Actually the architecturally very different and their strengths lie in different areas. To get the best out of each takes a lot of work and special consideration... and time. They may be able to put up similar results in many situations, but how this would be obtained is quite differently. Of course right now both are being used minimally in regards to featureset...

It takes 24-28 months to build a game from the ground up, even longer if you do your own engine and tools. Xenos has been in dev hands almost 10 months now. Looking at the core features we see many have not been touched: Hardware tesselation, vertex texturing, SM3.0, etc. Up until a couple months ago there were some API issues with tiling. It is a totally new hardware paradigm and devs are still struggling to get SM2.0-only games out. The first SM2.0 ONLY game is Oblivion. SM3.0 is minimally exploited at this time.

--

As for the interview nAo has never been really hot on Xenos even before he worked with Ninja Theory (he has not worked with the 360). It just does not line up with his vision of a game engine.

That said, note that his comparisons are all, "We did this to alleviate a bottleneck on RSX" -- yet nowhere in the discussion is the 360 given equal credit. i.e. "Ok, you spent 12 months tweaking the PS3 -- what would happen if you spent the same 12 months on the 360 doing the same thing?"

And that really is an issue IMO. With the PS3 due to the segmented memory and the bandwidth constraints devs have to do a lot of extra work on things that "just work" on Xenos. e.g. With the fillrate a developer could move their alpha blends of certain objects to CELL. Ok, but now you are taking CPU performance away from the game engine to make up for the GPU AND it requires more work. Ditto HDR. NOA32 takes shader cycles. It does not appear overly expensive, but again you are talking extra time and taking performance to do it.

Ninja Theory is a 2nd party being funded by Sony for exclusive PS3 content. They are not working on any 360 titles and they have blogged a number of confrontational posts in the past about how Xbox kung fu game sold poorly and was critically chastized was the result of MS's spinelessness, etc... That does not mean they are wrong, but it does mean they are just one opinion and should be treated as such.

I personally tend to consider what multi-platform devs say in regards to comparisons, and in regards to each system and how it is being used in exclusive content I listen to exclusive devs about their OWN system. The comments from Starbreeze, id Software, Valve, etc... are all multi-platform and tell us a lot. The CoD2 gushed on the 360, NT gushes on the PS3. It is what you would expect.

Back to the article, some of the comments are really vague. e.g. He would not give reasons why (i.e. NDA) he thought certain things. At other points, like the memory architecture, it was vague.

But as I was recently pointing out, numbers tend to lie. e.g. Fillrate

RSX 4.4B pixels/s
Xenos 4.0B pixels/s

Here is the catch: That number is cut in half when RSX enables 4xMSAA. Xenos remains the same (i.e. no logic penalty). Further, Xenos ALWAYS has that fillrate because it has 256GB/s of memory bandwidth for pixel fill. It NEVER is delayed.

RSX is not so clean. If ALL the GDDR3 is dedicated to fillrate it tops out at 2.8B pixel/s. And that leaves no memory bandwidth for the framebuffer, backbuffers (which have high bandwidth needs), texturing, geometry, etc.

Xenos? It has enough bandwidth for full fillrate with 4xMSAA, FP10 (HDR), and so forth. Further, Xenos has a huge Z fill because unlike any GPU on the market it can do 2 Z samples per clock, even with AA enabled (with 4xMSAA it does 64 z samples a clock).

This same type of scenario applies to texturing (e.g. Xenos' TMUs are totally decoupled and, unlike RSX, do not require PS ALUs being allocated to texture), flops (Xenos not only load balances but with an array of 'small' ALUs is very fine grained for very high utilization and Xenos is also very threaded), shader model (both are SM3.0, but Xenos batch size is 64 whereas RSX is over 1000 meaning Xenos is a lot faster at pixel shader dynamix branching), etc

It is easy to say: Xenos performs 217GFLOPs and RSX is 255GFLOPs, so they are close with an edge to RSX. But that totally ignores architecture (bandwidth, utilization, featureset, bottlenecks) and ease of use. Xenos is clearly designed for the console space and console problems. it also resolves some big problems natively; e.g. FP16 for HDR is 2x as expensive as Int8 and uses more bandwidth. Xenos has FP10 so you get the benefits of FP10 with the expense of Int8. No work arounds and time invested, no extra costs. It just works. All things even FP10 saves a lot of resources over FP16 which means in theory a FP10-system could need less raw power to get the same end result. Featureset is pretty important in that regard.

Of course the article never touches on how such bandwidth could be leveraged for the Xbox 360--and why would it? PLAYSTATION site, PLAYSTATION developer.

It does not mean it is unfair, it just means it is only one angle and opinion.

Of course if you asked a 360 dev they would have their own little tricks. Not to mention it has a lot of features (curved surfaces, tesselation, vertex texturing, coherant memory reads and writes, fine-grained SM3.0 that is much, much faster at pixel dynamic branching... in the order of 4-10x faster in some DB bottlenecked situations).

So take what any dev says in context and with a grain of salt. These interviews are good PR. MS does the same thing. But there is always more to the story.

What I can say is that it is nice to see PS3 overcoming some of the larger hurdles RSX presents. I look forward to tesselation and displacement mapping work arounds running on CELL as well. Hopefully Sony incorperates NOA32, Tesselation, DM, etc... into the SDK and allows these things to be accessed through an API as "featuresets" if they wish. Making next-gen features standard and easily accessible is important and with the current state of Sony tools devs could use the help.