PS3 Linux Benchmarks

I finally saw some benchmarks for the PS3. Geek Patrol

Highlights: 1.6 G5 vs. PS3 with Xeon benchmarks added. bZip and JPEG seemed like the most common real world part of the benchmarks.

bzip2 Compress (multi-threaded scalar)
PlayStation 3 - 124.1
Power Mac G5 - 168.4
Xeon 5160 - 1194.4

bzip2 Decompress (multi-threaded scalar)
PlayStation 3 - 99.5
Power Mac G5 - 133.1
Xeon 5160 - 1353.3

JPEG Compress (multi-threaded scalar)
PlayStation 3 - 94.8
Power Mac G5 - 103.0
Xeon 5160 - 877.6

JPEG Decompress (multi-threaded scalar)
PlayStation 3 - 72.9
Power Mac G5 - 119.2
Xeon 5160 - 788.9

So basically in performance the PS3 processor usually isn't going to be as good as a three and a half year old budget G5 processor.

The only thing that was good was its memory write speed, but it's memory read speed was slow. Why have such good write speed, but have slow read speed?

Just to let you know, its not using the full cell processor in these benchmarks (only the PPE, no SPE's). If you look at it that way, these are actually pretty impressive numbers.

BladeVenom said:

So basically in performance the PS3 processor usually isn't going to be as good as a three and a half year old budget G5 processor.

Click to expand...

Did you even read the article?

"Geekbench also isn’t able to exploit the eight vector processors on the Cell processor. Any program designed and optimized for the Cell processor should be a lot faster than one designed for a generic processor (like, say, Geekbench). So while the Geekbench results might seem disappointing, keep in mind that Geekbench can’t exercise the PlayStation 3 to its full potential."

This is what it comes down to. The benchmark suite they used was clearly not optomized for the cell and they didn't present it as such. The cell is a radically different processor that is gonna reach much farther than just the ps3 because it runs circles around other cpus in the right applications. It may not be truly 9 core like they presented it as at first, but it destroys 4 separate all purpose cores hands down in a lot of cases for a fraction of the cost, space, and heat.

Most games are cross platform so those cells aren't going to be used anyways. Name for me one game that uses them?

it destroys 4 separate all purpose cores hands down in a lot of cases

Click to expand...

Then name some. I'm awaiting your links.

BladeVenom said:

Most games are cross platform so those cells aren't going to be used anyways. Name for me one game that uses them?


Then name some. I'm awaiting your links.

Click to expand...

Name one game that doesn't. Have you personally seen every games' source code to verify that no threading is taking place? I recall hearing a few months ago that some developers planned to use separate cores for sound processing, network traffic, i/o traffic, physics, etc, which in and of itself would vastly improve performance.

As for benchmarks...that's proprietary information from my workplace that I'm not allowed to give specifics on

Bad_Boy said:

Just to let you know, its not using the full cell processor in these benchmarks (only the PPE, no SPE's). If you look at it that way, these are actually pretty impressive numbers.

Click to expand...

That assumes software that uses the SPEs is ever going to become commonplace on Linux. And, no, losing to a single G5 isn't impressive, period.

erwos said:

That assumes software that uses the SPEs is ever going to become commonplace on Linux. And, no, losing to a single G5 isn't impressive, period.

Click to expand...

It will absolutely become commonplace. Maybe not in the ps3 distributions of it, but on the server side IBM is pushing this thing really hard.

erwos said:

That assumes software that uses the SPEs is ever going to become commonplace on Linux.

Click to expand...

Depends on the application. As Spaceman_Spiff said, the applications IBM has planned for the cell will most likely be optimized for the cell. Why wouldnt they? Linux PS3 applications maybe no so much as he also said.

erwos said:

And, no, losing to a single G5 isn't impressive, period.

Click to expand...

This benchmark wasnt even optimized for the cell. If only a fraction of the cell's power is around par with it, I think thats pretty good IMO.

if the cell was so good, it wouldnt need optimized software for anything even benchmarking, if was truly so fucking awesome and all powerful as the PR has made it out to be.. but we all know that it is not the next greatest thing.. its just another thing.

The benchmark is also not optimized for the G5.

Geekbench’s benchmarks are written in platform-neutral C++, and have no platform-specific optimizations.

Click to expand...

Tetrahedron said:

if the cell was so good, it wouldnt need optimized software for anything even benchmarking, if was truly so fucking awesome and all powerful as the PR has made it out to be.. but we all know that it is not the next greatest thing.. its just another thing.

Click to expand...

As I've said, the cell kills other cpus in scientific based applications, but does that mean its the best chip out there for every situation? Absolutley not, because even if it was optomized its not an all purpose cpu. Its vastly different architecture that linux is not utilizing yet. Read the article please, they in no way suggest anything negative about the cell based on their benchmarks. While you're at it, read up on processor architectures and os design too.

BladeVenom said:

The benchmark is also not optimized for the G5.

Click to expand...

True, but the cell is vastly different architecturally than any other cpu out there. If it was 8 normal cores it would be fair to judge it this way, but this test essentially looks at one and only one core, which is not a fair assessment of the intent of the design because of the way it divies up tasks. All things are not equal, the G5 still has a large advantage in this environment.

Basically this thread comes down to 2 things:
1.) The testers in no way are disappointed or surprised by the performance
2.) The cell represents a new way of doing things, not the same old "more mhz, smaller die" rehash that x86 throws at us. Remember, the cell is IBM's baby, not sony's. Do yall know something about microprocessor design that everyone at IBM doesn't?

I'm not speaking at all to the decision to put this into the ps3, merely to the processor itself.

Spaceman_Spiff said:

As I've said, the cell kills other cpus in scientific based applications

Click to expand...

You can say it all you want, but it doesn't make it true. Lets see a link to some real benchmarks.

Do yall know something about microprocessor design that everyone at IBM doesn't?

Click to expand...

Do you know something about microprocessor design that Intel, who is number one chipmaker in the world, doesn't know? Don't forget AMD who is number seven. IBM is number twenty one.

Do you know something that Apple doesn't know? After they saw the Cell processor, they switched to Intel.

Intel used a coprocessor in 1980. They also used an in order only design until the Pentium Pro. The Cell processor isn't really anything totally new, but it does seem to make for good marketing.

Come on BladeVenom, are you actually saying these benchmarks are realisticly depictive of cell's power?

BladeVenom said:

You can say it all you want, but it doesn't make it true. Lets see a link to some real benchmarks.


Do you know something about microprocessor design that Intel, who is number one chipmaker in the world, doesn't know? Don't forget AMD who is number seven. IBM is number twenty one.

Do you know something that Apple doesn't know? After they saw the Cell processor, they switched to Intel.

Intel used a coprocessor in 1980. They also used an in order only design until the Pentium Pro. The Cell processor isn't really anything totally new, but it does seem to make for good marketing.

Click to expand...

You're right. I concede, the cell is a g4. I see it so clearly now. Thank you so much for not reading this article and then misconstruing the results into a conclusion completely different from that of the author.

Its clearly not worth the time to argue with people like this, let this thread die.

Bad_Boy said:

Come on BladeVenom, are you actually saying these benchmarks are realisticly depictive of cell's power?

Click to expand...

In most scenarios yes. Certainly for general purpose programs, and cross platform games.

BladeVenom said:

In most scenarios yes. Certainly for general purpose programs, and cross platform games.

Click to expand...

Then maybe you should read your own link again.

Spaceman_Spiff said:

Its clearly not worth the time to argue with people like this, let this thread die.

Click to expand...

Agreed.

Everyone should give this article a read to get a more in-depth idea of how the processors compare to traditional out of order execution desktop processors, which will also shed some light on performance benchmarks above.

I agree with spiff though, lets just let this thread die because it's borderline flamebait in this subforum.