StarDestroyer.Net BBS

PS3 hardware slow and broken

In words and pictures


By Charlie Demerjian in Taipei: Monday 05 June 2006, 03:35

AFTER BREAKING THE news to me about PS3 RSX speeds earlier on the flight to Japan, my row-mate said 'if you think that's interesting, wait till you see this. Cell is hurting, badly'.

For those of you that believe in religions with karmic tendencies, scoops like this meant one of two things, the wings of the plane are about to fall off and I am going to die in a fiery ball, or worse yet, the movie selection will be worrisome. Cell memory access appears to be broken, RSX has half the triangle setup rate of the ATI chip in XBox360, and the true horror, Big Momma's House 2 and a Queen Latifa movie.

With the movie selection still making my brain throb from the glances I caught, I furiously took notes on what the source was saying. He started out saying that the RSX can only write about half as much vertex data as it can fetch, not an ideal situation by any stretch, but survivable.

Then came the horrible news, RSX appears to be limited to setting up 275 Million triangles/second, anemic compared to the 500+ million in XBox360. When asked about this apparent thumping dished out by MS, the reply from one notable ISV relations boffin was a terse 'What a Piece of Junk'. Talk about a steak in the heart.

Half the triangle setup capability in the PS3, could things get worse? Yes, far far worse, how about another disparity of three orders of magnitude? No, I am not joking, looking at Sony's own figures, Cell appears to be pretty badly broken.



For main memory, it looks like Cell has about 25GBps of main memory bandwidth, and RSX is about 15-20GBps. Achievable bandwidth is between about two thirds of that and nearly 100%, clearly the elves in the caves surrounding Rambus central did something right with XDR. That is the happy news.

For local memory, the measured vs theoretical bandwidth is missing, I wonder why? RSX is at a solid 22.4GBps for both read and write, good job there green team. Then comes the blue team with Cell. Local memory write is about 4GBps, 40% of the next slowest bandwidth there. Then comes the bomb from hell, the Cell local memory read bandwidth is a stunning 16MBps, note that is a capital M to connote Mega vs a capital G to connote Giga. This is a three order of magnitude oopsie, and it is an oopsie, as Sony put it "(no, this isn't a typo...)".

If you can write at 250x the read speed, it makes Cell local memory just about useless. That means you do all your work out of main memory, and the whole point of local is, well, pointless. This can lead to contention issues for the main memory bus, and all sorts of nightmarish to debug performance problems. Basically, if this Sony presentation to PS3 devs shown to us is correct, it looks like PS3 will be hobbled in a serious way.

The next slide goes on to say "Don't read from local memory, but write to main memory with RSX(tm) and read it from there instead", and repeats the table numbers. This is very very bad. The number of times the presentation goes on to say that it is correct, and the lack of anything like "this will be fixed by production steppings, so take measures X, Y and Z" say to me that it is not a fixable snafu. Remember at E3 when I said that the PS3 demos there were object sparse? Any guesses why?

Someone screwed up so badly it looks like it will relegate the console to second place behind the 360. All the devs I talked to were lukewarm on the 360 architecture but universally negative on the PS3. Revelations like this go a long way to explain why you keep hearing about simmering problems from the Sony devs.

You end up with a console with half the triangle setup rate of the 360, a crippled CPU that is a bitch to program, and tools that are atrocious compared to the 360. To make matters worse, you have an arrogant set of execs telling us that twice the price is worth it for half the power, a year late. If it isn't already too late, Sony had better do something about this recto-cranial inversion or it may very well sink the console. µ

B) Ther reporter is relating something, the guy who he sat next to on the plane told him.

Aaron Ash wrote:I would like to see Sony take some losses in the new generation, however I can't find it in the darkest corners of my Sony hate to take this article seriously because:

A) It's the Inquirer
and
B) Ther reporter is relating something, the guy who he sat next to on the plane told him.

This story just reeks of wanting to be on the anti-Sony bandwagon.

Now I'm no 3D hardware tech head, but from what I've read on other forums people seem to be saying that this is basically and apples to oranges comparison. It is apparently true that the PS3 turns over fewer triangles than the 360, but it also applies various pixel shading effects to them during that process that the 360 has to do later in a more laborious manner.

Xon wrote:Holy fuck, 16mb/s read for the local memory cache?

Well this is Sony.

Arrow wrote:

Xon wrote:Holy fuck, 16mb/s read for the local memory cache?

Well this is Sony.

Obviously, someone didn't tell the kiddies designing the thing that whole point of a local cache is that it's faster than system memory! Idiots. What kind of retard does it take to make the read/write performance so lopsided. And why wasn't this issue corrected long, long ago?

Admiral Valdemar wrote:It's also "stake in the heart", not "steak".

LMAO. I had to laugh all the way to the bank.

Local storage's read/write at 16MBp/s is bad? How much memory are they expecting the SPE to address, 10 GB? The local SRAM is only 512KB... 16MBp/s is plenty enough.

Not to mention, the RSX memory is WAY off. That's for the FlexIO. The RSX memory is 22.4GBp/s, same as the Xbox 360's Xenon.

The CELL's 25GBp/s is actually faster than the Xenos's 22.4GBp/s, so.. who the hell cares?

As for the triangles... Don't make me laugh, the one thing that the RSX EXCELS at, if nothing else, is pumping out triangles.

Disregard this crap article entirely.

Crazy_Vasey wrote:I am in tears of laughter here. 16Mps for CPU cache!? What is this, a Commodore 64?

The article took the slide out of context.

The slide was talking about RSX, therefore "local memory" was the RSX local memory. Which means that when Cell needs to read the RSX's local memory it reads it at 16MB per sec. This is normal as GPUs are consumers, not data providers, CPUs rarely need to read back what the GPU generates as its all throwaway once your process the next frame.

Praxis wrote:Another quote.

The article took the slide out of context.

The slide was talking about RSX, therefore "local memory" was the RSX local memory. Which means that when Cell needs to read the RSX's local memory it reads it at 16MB per sec. This is normal as GPUs are consumers, not data providers, CPUs rarely need to read back what the GPU generates as its all throwaway once your process the next frame.

Fukui wrote: I can dress my links

From the report:
"The XDR controller actually consists of two independent controllers, offering more flexibility than one. The memory interface can support an incredible 25.6GB/s of bandwidth."

Its quite obvious, the enquire report is totally bogus.
Poor enquirer.

Praxis wrote:I asked for a source, he responded with this.

Fukui wrote:I can dress my links


From the report:
"The XDR controller actually consists of two independent controllers, offering more flexibility than one. The memory interface can support an incredible 25.6GB/s of bandwidth."

Its quite obvious, the enquire report is totally bogus.
Poor enquirer.

Praxis wrote:I asked for a source, he responded with this.

Fukui wrote: I can dress my links


From the report:
"The XDR controller actually consists of two independent controllers, offering more flexibility than one. The memory interface can support an incredible 25.6GB/s of bandwidth."

Its quite obvious, the enquire report is totally bogus.
Poor enquirer.

Edi wrote:Dress your goddamn links, idiots! I don't like it when the page format gets fucked up because someone was too lazy to put the url tags in...

Edi