Support NeoGAF

[3rdman]

http://www.extremetech.com/article2/0,1558,1816966,00.asp

Great article and well worth the read...

Some of the specs seem particularly impressive. We've all heard about how powerful the Cell can be if it's fully utilized. Sony pulled a little bit of a bait-and-switch on this one: A chart showed how the number of gigaflops the Cell can perform stacks up against some CPUs from other consoles, including the Xbox 360. It was almost twice as much: 115 GFlops for the Xbox 360 and 218 GFlops for the Cell at 3.8GHz. That's just the thing, though: Five minutes later, when listing the actual specs of the PlayStation 3, Sony listed the Cell CPU as running at 3.2 GHz. My "back of the napkin" math puts the PS3's Cell CPU at 183 GFlops. That's more like 50% higher than Sony's quoted numbers for the Xbox 360, not twice as much.

Memory in the PS3 is sort of segmented. There's 256MB of XDR memory running at 3.2GHz, for a total bandwidth of 25.6 GB/sec. It's connected to the Cell CPU. The Cell CPU has a very fast and wide bus to Nvidia's RSX chip, which is in turn connected to 256MB of 700MHz GDDR3. The RSX chip can render to either memory bank, but we're not sure what the performance penalty is for jumping past the Cell CPU to use the XDR memory. The PlayStation 3 seems to have a total system bandwidth advantage over the Xbox 360, but it doesn't use a large cache of embedded DRAM for the frame buffer as the 360's graphics chip from ATI does.

The conference closed with a reel of "coming games," that looked extremely impressive, but was all just film footage. Nothing was running in real time, and I'm almost positive that only one or two of the demos in the reel were actually running in any kind of game engine, using in-game assets. It reeked of those "prerendered videos of what we think the games will look like," and I've never spoken to a single developer who's far along enough in PlayStation 3 development to have visuals and audio of the quality on display. As I spoke to journalists through the night, this seemed to be a point of contention. My question is this: If that stuff was running in the engine, why couldn't Sony show it to us live? Why only in a video reel?

As for me, I was a bit underwhelmed by Sony's demonstrations. All of the really super impressive stuff looked clearly prerendered to me—cinematics designed to show what the PS3 will be capable of, but not using the real game engine or real game assets. I got in several arguments during the evening with those who thought it was all real game footage, but I laughed it off. I don't know a single developer far enough along on PS3 games to produce in-game trailers like what we saw. Still, the specs are impressive, and it should easily be able to match or supercede the Xbox 360.

When it comes to raw technical capabilities, I think the systems are more evenly matched than the marketing guys would have us believe. In the current generation, Microsoft came to market 20 months after the PS2, and could take advantage of further process technology advancements and price drops to deliver much more powerful hardware. This time, the PS3 will come to market only about 6 months after the Xbox 360, using the same process technologies and working within the same basic heat envelope. All "paper specs" aside, the sort of experience that each system will be able to put on screen in a real game will be very similar. At least close enough that the general public won't think of one as more powerful than the other.

 

[PhatSaqs]

These guys are obviously biased.

 

[gofreak]

It was almost twice as much: 115 GFlops for the Xbox 360 and 218 GFlops for the Cell at 3.8GHz. That's just the thing, though: Five minutes later, when listing the actual specs of the PlayStation 3, Sony listed the Cell CPU as running at 3.2 GHz. My "back of the napkin" math puts the PS3's Cell CPU at 183 GFlops. That's more like 50% higher than Sony's quoted numbers for the Xbox 360, not twice as much.

He completely messed this up. First, the slide doesn't say 3.8Ghz, it says 3.2Ghz, and second, he's leaving the PPE out of his gflops calculation. The Sony calculation is correct.

 

[Suikoguy]

While I know the public demands comparisons, we just don't have enough real world information to make a comparision.

 

[jedimike]

While I know the public demands comparisons, we just don't have enough real world information to make a comparision.

An unscathing remark from suiko? Impressive.

 

[3rdman]

He completely messed this up. First, the slide doesn't say 3.8Ghz, it says 3.2Ghz, and second, he's leaving the PPE out of his gflops calculation. The Sony calculation is correct.

I think you missed his point. What he says is that Sony introduced their numbers for Cell in general (at 3.8Ghz) and compared it to the XCPU...It wasn't t a comparison of the actual PS3 CPU (at 3.2Ghz). In other words Sony was using the math based on 3.8 rather than 3.2 to get to 218 Gflops.

True or not, I don't know but this is what he meant, I believe.

 

[jimbo]

Biased or not, I think they're idea of the difference between the two is a lot more realistic than what people are coming away with from the conferences.

 

[sonycowboy]

Biased or not, I think they're idea of the difference between the two is a lot more realistic than what people are coming away with from the conferences.

If wishes were horses
Beggars would ride
All dreams and desires would ride along side
Worries and troubles would fall off behind
If wishes were horses, beggars would ride

 

[3rdman]

If wishes were horses
Beggars would ride
All dreams and desires would ride along side
Worries and troubles would fall off behind
If wishes were horses, beggars would ride

Damn good arguement! I guess that about wraps it up then!

 

[gofreak]

I think you missed his point. What he says is that Sony introduced their numbers for Cell in general (at 3.8Ghz) and compared it to the XCPU...It wasn't t a comparison of the actual PS3 CPU (at 3.2Ghz). In other words Sony was using the math based on 3.8 rather than 3.2 to get to 218 Gflops..

No they didn't. The slide was marked as 3.2Ghz, not 3.8Ghz:

And the math works out at 218Gflops for 3.2Ghz.

1 PPE (1*12*3.2Ghz)

+

7 SPEs (7*8*3.2Ghz)

= 217.6Gflops or ~218Gflops.

 

[3rdman]

Ah, I see! Why would he leave out the PPE then? Not very understanding of Cell, but doesn't the PPE direct the data to the SPE's and its the SPE's that do the calculations.

 

[gofreak]

Ah, I see! Why would he leave out the PPE then? Not very understanding of Cell, but doesn't the PPE direct the data to the SPE's and its the SPE's that do the calculations.

The PPE can be used for SPE management, but it's interference can be minimised with some extra work on the programmer's part. The PPE can either way execute general code, I believe. Certainly, the VMX unit on the PPE (which accounts for 8 of the 12 flops per cycle in the PPE) isn't going to be used for SPE management

 

[Pug]

Gofreak the problem before the Cell for PS3 was announced the CELL Ggfop numbers nver took into accoun the PPE!