The cell is all hype.

[imported_kouch]

I would say WAYYYYY more than 75% hype, cause that would still put it at 250x the performance of a dual opteron server, lol what a joke. Maybe it doesn't even need electricity to run guys, can run 5 years on a single AA battery. Gimme a break, how is it possible to produce a processor that is faster at everything than all the proprietory chips specifically optimized for those applications. That is like saying software rendering is faster than hardware rendering.

edit: if you think otherwise, lets put money down right now that Cell does not kill a dual opteron server right now in any server apps.

 

[zakee00]

Originally posted by: kouch
I would say WAYYYYY more than 75% hype, cause that would still put it at 250x the performance of a dual opteron server, lol what a joke. Maybe it doesn't even need electricity to run guys, can run 5 years on a single AA battery. Gimme a break, how is it possible to produce a processor that is faster at everything than all the proprietory chips specifically optimized for those applications. That is like saying software rendering is faster than hardware rendering.

edit: if you think otherwise, lets put money down right now that Cell does not kill a dual opteron server right now in any server apps.

its almost a joke really. rofl they said it will start at 4GHz on 90nm, when Intel (which has been all about MHz) isnt even there yet on 90nm. and thats what they specialize in also, not making walkmans (jk i know IBM is making it)
they are so retarded...my bet is that by they time they release the PS3 my computer will play games almost as good.

 

[imported_kouch]

Originally posted by: zakee00

Originally posted by: kouch
I would say WAYYYYY more than 75% hype, cause that would still put it at 250x the performance of a dual opteron server, lol what a joke. Maybe it doesn't even need electricity to run guys, can run 5 years on a single AA battery. Gimme a break, how is it possible to produce a processor that is faster at everything than all the proprietory chips specifically optimized for those applications. That is like saying software rendering is faster than hardware rendering.

edit: if you think otherwise, lets put money down right now that Cell does not kill a dual opteron server right now in any server apps.

its almost a joke really. rofl they said it will start at 4GHz on 90nm, when Intel (which has been all about MHz) isnt even there yet on 90nm. and thats what they specialize in also, not making walkmans (jk i know IBM is making it)
they are so retarded...my bet is that by they time they release the PS3 my computer will play games almost as good.

Almost as good? PC's with latest hardware (which would be like $2000 range) will always easily outplay any console sold at $300 unless the console people are willing to take a $700 loss (assuming a extreme 50% margin on pc parts) on every part sold.

I bet an A64 and 6800 ultra system now will beat all next gen consoles.

 

[zakee00]

the only reason my computer wouldnt be as good is because nvidia is unsing the new NV50 for the PS3. that will probably be faster then my 6800GT @ Ultra.
I am confident that my winnie at 2.655GHz will be as fast or faster then the Cell for gaming.

 

[Schmeh]

Originally posted by: kouch
I would say WAYYYYY more than 75% hype, cause that would still put it at 250x the performance of a dual opteron server, lol what a joke. Maybe it doesn't even need electricity to run guys, can run 5 years on a single AA battery. Gimme a break, how is it possible to produce a processor that is faster at everything than all the proprietory chips specifically optimized for those applications. That is like saying software rendering is faster than hardware rendering.

edit: if you think otherwise, lets put money down right now that Cell does not kill a dual opteron server right now in any server apps.

I promise that a cell workstation will be able to outperform an opteron workstation when it comes to graphics rendering and design.

IBM says they already have a protype Cell workstation that can reach 16TFLOPS.

Link: http://www-03.ibm.com/chips/news/2004/1129_cell2.html

 

[xbdestroya]

I'm right there with you Schmeh.

 

[fsstrike]

Cell sounds too good to be true, and I doubt the claims are real IMO. However, I will wait for cretible benchmarks before I can truely have a real opinion.

 

[zakee00]

even if that is true, lets see how fast they can get them out the door or if they can provide a sufficent ammount for the hordes of PS3 buyers.
IBM had some trouble manu. the G5 for apple. Apple had to delay the launch of the new iMac, and it took them weeks to get up to date with powermac orders.

 

[91TTZ]

Originally posted by: Schmeh
According to CNet News.com, the cell, which is a cpu with a 64-bit Power Processor and 8 "synergistic processing units" can perform 256GFLOPS. By comparison a Pentium 4 3Ghz chip can perform 6GFLOPS. Yes, the Cell can not only meet the performance of Intel and AMD's CPUs but it can outperform them by quite a large margin.

They don't give enough information to arrive at an accurate conclusion. What kind of operations are they, and is it the same type of operation being tested on both processors? Or are they doing the typical marketing BS like finding a rarely used instruction which executes very fast on their chip but very slow on the P4?

Remember when the G4 came out? They had all this "supercomputer" talk and they showed how many more instructions it can process compared to the P3, yet when you looked at the details of the test, they set it up in a way that compared the G4 using Altivec to the P3 without using SSE. When you looked at the real world performance of the chips, it wasn't that good.

 

[91TTZ]

Originally posted by: Schmeh

I promise that a cell workstation will be able to outperform an opteron workstation when it comes to graphics rendering and design.

Promise is too strong of a word. Nobody outside of the companies involved has even seen one running. No systems have been demo'd, no benchmarks have been run.

Originally posted by: Schmeh

IBM says they already have a protype Cell workstation that can reach 16TFLOPS.

And I say that I have an alien spacecraft in my garage. I can't quite back up that claim though. Let's see if IBM can.

 

[zakee00]

rofl nice one about the aliens, so true.

G4 using Altivec to the P3 without using SSE. When you looked at the real world performance of the chips, it wasn't that good.

i own a G4 PowerBook, and SSE and AltiVec are different. Every good Mac app is AltiVec optimized, even the OS. it makes a huge difference in performance. the only difference between a G3 and a G4 is altivec added, and look at the performance increase in everything (real world).

 

[Megamixman]

Ok since it is based of the PowerPc Architecture, how could it suddenly scale to 4Ghz, when IBM has had trouble getting the G5 to even 3Ghz. Either that or that thing consumes enough power to run a city. The Cell is mostly hype. Secondly that theoretical 256Gflops is probably if all the computational units were running simultaneously crunching out useless numbers that have nothing to do with each other. So it might becomes usefull in something like datacenters and thats about it. Where else have you seen 64 cpu computers, etc.

*sarcastic*
It will also probably have lots of trouble with bandwith like the P4, consume enough power to drive an electric car, and require the quadruple phase change coolers as a minimum.

 

[VirtualLarry]

Originally posted by: zakee00

Now this is not to say that the Cell will take over the computer industry, but I do think that since it takes a couple hundered Opterons, Xeons, or Itaniums to get 1TFLOP and it would only take 8-15 Cells to get the same performance, the cell has a great chance at being dominant in the Super Computer market, as well as the workstation market for graphical development.

total BS, they is simply no way that it can hit 256GFLOPS. there is obviously a huge piece of info we are missing. if it was right now physicly possible to make a chip that did that you would be hearing alot more about it.

Actually, I did hear about one some time back, unrelated to Cell. It was a very specialized chip, designed specifically for vectorized numerical processing. It's not hard to come up with those throughput numbers for a specialize chip designed to do just one thing and do it well. That's how GPUs work for graphics. As long as you can parallelize the task, you can simply tack on more pipelines for more throughput. The problem is in handling generalized Von-Neumann-style computing loads, with lots of (fairly unpredictable) branches, etc. That's why no general-purpose CPU has yet to show that kind of numerical-performance numbers. So that's really the thing - you can develop a chip with that kind of specialized numerical power, but it's not applicable to generalized computing tasks, and so it will never displace general-purpose processors, no matter how powerful it becomes.

 

[VirtualLarry]

Originally posted by: kouch
Gimme a break, how is it possible to produce a processor that is faster at everything than all the proprietory chips specifically optimized for those applications. That is like saying software rendering is faster than hardware rendering.

At times, and in many cases, that is true.

Many high-end (at the time) arcade games, ones developed as offshoots of military simulator technology, were actually software-rendered.

Sega's Model2-based 'Daytona USA' game was software-rendered, for example, running on a setup with quad Fujitsu 64-bit DSPs that processed the display-list generated by the main CPU/game-program, and rasterized to a framebuffer. The FSAA effect on the final output might have been done in hardware though, not sure.

 

[Eug Wanker]

Originally posted by: Megamixman
Ok since it is based of the PowerPc Architecture, how could it suddenly scale to 4Ghz, when IBM has had trouble getting the G5 to even 3Ghz.

I dunno, but a couple of things...

1) It may be based on POWER5. The G5 is based on the POWER4.
2) Even if it's based on POWER5, Cell's main core may be a much more stripped down POWER5 than the G5 was of the POWER4.
3) Refinements in the process? The G5 was in some ways Cell's 90 nm guinea pig.
4) Maybe I'm talking outa my arse.

Anyways, the ISSCC slides are here. It seems like the specs are:

1) A single PowerPC core, with VMX SIMD and simultaneous multi-threading
2) 8 Synergistic Processor Elements
3) FlexIO bus interface controller
4) Integrated dual XDR memory controller
5) Elemental Interconnect Bus that connects all of the above

 

[zephyrprime]

Originally posted by: 91TTZ

Originally posted by: Schmeh
According to CNet News.com, the cell, which is a cpu with a 64-bit Power Processor and 8 "synergistic processing units" can perform 256GFLOPS. By comparison a Pentium 4 3Ghz chip can perform 6GFLOPS. Yes, the Cell can not only meet the performance of Intel and AMD's CPUs but it can outperform them by quite a large margin.

They don't give enough information to arrive at an accurate conclusion. What kind of operations are they, and is it the same type of operation being tested on both processors? Or are they doing the typical marketing BS like finding a rarely used instruction which executes very fast on their chip but very slow on the P4?

GFLOP are always marketing more than anything else. But this fact isn't unique to Sony so I wouldn't fault them for exageratting in this instance.

Anyway, it's pretty easy to figure out how they came to that number:
4GHz*8 processing units * 4 way vector instructions * 2 for multiply-accumulate = 256 GFLOPS.

The funny thing is, if you use the same calculation method for the 3GHz P4, you'd get 24GFlops - not 6 GFlops. You'd only get 6GFlops if you only considered the P4's FPU and not it's SSE unit. So the comparison to the P4 is highly misleading.

The cell is an insteresting architechture and I appluad it. I don't like the idea of "cells" very much since I think they're a huge security risk but I do like the idea of having all the little on-die vector co-processors. IBM likes to call their vector on-die co-processors"synergistic processing units" but I don't like that name because it's so uninformative.

The vector coprecessing units are a great way to add a lot processing power for minimal expense. They're great for matrix math but suck for b-trees. I expect that the PS3 will be able to handle very high polygon counts and more extensive physics with it's added math power.

It sounds like cell's vector units will only be single precision so they'd be just about useless for scientific computing contrary to IBM's claims. Perhaps IBM will make a more expensive double precision variant for the super computer market but unless it does so, I do not expect Cell to enjoy any success in that area.

 

[DrMrLordX]

Originally posted by: Megamixman
Ok since it is based of the PowerPc Architecture, how could it suddenly scale to 4Ghz, when IBM has had trouble getting the G5 to even 3Ghz. .

Last I heard, and this is just what I heard, the Power5 core on each Cell will be clocked at 1/2 the speed of the overall Cell itself. In other words, the vector processing units should be running at the touted 4.6 ghz, but the Power5 core will be running at a much-slower 2.3 ghz or so.

Any software that fails to make extensive use of the various vector processing units will likely not perform all that well on Cell.

 

[zakee00]

Originally posted by: DrMrLordX

Originally posted by: Megamixman
Ok since it is based of the PowerPc Architecture, how could it suddenly scale to 4Ghz, when IBM has had trouble getting the G5 to even 3Ghz. .

Last I heard, and this is just what I heard, the Power5 core on each Cell will be clocked at 1/2 the speed of the overall Cell itself. In other words, the vector processing units should be running at the touted 4.6 ghz, but the Power5 core will be running at a much-slower 2.3 ghz or so.

Any software that fails to make extensive use of the various vector processing units will likely not perform all that well on Cell.

thats weird

 

[arcas]

I find Cell's lack of instruction reordering abilities to be an interesting decision. On the one hand, this is similar to Itanium's approach and it will require a smart compiler to perform the instruction scheduling. On the other hand, since their target audience is game consoles, this decision was probably a good one. A game machine can get away with a less powerful CPU in favor of stronger SIMD abilities. Cell's designers might have saved enough die space by not implementing reordering that they could add more vector units.

Intel spend wads of cash on Itanium compiler development and one can argue that after several years of development they now have a pretty decent compiler. It remains to be seen whether Sony and IBM are going to devote similar resources towards Cell compiler development. I think this will be critical if Cell is going to be used for general-purpose computing otherwise the lack of instruction reordering is going to hurt.

 

[clarkey01]

we'r all adding to the hype, so dont blame sony

 

[mooncancook]

when my DBA (a really old man) modified some slow sql query he always claimes that its performance is improved by 10-100 thousand percent. He sounds dead serious and excited. well, either he is a genius or an idiot. go figure.

let's not discount the technology tho. I don't think it's going to revolutionize anything in the computer world, but i DO HOPE it will revolutionize computer pricing - no more $800 CPU, and no more $500 GPU! competition is good.

 

[Eug]

I see no reports saying the PowerPC core will run at half clock speed.

Anyways, it seems like it's confirmed it supports VMX and SMT, but some think it is likely not a stripped down POWER5 core, but another core that isn't in current use. Not a huge amount of info came out on the core today at ISSCC though. Too bad.

Further information/speculation here.

It sounds like this CPU is going to be EXTREMELY fast for stuff that can make use of the SPEs. However, for anything else, it's "just" going to be sort of like a fast G5. Note: I am just guessing, since I am not an engineer.

 

[DrMrLordX]

Originally posted by: Eug
I see no reports saying the PowerPC core will run at half clock speed.

It was mentioned in a post here on Anandtech by xbdestroya.

This should be it

Just read down a bit and look for the italics in a post by xbdestroya. The commentary quoted by xbdestroya indicates that the Power core in the Cell(referred to here as a PU) should be running at around 2.4 ghz. Here is the block of quoted text from xbdestroya's post:

Now the PUs on the other hand are probably a lot closer to ordinary PPC devices, albeit probably lacking in the SIMD portion since you have all these extra SIMD pipelines attached. My thinking is that since the APUs are essentially all independent anyway, they would also be independent of the PU, which is far more complex in nature. So since the PU is essentially almost a PPC in itself, I'd think that if the APUs run at 4.8 GHz or something, the PUs would run at 2.4. That's certainly a feasible speed for a PPC, and at 65nm, the effective power consumption would be low enough that you may not have to worry.

On a side note, here's some information from Ars Technica's Hannibal about Cell:

http://arstechnica.com/articles/paedia/cpu/cell-1.ars

 

[VirtualLarry]

Originally posted by: Eug
It sounds like this CPU is going to be EXTREMELY fast for stuff that can make use of the SPEs. However, for anything else, it's "just" going to be sort of like a fast G5. Note: I am just guessing, since I am not an engineer.

That's what I sort of expect too - for general-purpose computing tasks, the "Cell" will only be primarily running the PPE, and should be about the same sort of speed as a ~4Ghz Power-esque CPU core. But if you need to do a lot of vector floating-point calcs, it should FLY! (Btw, does anyone remember the Dreamcast? A 64-bit Hitachi SH-2 (I think, or was it -3/-4?), but with a custom 128-bit vector floating-point unit grafted on, so it could handle a lot more 3D-math type operations. Paired with a GPU, it made a nice little system. (It was eventually later all shrunk down onto a single chip, and clocked higher, no idea what that "Son of DC" chip is being put in, I think DVD players. Too bad, I would have loved to see a handheld portable DC, like the upcoming PSP unit from Sony.)