PS3: The GPU that will be vs the GPU that could have been

[Acert93]

You're doing it again, Acert

What the? You know, I do not react to a post of yours that may or may not contain an error in this manner. Maybe I should?? Since I am "doing it again" please refer me to a specific post. Thanks.

SPE's are not JUST vector units! They're processors. They can do floating point, they can do integer. They can do any code a general purpose processor can AFAIK. They are not specialised for general purpose, branched code, so take a penalty when running it, but you can't go around saying, like MS are, that Cell offers only a single core for general processing code. That's like saying a P4 can't do vector maths just because it's key strength is general purpose code.

Did I say they could only do FP/vector? If you read my posts carefully I say "FP centric". Some of you complain about long posts, but short hand gets nasty complaints. They are designed to be streaming devices. They are not general processing units as we know them either. They are aimed to be FP friendly and to be used in a streaming array and that is their primary benefit. While they are not limited to vector code they seem to have a lot of limitations that a general processing core does not have.

Does that make you happy?

And btw, your little rant does NOTHING to answer my question you quote. The PS3 CPU (heh, cannot call it CELL according to some ) still only has one PPC core and has a streaming array of "synergustuc processing elements" as the backbone of its performance while the XeCPU has 3 identical PPC cores--so how are we to cut through the BS and look at actual game design and begin to see how these designs work with games as we know them.

I do not mind a correction or a clarification, but your attitude is not helpful. Also, I am not running around "like MS". The problem is not everything MS has said is a lie, yet from just your last post bashing MS PR machine it does not seem to be pretty even (e.g. compare the Jen Hsun vs. KK conflict). They clearly have their motives but you are quick to defend Sony in their case. Something to think about.

 

[Shifty Geezer]

From this thread : http://www.beyond3d.com/forum/viewtopic.php?t=23327&start=40

I do wonder which is the best gamble: ~50% of the FP performance and ~300% of the general processing power or ~200% of the FP performance and ~33% of the general processing power (*Yes, I know these are all peak and multicore CPUs rarely get anywhere near the theoretical jump in performance compared to a single core.)

Is that a fair estimate of FP and General Purpose capabilities though? People talk of Cell's SPE's as being only capable of FP chugging. But they are proper processors, not just VMX units, and they can work as quickly on integers as FPs and can do out of order. They take a big hit with OOO as they lack the specific designs to cope with that tech, but the doesn't make them totally useless. If a SPE is half as good at general purpose as a PPE, Cell has 1+(7x0.5)= 4.5 PPC cores worth. Even if a SPE is only 25% as efficient at genral purpose code, that's 1+(7x0.25)= 2.75 PPC cores worth.

I don't know what the capbilities of SPE's are in regards genral purpose, but from what I can gather people totally overlook that and think of them as only good for FP crunching which is misrepresenting what Cell can and can't do.

No ranting or raving on my part, as I was querying your ideas on Cell general purpose performance.

Why I'm grumpy this time around is there's a lot of rubbish about Cell, mostly from MS's corner (repeated in the Blogcast) which isn't just PR 'bending-the-truth' but is totally wrong and untrue.

This being a smart forum, people can share ideas like we do and get corrected and not go holding wrong impressions. Anyone reading these forums entering with an idea that KZ was realtime or Xenos has 256GB/s bandwidth to eDRAM would be educated otherwise if they read thoroughly. B3D Forums are the last bastion of truth against the raging forces of marketting and PR speak!

Having already pointed out to you in another thread that it's not fair to count SPE's as just FP units, you have repeated this idea flatly, which is spreading the misinformation. Maybe you just missed my point earlier in that different thread?

I haven't answerd your question on processing because I don't know the answer. I just wanted to point out your presentation of Cell's architecture was wrong, as I had mentioned to you previously.

And I'm not slagging SONY off as much because at the moment they're not putting out bald-faced lies. When KK or Harrison starts saying 'Xenos only has 12 GB/s bandwidth available' or 'XeCPU's VMX units aren't true vector processors' then I'll give 'em an earful, but at the moment SONY are using they marketting skills to be derogitary without using untruths. MS at the moment are saying totally tripe, which you've repeated, though perhaps inadvertantly.

And I apologise if I caused offense. Primarily I'm angry at the misinformation, not yourself

 

[Titanio]

e.g. polygon throughput and number of shaders.

Sorry if already picked up on, but...

Polygon throughput should be higher on RSX if only because of the higher clockrate.

Number of shaders..? Not exactly comparable, is it, in terms of units? You've got two different paradigms. The number of shader ops per sec based on publicised figures is actually higher for RSX. For any more detail we'll just have to wait and see.