IBM's Cell chip hype is total junk ?

[erhan28]

Interesting article.. Just thought I would like to share. any ideas on this?

--------------------------------------------------------------
By The Letterman: Tuesday 17 May 2005, 12:06

SO IBM hypersters are claiming their cell chip can do two teraflops per second. what about the lastest power5??? Go to IBM.com and find out here.

There you can see the latest and fastest power5 dual core processors can do 56.78 Gigaflop/sec in an eight way configuration. That's 16 Power5 cores combined, and the total is only 56.78 gigaflops for 16 lowly Power5 cores, or, about three gigaflops/per core. which makes sense, a float point operation takes more than one clock to finish, and each power5 core has probably two or more FP units...

So, there is no magic, you can't possibly do more than one instruction in one clock in one unit. You can do multiple if you have multiple units. To have 1000gigaflops, you will need 500 units each operate at two gigaflops.

A teraflop means 1000 gigaflops, so IBM is claiming that one PS3 is 2000% faster than an eight way power5 server selling at half a million bucks??

This is stupid.

Dave Y
==================
Source from the Inquirer
http://www.theinquirer.net/?article=23285

[mateo leet-o]

hmmmm interesting....but i don't know enough to comment

[Bizkitkid2001]

Two totally different processors designed for two totally different jobs.

The Cell packs all of its cores into the space the size of a thumtack, the Power5 processors are spread out among a larger distance. Closer is faster, especially as close as all cores on one chip. Not to mention the bus between each Power5 processor slows down the process alot.

But of course I will probably be wrong, and a more intelligent person will come into this thread and explain everything correctly.

[RamonGTP]

I don't think processors are limited to a single instruction per clock cycle as they were several years back. IIRC, the Athlon64 itself can do 3 instructions per cycle with a single core.

[Epyon9283]

The ps3 specs have the performance of the cell processor at 218 GFLOPS. The nvidia RSX is doing 1.8 TFLOPS. Add those together and you get the 2 TFLOPS everyone is talking about.

The number still seems really high though since the 6800 ultra only does 40 GFLOPS (http://download.nvidia.com/developer..._05_GPGPU.pdf).

[RamonGTP]

I have no doubt the figures put forth by sony are indeed inflated to fuel the hype. That said however, it will still be much more powerfull than anything we have right now.

[Dj-Icer]

I agree with Ramon.

[erhan28]

Yea, that is true but where how do they plan to sell this chip for this cheap price, it would cost like atleast $500

[Postal Dude]

I dont kno where i read it, Somewhere on e3 i think. But the cell processor has 8 of something (cant for the life of me remember what theyre called... SEP's maybe) Only 7 of these are activated during the use of the cell in the PS3, what would the 8th be for? anyone kno? or even have a clue what im talking about?

[jpiermarini]

i know what youa re referring too and its spe's i believe and not sep's. i myself was wondering about this. perhaps its time to google.

[butch81385]

i read the same thing postal dude, IIRC it is for "redundancy" or something like that...

[coolguy1986]

I question the specs of the Ps3, i mean they said that the RSK(its GPU) is more powerful than 2 6800ultra's which currently run for 1000 bucks.

[elmers]

So they sell the console at a loss, and overprice games to compensate. What else is new?

[coolguy1986]

Selling a console worth 1200bucks for 400 or 500, selling games at 60bucks doesnt bring much profit if any at all.

[butch81385]

Selling a console worth 1200bucks for 400 or 500, selling games at 60bucks doesnt bring much profit if any at all.

ahh, but licensing the technology, licensing the game devolpment, as well as other side things will bring in a profit...