x86-64 a viable choice for Xbox2?

x86 has R&D like you wouldn't believe (Why do you think other architectures struggle against x86?) and not just in hardware, but in the software end as well.

Intel's new version of Hyper Threading will be available in the Prescott. Chances are the FPU will be improved, especially the SSE2 units throughput (I suspect it'll be 1 instruction per cycle rather than every other). The CPU will boast a lot of bandwidth and cache. Not to mention Intel's process technology is is about as good as it gets.

Well dude that is nice... but as i've said MS is likely to buy a cheap outdated downgraded intel cpu, if the xbox is anything to go by... I'm sure that with equal R&D into a gaming centered processor intel could very well equal Ibm... but alas they haven't started such a project AFAIK, thus if they were to start a yr from now, they'd have to be pretty good to come up with a significantly superior product with far less R&D time... and still i believe that a cpu and gpu designed to work in conjuction are likelier to work together better, than cpu's and gpu's done separately.

never said XB 2 would be a single chip design, in fact I argued that it will NOT be. Does Sony make their own chips? I thought Toshiba and NEC took care of that. Anyway, Intel and AMD are king of the hill in process technology, which is also reflected in performance figures:

Neither did I. I just mean that already HALF of the important components a.k.a. cpu, aren't likely to be up to par transistor budget wise... just telling u where I believe the Sony transistor budget advantage could come from...

So it boils down to the price/performance of the GPU and I don't see the Japanese foundries being much ahead of the Taiwanese.

And last: While Sony is a big company I seriously doubt they have deeper pockets than Microsoft.

Those taiwanese sure are having fun with their geforce :LOL: .... anyway AFAIK IBM and Intel are above any Taiwanese cheese, and with sony all cuddled up with IBM tech... yes IBM powered japanese foundries, are better than u'r average Taiwanese ones(heck, they're having trouble now.... let's not even begin to imagine what nasty little horrors await in the future.)

As for pockets and green dough.... As i've said Sony invested billions into the ps2, and now with two successes in a row, they're likely to invest even more... MS can't just throw 10B$ out of their piggy bank for no reason, without seriously hurting their stock... So MS isn't in another league or anything like that... their value falls low enough.... and things could get nasty.

... have any of you seen a sample CELL running anything? Do they even EXIST yet?


The confidence of the sony big heads in their many ps3 speechs says it all... I mean they obviously know how the thing's progressing, so it's obviously going the way they intended... Surely if they were getting some cr@ppy perf, or came upon an obstacle they couldn't overcome they wouldn't be that confident...
 
Uhm... to all the people talking about 1-billion transistor chips and such...

WTF? Do you have ANY IDEA how abysmal the yields will be with that many transistors? Forget about it! The potential losses Sony would incur are awful.

Even on 90nm, 1 billion transistors is INSANE. Look at NV30, that's "only" 125 Million (one eighth of one billion), and it's already pretty big at 130nm.

Well we all know Nvidia will likely be forced to use their so called 300M transistors figure for their 2005 gpu... or the taiwanese will crack...


As for Ps3 it isn't coming out till 2005, and it has IBM backing it, so like archie i expect they use 65nm<.... U know those are the benefits of not being encumbered by taiwanese wannabes...

PS: i've heard intel is throwing 3times gates, or something to make up for leaks at 65nm and below thus wasting more space, while IBM has managed to succeed with only 2times something... any truth to this?(yeah, heard it a while ago, don't remember where, no clue....)

Actually both are publicly traded companies, so it's really not their money to burn, but their shareholders... Shareholders usually don't spending wars (especially since Microsoft stock doesn't pay dividends)...

Yes, but as many have said on this and many other boards... sony money counts as investments since it's mostly internal stuff... While to MS its losses since it goes to nvidia and intel,etc... or so i've heard.
 
BenSkywalker said:
Over a factor of four, where exactly do you get this figure from?

In early 1999, the Graphic Synthesizer was shown with 40-odd million transistors. The nVidia equaivalent at this time was the TNT2U with just over 10M tranistsors.

In 2001, the I-32 GS was shown with 280-odd Million transistors. The nVidia equivalent was the Nv20 with just over 60M tranistsors.

Sony being behind the curve in fabrication technology then they are four times ahead of the chip that nVidia released in prior years....?

Um, wake me up when another IHV can get eDRAM to work commerically, ok? Seriously though, they had problems at launch, but the GS is being produced at 0.13um , isn't it? I thought Simplex put out a PR a few months ago... nVidia is producing [or lack there of] at... I seem to forget. ;)

Would the relatively miniscule amount of transistors it would have taken to add S3TC been worth losing a couple KBs of eDRAM? What about adding in some decent useable texture filtering? How about using a 8x1 or 4x4 pixel pipe arrangement instead of a 16x0? What would they have done is a good question, maybe 6GTexels raw fill while still using CLUT and bilinear?

It was a 3D philisohical choice made based on time to market needs [Simplex has a nice presentation on this and IP reuse] and their idea of the future of 3D graphics.

16*0? Um.. alrighty then.

What to know whats even funnier? Even it it was 16*0 - thats a remakable modern and contempoary approach when you think of it [not talking of the features -blends, DOT3- that are MIA] 16*0 would be a effecient usage of tranistor counts: Very computationally intensive for shaders with much flexiblility in what was tradiationally the 'fixed pipeline.' Use pipe combining and loopbacks. Would be an awesome architecture for using Stencil.

4*4? Are you serious? What a waste of transistors, welcome to the failed ideas of last generation. The whole industry is moving towards the ideal of going heavy on the pixel computational ability and less on the idea of a fixed "TCU." I'm going to assume your joking.
 
How about using a 8x1 or 4x4 pixel pipe arrangement instead of a 16x0?

"16x0? You lost me here..."

Faf, when he said 16x0, I think he means the PS2's Graphics Synthesizer has 16 pixel pipes and 0 TMUs per pipe. Of course you know that it takes a full extra pass to do texture mapping on GS. something like 8 pipes to do textures.
 
Is it really that insane? The I-32 was over 280million transistors at .18um... A Power4 in the same ballpark at .13um (and that's 2 cores, 3 L2s, and chip-interconnect logic). An IC with a billion transistors at 90nm really doesn't seem all that far-fetched... Plus you have to keep in mind that 90nm is right around the corner (Intel, IBM and Toshiba will be sampling devices at various stages on that design rule early next year), you have to start thinking about 70nm and 65-55nm...


agreed. The I-32 (that is, GS with 32MB eDRAM for those that wonder what is being talked about) was out in 2001 (i think) with over 280M transistors. by 2005-2006, I don't see why Sony wouldn't have a 1B+ transistor GS3, with many more pixel pipelines, features, and more than 32MB eDRAM.


So what process will the GS3 be on? .09, .07, .06? - Just depends on the timeframe of PS3.
 
my guess on GS3: 32-64 pipes, DX9 equivalent features, 64MB or more eDRAM, multi-TeraByte (at least 1 TB) bandwidth, 1 billion+ transistors.


then again, Sony might increase the number of rendering pipelines by the same amount they increased from PSX==>PS2. that is, 16x. ok, ok, so 128 pipelines sounds completely and totally insane. but then again, so did 16 pipelines in March 1999 when PS2 specs were annouced.
 
The nvgpu for 2005 i've heard is to be 300M transistor

curious as to where you got that from.

so that would be like the NV50.

if XB2 uses a variant of NV50 or NV55, it could be over 400M transistors if you're right about Nvidia's 2005 GPU (the basic one for PC cards) being 300M.


someone also mentioned that XB2 might not be a single chip (meaning the GPU) design. I wonder if perhaps Nvidia/MS will go for a multi GPU approach to XB2's graphics. I predict they will, because they might have to in the face of a monster EE3/GS3 rendering/fillrate/poly rate combination in PS3.
 
Mega, If they used that many pipelines they would have to cut corners again. So no floating point pixel shaders with looping constructs in that case.

Marco

PS. floating point is overrated though, Id rather see them go with block floating point :) (That means you use semi fixed point, you change the decimal point programmatically when necessary.) I wonder if you could get much die area gain from that.
 
GS was only demonstrated in 1999. Nvidia had GF available at the end of that year.

GS32 has never gone into mass production AFAIK - at least I've never seen any commercially available hardware on it. I believe that the yields had to be terrible.

Also, even 32MB of embedded memory cannot compensate for lack of features. A GS32 could never produce the kind of visuals that a GF FX can - it can only push lots of cartoon-like polygons. Sony should better increase programmability and add floating point accuracy to their graphics subsystem if they want to compete.

As an artist, I'd definitely prefer to work on DX9 hardware as I would have a lot more possibilities to maximize the output's quality. You see, most of the guys you need to do cinematic quality content alrady have experience that you don't want to dismiss - and their approach is quite close to the direction that PC 3D hardware is following...
 
"Mega, If they used that many pipelines they would have to cut corners again. So no floating point pixel shaders with looping constructs in that case"

Ok, that's why I said Sony might use as few as 32 pipelines. still highly parallel, just not as many as 64 or 128 pipes. I don't think Sony will decrease the number of pipes, or keep it the same as GS. I think Sony can have modern floating point pipelines while still being very parallel. but not with a rediculas number of pipes like 64-128.

By 2005, Nvidia and ATI will no doubt be at 16 pipes, at least.
 
"Also, even 32MB of embedded memory cannot compensate for lack of features. A GS32 could never produce the kind of visuals that a GF FX can - it can only push lots of cartoon-like polygons. Sony should better increase programmability and add floating point accuracy to their graphics subsystem if they want to compete."

I agree there.

I think Sony will have DX9 features in GS3 while being much more parallel... more performance, more bandwidth, more fillrate than PC GPUs. While MS/Nvidia will have DX10 or DX11 features in XB2.
 
OK I give in what feature did V2 have that TNT2 didn't?
Well, from what I recall TNT series didn't even have basic Clut support - smallest texture format was 16bit. Considering how much people like to bitch about compression around here... I'd think that would be relevant :\

Faf, when he said 16x0, I think he means the PS2's Graphics Synthesizer has 16 pixel pipes and 0 TMUs per pipe. Of course you know that it takes a full extra pass to do texture mapping on GS.
It doesn't. 8pipes texture, it's single pass.
 
Well, from what I recall TNT series didn't even have basic Clut support - smallest texture format was 16bit. Considering how much people like to bitch about compression around here... I'd think that would be relevant :\

OK fair enough I'd forgotten about that.
But it also wasn't limited to 256x256 textures, supported 32 bit color, and had a much better color combiner implementation. I'd hardly call it sub V2 in terms of features.... V2 was fast for the time especially the SLI variant, but it wasn't what you'd call a featureset monster.
I'm just being pedantic here....
 
Faf-

Well, from what I recall TNT series didn't even have basic Clut support - smallest texture format was 16bit. Considering how much people like to bitch about compression around here... I'd think that would be relevant

The TNT2 supported 32MB on board and AGP texturing both of which the V3 lacked, if the PS2 has twice the RAM the XBox had(with comparable bandwith) the lack of texture compression wouldn't be that big of a deal. I think it's a big deal on the console side of things, I almost always shut it off on my PC when it is available(although some games make it non viable as I only have a 64MB gfx card). I did mean 16 pixel pipes with 0 TMUs btw.

Vince-

In early 1999, the Graphic Synthesizer was shown with 40-odd million transistors. The nVidia equaivalent at this time was the TNT2U with just over 10M tranistsors.

The GeForce was being mass produced in 1999, that was a 23Million transistor part.

In 2001, the I-32 GS was shown with 280-odd Million transistors. The nVidia equivalent was the Nv20 with just over 60M tranistsors.

At ~450mm it was never viable for a consumer part nor was it ever mass produced.

4*4? Are you serious? What a waste of transistors, welcome to the failed ideas of last generation. The whole industry is moving towards the ideal of going heavy on the pixel computational ability and less on the idea of a fixed "TCU." I'm going to assume your joking.

Why would it have to be fixed function to be 4x4? The DX9 rasterizers support sixteen texture layers in a single pass, even then a lot of people aren't too pleased the NV30 doesn't have at least 2TMUs per pipe.
 
But it also wasn't limited to 256x256 textures, supported 32 bit color, and had a much better color combiner implementation. I'd hardly call it sub V2 in terms of features.... V2 was fast for the time especially the SLI variant, but it wasn't what you'd call a featureset monster.
Damn texture size limitation slipped my mind :p But anyway, IMO until GF series, the 32bit color in NVidia cards was mainly a gimmick given what performance implications for it were(It might have helped if they allowed use of 16bit Z with it).
I know I exagerated there (it was on purpose after all ;)), but didn't think of TNT series as a particularly strong featureset either - their strongest feature in the end was speed as well...

if the PS2 has twice the RAM the XBox had(with comparable bandwith) the lack of texture compression wouldn't be that big of a deal.
If by "not big of a deal" you mean "it'd be better off then any competition" then yeah, sure - I am quite confident that I can stash more content into a 128mb DTL10k then a 64mb Box right away, without any bandwith change.

I did mean 16 pixel pipes with 0 TMUs btw.
And as I've said the actual configuration used is different.
Not to mention that 0 TMU would imply you have no texel fetch at all, so texturing should be impossible. ;)
 
BenSkywalker;

Ok, I give you it - I admit when I'm wrong. NV10 was ready for XMas of '99 in mass.... good call. So their advantage is 3X on average.

Concerning I-32 GS, I remember PixelFusion having over 100M on a 0.25um process and claiming costs in the hundreds after a year or so of production. Do some cost and die shrinkage normalizing and tell me if it's feasible.

Why would it have to be fixed function to be 4x4?

Really doesn't matter as a 4*4 architecture is inheriently imbalanced and totally wrong for anyform of advanced DX8+ raster function.

Even then a lot of people aren't too pleased the NV30 doesn't have at least 2TMUs per pipe.

Alot of people are stupid and look at nothing big which has the bigger numbers and nomenclature. As the people who bitch and they'll describe the 3D pipeline like it was in the DX5/6 generations. They have no idea other than two "TMUs" are better than one - Hell, by right their not even called a TMU... TCU is a bit better.

The NV30 and R300 are well balanced architectures - heavy in pixel computation for the shaders and extremely flexibility in the texturing, all while remaining within bandwith constraints. Now something like the Matrox G800 is nothing but crap - the sooner that type of architecture dies the better.

my guess on GS3: 32-64 pipes

Actually, this is interesting. The GSCube has 64 pipelines IIRC - GSCube in a chip would be quite interesting indeed.

Mega, If they used that many pipelines they would have to cut corners again.

Hmm... just thinking and typing at the same time, but didn't the GSCube render FF:TSW - and not a hacked to shit remake? Didn't Faf say onetime that SCE's engineering were keen on the idea of rendering in layers like they did on the Cube? Just speculation

EDITed out a [/quote] -> [/quote] error
 
Actually, this is interesting. The GSCube has 64 pipelines IIRC - GSCube in a chip would be quite interesting indeed.


no... GSCube (2nd version) has 64 sets of EE+GS. that's 64*16 pipes.
each GS has 16 pipes. -that's 1024 pixel pipes in total.
2400M pixels/sec *64. half for textured. 4 billion+ polys/sec.

The first version of GSCube had only 16 sets of EE+GS. so that's 256 pixel pipelines. 2400M pixels/sec * 16. half for textured. 1 billion+ polys/sec.


The PS3 (EE3+GS3) should have more polygon rendering performance than the 2nd version of GSCube...(PS2 * 64 - 4Bpps), but I doubt it will have the same fill rate as the 64-way GSCube (2400M/pixels * 64) but the pixels on PS3 will no doubt be much better in terms of features than the simple pipes of PS2/GSCube.

Even with the 2nd GSCube's massively parallel design, 64 sets of EE+GS, I doubt they got anything near peak performance. there must be horrible inefficiencies. I'm sure GS3+EE3 will be a much much better solution for rendering realistic images than 64 sets of GS+EE.
 
GS32 has never gone into mass production AFAIK - at least I've never seen any commercially available hardware on it. I believe that the yields had to be terrible.

They were never meant for mass commercial production. A lot of it was just experimentation and demonstrate their .18um prowess of the Nagasaki fabs (after they had gotten the .25 to .18um migration bugs ironed out)... The point was not that it wasn't targetting consumer use, but that ultra large scale integration of that magnitude was and is possible...

The DX9 rasterizers support sixteen texture layers in a single pass, even then a lot of people aren't too pleased the NV30 doesn't have at least 2TMUs per pipe.

Probably because they don't realize that staged arrays of traditional TMU's (or TCU's if you prefer) a bit of a dinosaur of parameter-based combiner type renderers that we've been using up 'till now...

Comparing the GS to a Voodoo2 or TNT is rather silly though as DRAM bandwidth and latency differences are in orders of the magnitude. Having single-cycle or less access to any address in your frame-buffer *does* change your perceptions on how approach certain tasks...
 
Back
Top