core..... never.

I do follow you in the sence that when a program is just written for single core than you dont feel more than the same speed. I would argue my e6600 is just as fast as the 3.7 due to better chip, lower nm etc. But when the gamefirms etc start to make games with dual-quad etc games there is just no one on earth that would even think on single core processors.

Just imagine if we get games that are optimised to deliver the workloads to different cores. Like EA when making a game distribute the sound to core 1, the background graphics to core 2, the character animation to core 3 etc etc. Its gonna be INSANE fast.

The biggest obstacle is getting programs and games that are written to take advantage of multicore systems. I also believe in the future the programs can themselves detech your system and take advantage of it.

Single core machines has looked great and beat the multicores but its over with modern games and soon even the most die hard fans of the old single core machines is gone too.

The biggest fear is how much power does each core use. The current use of watts is not good enough. imagine that 80 core system with 75 watts per core..... 6000 watts....lol

But perhaps you can shot cores down in real time... I need only 10 for this job. so the 70 other is sleeping so to say and if intel gets it down to a third in power use its still around 2kw but its possible with better and better psu´s etc.






RE: c2d
By masher2 (blog) on 1/5/2007 7:47:08 AM , Rating: 3
> "if intel gets it down to a third in power use its still around 2kw but its possible with better and better psu´s etc."

No PSU in the US is going to supply 2kW unless its hardwired to mains. There's a 15A limit on plugs and receptacles, which (assuming an 80% efficient PSU) limits you to around 1300W.


RE: c2d
By MrBungle123 on 1/5/2007 2:18:13 PM , Rating: 2
that's not true, there are 20A 120V receptacles available. They have a sort of sideways T for the hot side and are compatable with standard 2 and 3 prong residential electrical equipment.


RE: c2d
By masher2 (blog) on 1/5/2007 2:29:05 PM , Rating: 2
> "there are 20A 120V receptacles available..."

Sure, and there are also 30A and 50A ones. However, the standard NEMA 5-15 plug and receptacle (found on 99.9% of all 125VAC US households and equipment) is limited to 15 amps. So as I said, you're not going to see desktop computers utilizing more than 15A anytime in the next couple decades. No one is going to rewire their home simply to be able to buy a faster PC.

Servers, now, are a different story, many of which already ship with power needs far above 15A, and thus require special wiring to utilize.


RE: c2d
By Comdrpopnfresh on 1/5/2007 4:28:49 PM , Rating: 2
i agree with you almost completely. But what if the concept of what a 'modern' game is were to be changed? What if it wasn't the job of developers to make their software adapt to efficiently run dualcore?
IMO, it would be far more efficient, power, and adaptive to make a multicore chip that can (ON THE HARDWARE LEVEL) teem processors where needed to , and turn off the others like you said. This would allow for sheer speed in single processes. The same chip would be able to teem sets of a few processors to make less, faster ones- where some multitasking is needed, but not on an 80-core scale. Or it could flat-out run all the processors at default for things like rendering- everyone would win here.
The processor would adapt to meet the needs of the task, instead of software and users having to adapt to hardware, which is almost always obsolete as soon as it is sold. Considering the fact that revisions to hardware are so frequent, is it not to much to ask that the hardware adapt to the market instead of the other way around?


225mm chip!
By dice1111 on 1/5/2007 9:23:32 AM , Rating: 2
Granted it's 80 core's but thats huge!!! Imagine the heatsink required for that thing? Square out a peice of A4 paper and it's just a little bigger then that. How big would the motherboard be? I can't even picture it.




RE: 225mm chip!
By masher2 (blog) on 1/5/2007 9:25:31 AM , Rating: 3
Err, 225 sq. millimeters is not 225 millimeters squared. This is a chip about 1.5 cm on a side.


RE: 225mm chip!
By dice1111 on 1/5/2007 10:45:51 AM , Rating: 2
Ah, my bad. Needed to reread that. Thank goodness then.


YAY!
By S3anister on 1/4/2007 11:46:10 PM , Rating: 2
Now I can finally run a computer at 5 or 6GHz without having to use liquid nitrogen!




RE: YAY!
By averaesaveraesky on 1/5/2007 12:45:29 AM , Rating: 3
if you read the fine print very carefully it says "phase change cooling required" ;)


I stand vindicated
By masher2 (blog) on 1/5/2007 7:35:38 AM , Rating: 1
Several years ago, I predicted Intel chips with "hundreds of cores" would be available by 2020. Last year I repeated the prediction, in my blog here at DT. People thought I was nuts...but it seems that, if anything, I was too conservative in my estimate.




My vote for the win goes to....
By Comdrpopnfresh on 1/4/07, Rating: -1
RE: My vote for the win goes to....
By Operandi on 1/4/2007 9:06:59 PM , Rating: 3
Dual & multi core may have limited benefits (gaming) now but that is going to be changing quickly. Just look at what Valve is doingp; every machine they purchase is said to be quad core minimum.

Multi threading for gaming isn't exactly a new concept either. In the console world the Sega Saturn was modified to a dual CPU configuration late in it's design to compete with the Playstation 1.


RE: My vote for the win goes to....
By Comdrpopnfresh on 1/5/2007 3:33:36 PM , Rating: 2
why should developers have to write multicore software? They had enough trouble meeting deadlines when they only had to work on one processor. I don't like the fact that computers are heading towards what IBM and Cray make. The greatest challenge with supercomputers (extreme parallel) is not the engineering or manufacturing of these machines- its the programing. We are using computers based on machinecode and assembly language that was not meant to run on multicore chips. It creates an inefficiency. Thats why even in multicore environments running multiple processors efficiency does not scale- a dual core does not provide 200% of the performance of a singlecore, nor does a quadcore result in a 200% increase in anything over a dual.
My point is that teeming is required to keep personal computing on track with its foundations, or simply finding better single-core processors.


By masher2 (blog) on 1/5/2007 9:14:36 PM , Rating: 2
> "Thats why even in multicore environments running multiple processors efficiency does not scale- a dual core does not provide 200% of the performance of a singlecore..."

Depends on the task. There are many applications that can provide nearly linear scaling up to dozens if not hundreds of cores...which is why servers and workstations. have had multiple cpus for a very long time.


RE: My vote for the win goes to....
By Url on 1/4/2007 9:08:28 PM , Rating: 2
Just because the Pentium 4 EE has a higher frequency speed, it isn't the faster processor. The C2D have slower clock speeds but they also execute more data per Hz. Another factor that make single core chips better than multi-core is the software. Vista will change most of that because it will be able to manage multi-core processors better than XP.


By Comdrpopnfresh on 1/5/2007 3:21:30 PM , Rating: 2
although C2D may be able to handle more data per clockcycle, all that data is not directly from a select few processes. I'm simply saying that our perception of 'speed' is being skewed away from what it has been since the conception of the PC. Is a 1000-core processor with each processor running at 50mhz going to be faster than a C2D? It would seems to have more floating-point power, and a total bandwidth/speed of 50ghz- but its not all used together- its simply parallel processing. By your reasoning the answer is yes. But with majority of programing not multithreaded, this theoretical processor can run lots of stuff, but a lot slower per process than a C2D- such is the case with mutlicore/dualcore vs single core.


By Comdrpopnfresh on 1/5/2007 3:58:53 PM , Rating: 2
vista can run more- which is attributed to you truth that it is optimized for the multicore track. But isn't it also true that all reviews of vista so far have also pointed out its slower speed across the board?


RE: My vote for the win goes to....
By Spartan Niner on 1/4/2007 11:36:22 PM , Rating: 4
... and analysts can be dead wrong. Games run faster on single cores with higher speeds? Little to no benefit for quad core? Preposterous.

Let's see, multiple cores are great for highly parallelized tasks... hmm, games (check), video editing (check), rendering (check), mathematical calculations (check), encryption (check), database applications (check)... and the list goes on. The only reason multiple cores aren't showing that great of a performance boost yet is because much of the software we as the consumer use hasn't been made SMP/multi-core aware yet. Once multi-threaded apps are mainstream we'll wonder how we ever did anything with a single core.

As for games, let's assume a game is programmed to be multi-core. A quad core 2.4GHz versus a single-core 2.8GHz. Your poor little single core that could is being forced to do all the calculations, including physics, positional updates, etc. Your quad core on the other hand can have a quad doing physics, a quad readying textures and such to feed to the graphics card, and the other two quads processing other game engine necessities. Much better overall.


RE: My vote for the win goes to....
By Dfere on 1/5/2007 11:17:02 AM , Rating: 2
You forgot p8rn. P8rn apps using multiple cores would greatly speed consumption.


By Comdrpopnfresh on 1/5/2007 3:56:03 PM , Rating: 2
lol- I really don't think this humor is good for moving away from the topic at hand...


By Hydrofirex on 1/5/2007 2:04:29 AM , Rating: 2
Yeah, I have to agree: Multi-core is the future. I don't think anyone is saying you can solve the problem by throwing a million slow processors at a problem. Everything I've read leads me to believe that the idea for CPU's is that you refine single-core performance while adding additional cores - recognizing that no matter which route you take you're going to see diminishing returns. These are the main two routes to upgrading PC speed, and I think both are going to be necessary to keep performance gains up.

I have to say going from single-core to dual-core i see a HUGE benefit. Maybe eight-cores is a scam, but two or four very fast cores will definitely be able to handle modern multi-tasking and thread intensive applications at throughputs way beyond a single ramped core. Didn't the P4 prove this?!

HfX


RE: My vote for the win goes to....
By Etern205 on 1/5/2007 9:34:59 AM , Rating: 3
Multi-core is useless?!

If one day, the lowest processors are dual cores and
they have decided to stop maufacturing single cores, then
what are you going to do?
Protest and demand they bring single core back?
What the whole world laugh at your a55 for being
ignorant moron who does not know how to use a computer.

If you don't see a benefit to multi-cores then what
it probably means is you don't know how to use it!
Why don't you go back to a 486 because you don't
even derserve to use a Celeron!

BTW: running CineBench on a single core takes
about 1min 50sec
on a dual core: 38sec!
As you can see, it reduced the time in half!


RE: My vote for the win goes to....
By Comdrpopnfresh on 1/5/2007 4:12:42 PM , Rating: 1
what about benchmarking utilities made before dualcores? The benchmarking utilities have been added to to show impressible increases in the score of multicore processors! Didn't you ever read animal farm? Right now people with your view, baaahhing away whats been fed to them are playing the role of meat on wheels.....have fun at the glue factory.
I agree that dualcore is amazingly useful. But for the most part, it should stop there. The reviews of quadcore show little benefit unless you plan on rendering 12 things at once. The processors no longer fit the average user- they are straying away, towards the business market of servers which is a cash-cow for chip manufacturers. This is also the reason why intel switched to a unified architecture- gear towards servers, downgrade for users, and finally isolate problemed processors for value binning. This is how its working, not the consumer-centralized tier we've been told they are following.

Case in point:
The development of multiple quadcore Xenons for the server market, which will lead to quads for consumers thinking a corporation's computing needs are matching to their own, and those that have defective cache or cores will be binned to the value sector.

Dual cores have been around for a very short time in the scheme of things, and already fit the multitasking needs of majority of users. Why are companies continuing to plan a ramp up in the number of cores when the needs have been met?
Dual cores are wonderful devices, but multicore is not for the average user.


By ThisSpaceForRent on 1/6/2007 4:48:40 PM , Rating: 2
Without an installed hardware base, software developers have no motivation to develop multi-threaded applications. While multi-threaded apps are harder to develop, like any process this will improve as the industry moves towards programming those applications.

Pricewise the cost for the extra cores is negligble. Neither Intel or AMD wants to be more expensive than the other in their mainstream offerings, unless of course you're offering something the competition can't. Once they stop making single core processors this will also open up extra production capacity for multi core processors dropping the price further.

While I think the amount of system memory plays the most important role in determining the "speed" of any system. You still can't argue with having the processor power to crunch data, or multitask. I mean after all you can have all the RAM in the world, but once you reach 100% processor usage you're stuck waiting.


"This is from the DailyTech.com. It's a science website." -- Rush Limbaugh

Related Articles
AMD 2007 Server Roadmap
December 14, 2006, 9:47 AM
Intel Shows Off Teraflop-on-a-Chip
September 26, 2006, 2:49 PM













botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki