Sony delays PS3 to 2006, concentrates on PSP !?

[Panajev2001a]

>DX9? If that's true, then why can't a PS2 compare lighting/image quality wise with a DX9 part?

First, clockspeed... overclock the EE to like 450 MHz and watch it kick butt

Second, Why do you bring Image Quality into this ?

The VUs do not do AA or texture filtering, etc...

Second find a DX9 GPU with Vertex Shader as flexible as the VUs in the Emotion Engine... look at even DeanoC comments... VUs are very nicely designed units

Josiah, the Xbox also came out 1 year later AND I like the VUs a little bit more than a 8 registers SSE unit ( with no FP MADD support )... but yes the peak is 3 GFLOPS...

If you wrote a pure T&L bound benchmark that focused on the XCPU and the EE you would see the EE pulling ahead...

Holistically, I think the X-Box 2 will not be signifcantly disadvantage. Many architectures have tried to compete against x86 in the price/performance arena in simillar applications -- consoles are very high performance desktopish these days-- and they got whooped. Not to say Cell will suck, but there will be stiff competition. At the same time it should be noted that Intel places little emphasis on FP performance --in a historicaly sense-- as they should since the Windows work load doesn't demand it often.
Josiah
Posted: Sat May 17, 2003 7:20 am

Intel did not whoop them all on its own merits... as far as RISC opponents Intel fought them off with backward compatibility... not a problem, Cell is not being release in the same market...

SMT is an interesting weapon, but Intel plans in multi core are not comparable to Cell, not yet... Sony, IBM and Toshiba have pulled ahead a little bit as they have explored a way Intel has not...

Intel has different plans and for now they enjoy having different architectures target different markets ( IA-32, XScale and IA-64 )


It is not a problem if PlayStation 3's CPU is a FP powerhouse ( and more )... the Xbox 2 has XGPU 2 for most of that ( graphics wise )

 

[V3]

PS2 is the better system but not completely bettered DC.

Please let the DC to rest in peace

It doesn't matter how good the DC was, it died because it can't stand up to the PS2 hype. Thus it was never good enough to begin with. Its another one of Sega crap, that Sega knew and soon abandoned it. That's just sad, when you're reminded over and over again.

So please let it R.I.P.

 

[chaphack]

Second find a DX9 GPU with Vertex Shader as flexible as the VUs in the Emotion Engine...

i thought faf or someone else, said that while the VUs are as flexible as DX9 VS, but the VUs performance is not comparable to VS?

 

[Gubbi]

It is not at all clear to me that XBox 2 will have an AMD or Intel CPU. All that is required is that the host CPU can emulate x86 binaries at 733MHz P3 speeds. And as Intel themselves has shown, that is not at all impossible (I2 emulating x86 @ 1.5GHz P4 speed).

For all we know the XCPU 2 might be a wide issue RISC with multiple SIMD units, -and multiple cores to boot. With a Transmeta emulation layer.

Cheers
Gubbi

 

[Panajev2001a]

Gubbi... fact is that you cannot design and build such a huge CPU as you are thinking in two years... Neither Intel nor IBM, two of the BEST chip makers in the whole world, could... in only two years...

Money can just buy so much engineering time...

 

[Panajev2001a]

Second find a DX9 GPU with Vertex Shader as flexible as the VUs in the Emotion Engine...

i thought faf or someone else, said that while the VUs are as flexible as DX9 VS, but the VUs performance is not comparable to VS?

It depends, basic transforms are only 1 clock faster ( it takes VS 4 cycles and VU1 5 cycles while VU0 7 cycles ) in the Vertex Shaders and there are still things VS cannot do... like the ability of creating Vertices on the fly with relative ease, for example...

This is from DeanoC's post...

The VU are lovely once you understand their complexities, they are the best geometry shaders available, its fairly trivial to run tesselation programs as well as tradionational vertex programs. Pity the PS2 doesn't have any pixel shaders really.

 

[Saem]

SMT is an interesting weapon, but Intel plans in multi core are not comparable to Cell, not yet... Sony, IBM and Toshiba have pulled ahead a little bit as they have explored a way Intel has not...

I suppose I should have been clear in my posts.

Intel does a fair bit of research in all sorts of technologies, we just don't hear about it.

Currently there are rumours floating about multi-core systems from Intel, the question is what will be the base core. A P4 is no slouch and considering Prescott seems to be bringing a fair bit to the table, IIRC, SSE3 is getting the I2 FMAC or two of them -- Read it on Aces' but I haven't followed it up. In anycase, if this is the case a single Prescott will be quite the beast and a multi-core version will be a significant improvement.

 

[Gubbi]

Gubbi... fact is that you cannot design and build such a huge CPU as you are thinking in two years... Neither Intel nor IBM, two of the BEST chip makers in the whole world, could... in only two years...

Money can just buy so much engineering time...

I can't see why they had to start from scratch.

1.) Strap a couple of SIMD units on Sibytes BCM1250 (MIPS64)
2.) Intrinsity's FastMATH core, modified to support 32bit Floats (MIPS32)
3.) Upscaled Transmeta core (VLIW)
4.) Itanium 2 based, possibly with SIMD units (IA64).
5.) Ev6 core with SIMD units (Alpha, YAY!)

Bolting SIMD unit(s) on an existing core could be done in the timeframe we are discussing.

Cheers
Gubbi

 

[Paul]

i said "Bettered at IQ/textures by DC"

you said "Oh please.. DC cannot compare to ps2 in any way."

i said "give me a full 3D PS2 game running at 60fps at 480p with vibrant/varied/viewtiful SA2 textures"

I have no idea why you think i am saying DC is a better system overall???

You still havent showed me games that can best those PS2 games graphicly.

Maybe MS has the sweetsweet deal? Maybe KCET/Rockstar/Square/Capcom can move on to next gen platform for their next gen dream games? Maybe XBL will be mucho mature?

The developers will go where the money is, and I don't see xbox2 outselling PS3 at their launches.

i have no idea why are you hanging on to that 1tflops thing??? it is just another marketing biz word.

TFLOPS a marketing word? LMAO, so I guess all those 40+TFLOPS super computers are really just a joke then? Please..

PS2 was no revolutionary hardware.

Yes, actually it was when it was announced. It blew the tar out of everything avaliable spec wise, and everything avaliable for some time.

Sure if Xbox2 is before PS3, it might not stack up but it is never going to be that revolutionary destroyed by CELL.

And how are you so certain? With the polygon power ps3 will be able to push, you could use a REYS approach. Polygon wise, PS3 will blow the shit out of every next gen console, just like with PS2.

The PS2 GS was also very impressive when it was announced, and GS3 will obviously make more of an impact than GS did. Can you see Nvidia having something that can do over 10B polygons in two years? Hell they don't even have half a billion yet with their NV35.

You will obviously pull the "ps3 will have no AA and shit texture "

I'm pretty sure Sony isnt going to be stupid when making the GS3 chap..

Of course i knew, but you have to read more carefully to what i said.

And what did you say? This is what you said.

so what if PS3 have better cpu, not like that matters with the PS2. Bettered at IQ/textures by DC, overtaken by a cheaper GC and overpowered by MS XBOX.

I have been saying, the only reason why i am taking shots at Sony PS2 is the underwhelmingness of its hype vs graphics crap they throw around.

And guess what? You fell for it, it's your fault your disapointed with PS2 Chap, I could imagine you were one of those people who were having orgams over ps2's specs and then got dissapointed correct?

It's your fault, grow up, stop crying over spilt milk and get over it. You don't see me disapointed with ps2 do you? When I saw the MGS2 demo on my ps2 long ago, that was what Sony promised in my mind. Those graphics, and they were amazing.

 

[Paul]

IGN has a few more SA movs, you can check those out and take note to the smooth and detailed texturing, something yet seen on the PS2.

Now I suppose your going to say, "oh but those shots have AA on them"

The first rachet and clank had no jaggies and this one won't either.

Smooth enough detailed texturing for ya chap? I can get more examples if you want to look like a fool further.

 

[Phil]

Personally, I think Jak & Daxter looks miles better in the texture department than Ratchet & Clank. Regardless of R&C or J&D though, both games push way more geometry around than SA2.

That's something certain people just don't want to realise. It's easy to make a simplistic game with limited geometry look impressive thanks to textures - once you up the geometry though, texture demands increase rapidly.

 

[Panajev2001a]

Gubbi... fact is that you cannot design and build such a huge CPU as you are thinking in two years... Neither Intel nor IBM, two of the BEST chip makers in the whole world, could... in only two years...

Money can just buy so much engineering time...

I can't see why they had to start from scratch.

1.) Strap a couple of SIMD units on Sibytes BCM1250 (MIPS64)
2.) Intrinsity's FastMATH core, modified to support 32bit Floats (MIPS32)
3.) Upscaled Transmeta core (VLIW)
4.) Itanium 2 based, possibly with SIMD units (IA64).
5.) Ev6 core with SIMD units (Alpha, YAY!)

Bolting SIMD unit(s) on an existing core could be done in the timeframe we are discussing.

Cheers
Gubbi

Sometimes to do the things right you do have to start from scratch...

So basically you are saying that Intel in 2 years could take an IA-64 chip ( what Madison, Merced ??? Well the ex-Alpha guys are already at work on another brand new IA-64 core... who would work at this super IA-64 chip you are talking about ? ) and do what it took IBM something like 7-8 years ? ( Cell base ideas started a long while ago in IBM labs ) including a new OS and programming languages supporting it ?

Remember you need to give the game developers some kind of tools that are a bit close to what you intend to implement else you are giving an advantage to the competitors which will have more games for their launch...

Simply adding SIMD units will give you a giant chip ( expensive ) which will probably have quite bad efficiency and bugs ( see recent McKinley speed bug )...

we will have to agree to disagree...

I also do not understand why would Microsoft take part in such an expensive exercise in futility ( I apologize if I am sounding rude [no offense was meant Gubbi] ) as they do not need a CPU that matches PlayStation 3's Cell CPU... they are not the ones pondering on the micro-polygons road, they are not the ones who need a big and powerful CPU... Xbox 2 will rely on a ultra powerful GPU and a familiar to develop for Intel CPU ( Pentium IV or Pentium V )...

 

[zurich]

Isn't that more or less what SCE did with the Emotion Engine though? Take an off-the-shelf MIPS design and tack 2 uber-vector-engines onto it?

 

[Paul]

I have a question Pana, what are the advantages over a Micro polygon approach as Sony seems to want over the traditional way we are seeing with Xbox/PS2?

I'm at a loss at how this works..

 

[Panajev2001a]

It is another way... some disadvantages that you have to deal with and some advantages like more exact displacement mapping and nice AA and motion blur coming almost very cheap

Another "advantage" is the simplier Rasterizer... it needs to be fast ( fill-rate wise and single texturing is ok... texture filtering should still be implemented ) and with lots of bandwidth...

I will post more later on, there are some very interesting papers you can find by searching google...

The pipeline for a micro-polygons based render works differently from the standard graphics pipeline used in OpenGL or DirectX... it is suitable to a design like PlayStation 3 could be... very fast FP wise CPU with lots of bandwidth feeding a streamlined very fast Rasterizer...

 

[Saem]

Simply adding SIMD units will give you a giant chip ( expensive ) which will probably have quite bad efficiency and bugs ( see recent McKinley speed bug )...

I'm with you in the begining, but the Mckinley speed bug is an edge case and this stuff happens when you're pushing a process to it's limit and using async logic -- I believe this is async data transfer logic which isn't performing quite right in some very rare edge case.

 

[Panajev2001a]

You are right about McKinley, but my point can relate to it...

Such a Cell killer would have to seriously push Intel's new 65 nm process ( they will get to it too ) and they will have less time than Sony, IBM and Toshiba to test it and discover similar bugs...

Shorter development cycle also means less time for testing and debugging...

 

[Saem]

I understand the issues when it comes to validating and testing.

I think one thing should be kept in mind that the Cell implementation probably won't have the huge amount of custom work that Mckinely would have and if Intel would be designing something to compete against it they won't have to go to as much design effort as they do, so it'll be a fair bit easier for them to work on such a MPU.

In anycase, I don't think there will be a custom part in the X-Box, actually, I suspect that Intel will put out a mildly modified Prescott. Maybe a bit more FP power.

Additionally, I'm not convinced that the PS3's Cell chip will hit particularly high clock rates in comparision to what Intel could produce. I know this isn't particularly sound, but if we take the PS2 example, Intel should have near double the clock rate and I don't think that's far off base.

 

[Paul]

Then that means Xbox2's clock speed would be around 6-8Ghz..

 

[zidane1strife]

PS1 was no revolutionary hardware.

Some game had dozens of characters on-screen at 60fps, and the main character was double textured....

Others featured 640i 60fps gameplay....

The char models in some games was comparable to many of the low-quality models for ps2/dc/xbox/gcn...

It had the best sound, and video playback of the last gen...

PS2 was no revolutionary hardware.

... I think old george would beg to differ....

Anyway, the ps2 was indeed quite nice for its time...

It caused a mad panic, and killed a console.... before it was even released...

The h/w featured very nice flexibility, and some unsurpassed to this day gpu vram bandwith, and also had good fillrate....

There's no official information saying the CELL going into PS3 is going to be a 1 TFLOPS configuration

Indeed, at 65nm the h/w could run significantly faster than what many have thought. I doubt it will be limited to 1Tflops, I'd say 3+Tflops or beyond is very likely...

ps so the nv35 is only about 5x the ps2 raw performance.... just 100-200x(depending on how much is delivered...) more to go nvidia...