PS3 Open Platform - some big news

It means the PPE is about what we expected in performance compared to a desktop CPU. Though there is some curious data here. There seems to be a huge disparity between memory read and write performance on the PPE according to this benchmark. And some numbers like stdlib write seem too high to be telling the full story.


I just ran the same benchmark on my A64 X2 3800.
There are 3 areas where the PS3 PPE is faster.

Dot Product single-threaded scaler is almost 3x faster.
Stdlib Write single threaded scaler 5.3x faster
Stdlib Write multi threaded scaler 4.3x faster

 

The write scores are a result of XDR?

 

Right, fantastic at arithmetic, pants at most everything else.

Don't know if it's non-temporal streams, but either way: reads are probably latency limited, even for the multi threaded approach.

I'm guessing writes are coalesced into large transactions with write-buffers, so are much closer to actual bandwidth peak.

cheers

 

Hopefully we see SPE benchmark in the near future including the one that has many branches!

 

According to this link:
http://www.systemshootouts.org/processors.html
a 1.6GHz G5 ia about as fast as a 2.53GHz P4.

That is perfectly usable for a general purpose desktop, especially when you consider that for a desktop OS, it is media acceleration and not integer performance that is the system bottleneck. Use the SPE's and RSX for media/graphics acceleration (are you listening Sony?) and you have a pretty decent desktop machine.

 

Any idea what he means by this:

When the 1.6GHz G5 doesn't score 100 in any of the tests? Presumably the baseline is actually for some other system, an Intel one?

 

If true, this is really good news for me !

 

but linux on ps3 use spe or only ppe ?

 

According to this, in gaming a 2.5Ghz G5 (dual) is quite significantly slower than a 2.6 Ghz Athlon64 or dual Opteron.

The closest they get is in the multithreaded and very mac optimised (as well as PC obviously) Quake 3 test. Here the G5 is only 10% slower than the Opteron.

In the CPU tests the G5 fares better, trading places with the 2.6Ghz Opteron in most areas.

http://www.barefeats.com/macvpc.html

I think we need a wider array of tests to really tell but I guess overall we could estimate that the PPU is roughly as powerful as a 1.6Ghz Athlon64 from the benchmarks above - assuming the A64 is roughly as fast as the G5 clock for clock. Thats actually a lot faster than I would have thought and would suggest that even two of them in tandem would be at least a match for the fastest single core A64 out there.

A 1.6Ghz A64 would be equivilent to what? Say a 2Ghz AXP? Thats the AXP 2400+ so I guess the comparison above is roughly accurate when is says a 1.6 Ghz G5 is equivilent to a 2.53 Ghz P4. Thats also supported by the rough parity between the 3.4Ghz Xeon and the 2.5Ghz G5 (ignoring the gaming tests).

 

Shouldn't post before I've had my morning caffeine - the 100 is for a 1.6GHz G5 running Mac OS X, the scores in the article are for a 1.6GHz G5 running Fedora Core 4. That actually makes me think the benchmark is pretty poor as a measure of raw CPU performance - there is a pretty big difference in some tests between the same processor under the two different OSs which you wouldn't expect if the benchmarks were good measures of straight CPU / memory performance. He doesn't say in the post what compiler he's using or what settings - is it the same version of GCC on all platforms? Given the G5 under Linux scores as much as 168 vs the G5 under Mac OS X's 100 in one test and as little as 55.6 in another you have to question the value of these benchmarks as a measure of CPU or memory performance.

The PS3 PPU and 360 cores are much more dependent on compiler optimizations than the G5 and presumably the version of GCC used on the PS3 is just a general PPC version and not specifically targeted at the PPU so it's going to give significantly worse results than if you used Sony's version of GCC that ships with the devkits.

 

OSX has had shit OpenGL drivers from Apple forever, and applications where the Mac has a negligible presence (like, for instance, games) have been tuned specifically for x86, with comparable PPC performance being at least a total rewrite away (Doom 3 is a glaring example).

This is a shootout with too many intermediate layers to come to any conclusive ratio of what clock G5 matches up to whatever other clock on whatever other architecture, and certainly has even less relevance for a PPE running Linux.

 

To be fair - these are binaries optimized for dramatically different PPC then what these consoles have. Particularly when some 3 letter company implemented a "highly efficient" memory subsystem that will happilly stall on cache hits when the stars aren't in the right alignment.

I doubt many anti-SPE proponents really want to see those.

 

Have you guys seen this?

http://www.neogaf.com/forum/showthread.php?t=130532

It looks to my untrained eye like linux is making PS3 piracy a lot easier. I always wondered about that, but just assumed Sony wouldn't do anything that stupid.
________
Ocean View Condos

 

Isn't there something strange related to BluRay burning and disk compatability? The formats are quite different, aren't they (and can easily be distinguished by readers)?

 

PS3 l33t hackerz guy seyz

But yeah, it (should) mainly be an offline situation, much like Xbox/360. I'm just wondering if linux is going to somehow make it a whole different ballgame, and for that I have no knowledge.
________
MiAmore cam

 

I am sorry, but that guy is just plain wrong. He's making backups because his disc get scratched? Last time I checked, you can get your scratched disc changed by the publisher.

 

Cheers