Cell, 4,3Ghz with 1,3 vcore

[impar]

Greetings!

"In a disclosure to the International Solid State Circuits Conference, IBM said it has made Cell chips that run at 4.6 gigahertz and operate at 1.3 volts."
in, http://www.siliconvalley.com/mld/silico ... 323259.htm

[envoid]

Didn't realize they ran that quick! But they don't run x86 code, right?

[powerarmour]

Didn't realize they ran that quick! But they don't run x86 code, right?

Right, but not that that isn't the reason anyway.

Think how the P4 can scale further due to it's low IPC and longer pipeline, now imagine a cluster of them with probably a lower IPC and yet longer pipelines each with more registers and also working as a single CPU...

Something along those lines IMO....

If this chip is even half as powerful as they say it is, then the x86 PC format as a whole will be under serious threat, and even Microsoft with the Xbox2 could ring the death knell also.....

[Solarius]

Think how the P4 can scale further due to it's low IPC and longer pipeline, now imagine a cluster of them with probably a lower IPC and yet longer pipelines each with more registers and also working as a single CPU...

Something along those lines IMO....

If this chip is even half as powerful as they say it is, then the x86 PC format as a whole will be under serious threat, and even Microsoft with the Xbox2 could ring the death knell also.....

Lower IPC and longer pipelines? How does that ring anything? Lower IPC and longer pipelines is the reason a 2.2ghz AMD chip bends a 3.2ghz P4 over and makes it its b****... Lower IPC and longer pipelines are the reasons Intel is running into trouble. I fail to see how the x86 PC format should even bat an eye. If you reduce your IPC to zero I bet you could make a chip running at 10,231,123,123 PHZ.. it just wouldn't do anything.

[Doc Oc]

Lower IPC and longer pipelines? How does that ring anything? Lower IPC and longer pipelines is the reason a 2.2ghz AMD chip bends a 3.2ghz P4 over and makes it its [beach]... Lower IPC and longer pipelines are the reasons Intel is running into trouble. I fail to see how the x86 PC format should even bat an eye.

Keep in mind that Sony is going to sneak it in through the backdoor... using it in Playstation 3, TV's, PVR's, portables... they're starting off attacking everything BUT the pc market. If in a few years time everything (except the desktop PC) is Cell, and cell *is* flexible enough to replace x86, it will stand a good chance of replacing the PC CPU as we know it today. Between it and that day stand Intel's billions... but Sony's no small player either, nor is IBM.

Interesting times...

[powerarmour]

Think how the P4 can scale further due to it's low IPC and longer pipeline, now imagine a cluster of them with probably a lower IPC and yet longer pipelines each with more registers and also working as a single CPU...

Something along those lines IMO....

If this chip is even half as powerful as they say it is, then the x86 PC format as a whole will be under serious threat, and even Microsoft with the Xbox2 could ring the death knell also.....

Lower IPC and longer pipelines? How does that ring anything? Lower IPC and longer pipelines is the reason a 2.2ghz AMD chip bends a 3.2ghz P4 over and makes it its [beach]... Lower IPC and longer pipelines are the reasons Intel is running into trouble. I fail to see how the x86 PC format should even bat an eye. If you reduce your IPC to zero I bet you could make a chip running at 10,231,123,123 PHZ.. it just wouldn't do anything.

Lower IPC and Longer Pipelines = Lower Heat per Mhz, I doubt that if a single Cell is running at 4.6Ghz that it is running a higher IPC than an Athlon64 for example, but it should have a heck load more registers via it's distributed computing. Are you telling me that eight 3.2Ghz P4's in a cluster wouldn't beat a 2.2Ghz A64 just because they are "Lower IPC"...?

The Cell chips consist of many low-power PowerPC processors on a single chip, with the number depending on the chip's particular application. For instance, a server might have four groups of cells, while a handheld computer might have one. Each cell has its own connection to memory, a control processor and eight attached secondary processors.

The Cell chips also use distributed processing. That is, they divide a problem into parts and parcel the chores out to different cells. Each parcel contains both data and the software application that has to be processed.

Anyway, that's not the point, the point is these "Cell's" are designed to be multiple processers (maybe 8 or 16 way), and as such you need a silicon design that dosen't require a HSF the size of a suitcase to cool it.

They must be running pretty cool at 4.6Ghz if they are designed to be clustered....

[powerarmour]

Keep in mind that Sony is going to sneak it in through the backdoor... using it in Playstation 3, TV's, PVR's, portables... they're starting off attacking everything BUT the pc market. If in a few years time everything (except the desktop PC) is Cell, and cell *is* flexible enough to replace x86, it will stand a good chance of replacing the PC CPU as we know it today. Between it and that day stand Intel's billions... but Sony's no small player either, nor is IBM.

Interesting times...

Nor is Microsoft....

I'am sure if Microsoft/IBM really wanted the Xbox2 to be the "next" PC it could force the issue without too many problems as long as the backwards compatibility was there.

Intel and AMD would both be left without a platform to develop for, now that is a crazy thought.....

[Harsh]

Yes, lower IPC and longer pipeline did allow the P4 to scale to higher clock speeds (a marketing dream come true). It was also the reason that a 1.4GHz PIII could outperform a 1.4GHz P4. It is also currently the reason that an AMD chip can be rated at 3200+ when it runs at 2.4GHz.

I must also agree that less work per clock = less heat, but there are many other factors involved in heat generation and it is not directly proportional, just like clock speed is not solely responsible for heat. So if you cut down the IPC by 50%, you would probably get less than a 50% reduction in heat generated.

You can't simply reduce the heat per core and solve the overall heat dissipation issue. For example, four or eight P4 3.06 CPUs on a single core would, all else being equal, yield four to eight times the heat genereated by one core. You could reduce the IPC of the P4 cores and reduce the amount of heat generated by the group, but the overall effectiveness would also be diminished.

It will be interesting to see what happens with these Cell chips. If this behavior has been overcome in these Cell chips, they will indeed be able to rival desktop CPUs.

At this point in time, it might be possible for AMD to produce a core that could run at 4GHz (might, since Intel couldn't), but the reduction in IPC and the increase in pipeline stages would result in something that could likely be labeled 3200+. It's almost like having an engine that can crank up to 13K RPM and a transmission that only has a really low first gear.

I could then pose the argument that the 2004 ZX-10R that can do 90MPH+ in first gear, but I don't want to crap on this thread any more than I already have

EDIT

BTW: I do think it will be interesting to find out if many scaled-down processors can perform better overall than one souped up > 100W monster. The multiple processors would be handy, but not in every application. The old 1 + 1 can't be broken down that much...

[Solarius]

Lower IPC and Longer Pipelines = Lower Heat per Mhz, I doubt that if a single Cell is running at 4.6Ghz that it is running a higher IPC than an Athlon64 for example, but it should have a heck load more registers via it's distributed computing. Are you telling me that eight 3.2Ghz P4's in a cluster wouldn't beat a 2.2Ghz A64 just because they are "Lower IPC"...?

They must be running pretty cool at 4.6Ghz if they are designed to be clustered....

No but 8 3.2ghz running in a cluster requires $1m's a year in air conditioning to reduce frequency of fire engine visits.. and that's already with lower IPC. Imagine how far you'd have to lower it to be able to fit 8 cells (and the faster, the lower) into a consumer sized unit?

I'm not saying its not a cool idea. But I don't see it giving a Dual-Core 939 90nm A64 processor a run for its money this decade.

[impar]

Greetings!

Related:
http://news.teamxbox.com/xbox/7511/Cell ... at-46-Ghz/