Xbox2 graphics 10 times higher geometry perf. than X800 XT ?

[Dave Glue]

I will tell you this.. If MS releases a XBOX 2 that is a fully functioning PC, I would buy it in a heartbeat and join the darkside!!

Ideally, MS could offer OEM's a reference design where they could make their own Xenon/PC hybrids. Perhaps a year after the Xenon debuts so that the hardware prices have dropped to more reasonable levels.

The difficulty in that though is security - preventing piracy would basically mean that the Xenon internals would be completely walled off from the rest of PC, which would be a waste considering the graphics and CPU power you would have sitting untapped for most PC apps. But hey, I don't have the technical background to really explore this.

Doubt it will happen, but a man can dream.

 

[Doomtrooper]

Consoles offer much better gaming experiences over PC gaming now, when I played in ladder matches for UT the number one complaint was reliable voice communication for games, and less cheating. UT 2004 was hacked within weeks with ping hacks that allowed users to run 10x faster than any other player, see through walls hacks, aim bots.

Much harder to hack a console game and voice communication is flawless for what I've seen, so if you get your ass handed to you one on one on a console, you can rest assured the player you were facing was simply better vs. better at writing scipts to ehance their lame ass skill on PC games.

Although I have not made the switch yet, I am preparing to move to console gaming with X-box 2. PC games sales will suffer, as from my experience almost every Clan member I've played with or against doesn't use the PC hardly at all for gaming anymore.

 

[Dave Glue]

Consoles offer much better gaming experiences over PC gaming now,

Yep, that's how I judge PC games - solely by the early days of voice communication and unregulated public servers. Graphics/modability/control - screw 'em. Damn, Half Life 2 sure is gong to suck then when I pick up my copy today.

Anyone actually want to intelligently comment on the future PC->Xenon situation, without this turning into another typical fucking console vs. PC thread?

 

[jvd]

You have to wonder how this would affect ATI's high end sales, it seems like they're helping to produce a product that would effectively kill the demand for their PC cards (aside from the professional workstation cards).

Well . Current rumors have the system coming out x mass 2005 . So most likely a sept - nov launch.

THen u figure we will have 512 meg cards in early 2005 in the desktop area.

This is with the fastest ram they can put on it with a 256 bit memory bus .


THen you figure the xenon/xbox2 will have 512 megs of ram total shared with the cpu and whatever else needs it.

Next you take into account the space and cooling ability of the console. Then the space and cooling ability of the pc .


Basicly whatever they put in the xenon they can then put in the pc with more ram / faster ram and clock it higher. Not to mention 6 months down the line they will refresh it to clock even higher and then another 6 months down the line they will replace it with new tech or an even more enhanced version of the core.

That is why they don't have to worry about pc high end sales. IF anything having an ati card in the xenon can only help ati. All the xenon games will be developed with that cards shader abilitys and other abilitys like 3dc . So all ports should have 3dc support and other ati hardware feature support.

IF ati was smart they'd put 3dc and whatever other thing they want to push foward in that chip.

 

[Dave Glue]

Sure, high-end PC GPU's will still be around for a while - but the thing is with native HDTV res, the PC loses one of its largest advantages - resolution. The PC is still the only place to go for certain games, my point is that with the res and ram of these new consoles, that will likely no longer be the case. ATI can make them - the question is, will anyone buy them?

A further questionmark is the CPU - that's one area where PC's have always dominated, even at the consoles launch. With the slow growth of CPU speeds in the past 2 years for the PC, this may not be the case. Right now, a tri-core 3.5ghz CPU sounds somewhat implausible, but if it's streamlined with the purpose of pushing polygons and little else, it may be for real. The PC will be undergoing a difficult transition to multi-core CPU's, and still have millions of machines out there with single CPU's - that's an area where a fixed platform may really have an advantage.

 

[jvd]

Sure, high-end PC GPU's will still be around for a while - but the thing is with native HDTV res, the PC loses one of its largest advantages - resolution. The PC is still the only place to go for certain games, my point is that with the res and ram of these new consoles, that will likely no longer be the case. ATI can make them - the question is, will anyone buy them?

Yes hdtvs make the res for the consoles nicer. That is true and lots of people like big screens.

But the same is true for the pc. People can now hook up a 60 inch hdtv to thier pc . Not only that but because of the high end nature of the pc you can turn on fsaa and aniso. Not to mention browse the web , do reports, mod games and other things consoles can't do.

Also the pc will allways be the bleeding edges playground.

Even if it takes a year and a half to two years for the pc to completely surpass the next gen systems . once it suprases them it will just keep on putting distance on the graphics between them till a new console comes out. Not to mention that many large console publishers like namco and sega are moving to the pc. Namco claims that they will loose millions in the transition between the ps2 to the ps3 . (this was in an article btw , i believe pc gamer or pc stratgy or something like that last month) Consoles will allways have that transition. Not to mention the fact that you can get away with using last years engine on the pc by just pumping up polygon counts and effect usage. With a console a ps2 engine just isn't going to cut it on the ps3. Where as a suped up half life2 will look just fine compared to first gen or second gen xenon tittles.

A further questionmark is the CPU

I dunno about that. I think with the dreamcast that was offical over with. Mabye even the nintendo 64.

But i know the dreamcasts cpu which came out in 1998 the sh-4 did 1.4 gflops i believe. Which is certianly more than the p3 400-500s that were avalible then .

Just like the ps2 emotion engine was much more powerfull than the pcs in 99 and some might say are still more powerfully in graphical apps .

Its the ram and video cards that will give pcs the edge. there will be another jump soon where 512 meg dimms are the norm where dells and gateways sell that on the low end models and with the average gamer at 1.5 gigs of ram.

The games demand it and with intel ready to put out a x86 -64 bit chip like tha a64s pcs will again be able to inch higher stll .

 

[Nappe1]

Anyone thought about it having a Geometry shader (means continuos tesselation with dynamic mesh LODing ála Matrox Displacement Mapping.)

this would easily get the 10 times more geometry and without too much CPU overhead. But then, there should be enough more vertex shader power to handle extra vertices that Geometry Shader generates before that. Manufacturing processes are developing, so is this all possible on reasonable heat/size/price budget, is question that remains to be seen.

 

[Megadrive1988]

The shader core of the R500 was reported to have 48 Arithmetic Logic Units (ALUs) that can execute 64 simultaneous threads on groups of 64 vertices or pixels. ALUs are automatically and dynamically assigned to either pixel or vertex processing depending on load.

okay I suppose for now, this answers my question. '64 vertices' whereas R420 / Radeon X800 can only work on 6 vertices per clock because it has 6 vertex shaders and the 16 pixel pipes cannot work on vertices because X800 lacks a unified shader pool.

please correct me if I have that wrong.


btw, I also accept the idea that R500 / Xbox2 can reach 10x more usable geometry than R480 / Radeon X800, which was mentioned on the first page.

I am also discounting any help from the CPU(s) because the original quoted statement was only talking about the R500, not the whole Xbox2.

 

[MfA]

48 shader units with 64 thread contexts each for vertical MT is semi believeable.

 

[DegustatoR]

Anyone thought about it having a Geometry shader (means continuos tesselation with dynamic mesh LODing ála Matrox Displacement Mapping.)

this would easily get the 10 times more geometry and without too much CPU overhead. But then, there should be enough more vertex shader power to handle extra vertices that Geometry Shader generates before that. Manufacturing processes are developing, so is this all possible on reasonable heat/size/price budget, is question that remains to be seen.

Is the only problem is CPU overhead? I thought that you still have to process all these vertices via regular vertex shader, the only difference is that vertices could now be created by the GPU too, no? It won't give you hundreds of millions of polys for "free", right?..

 

[eSa]

The shader core of the R500 was reported to have 48 Arithmetic Logic Units (ALUs) that can execute 64 simultaneous threads on groups of 64 vertices or pixels. ALUs are automatically and dynamically assigned to either pixel or vertex processing depending on load.

okay I suppose for now, this answers my question. '64 vertices' whereas R420 / Radeon X800 can only work on 6 vertices per clock because it has 6 vertex shaders and the 16 pixel pipes cannot work on vertices because X800 lacks a unified shader pool.

please correct me if I have that wrong.

Yes, the first quote tells where that projected 10x and 4x performance increases comes from. But do note, that there are max 48 units WORKING with 64 vertices at the time. That is, the throughtput is still just 48 / 4 = 12 fully transformed vertices per cycle.

In a sense it is very logical, because all these "calculation units" will have memory address unit with SIMD unit and scalar unit, one will also be able to access memory in vertex shaders. Just like with pixel shaders can now with current generation.

Apart from EDRAM, that Xenon GPU is kinda boring really. Well fully 32-bit IEEE FPU and so on... But still very much what was expected.


Edit: Ah, sorry so the rumour is that it is 48 "calculation units", not 64.

 

[ERP]

That is, the throughtput is still just 48 / 4 = 12 fully transformed vertices per cycle.

To put that number in perspective, at 500Mhz that would be the quivalent of 6Billion verts/second. To me that doesn't deserve qualifiers like "just".

 

[DukenukemX]

What I don't get is why does the Xbox2 need 3 PowerPC G5 processors? It's as if Microsoft plans to emulate the P3 733 CPU for backwards compatibility.

That is if the lastest rumor says that the Xbox2 is going to have 3 G5 chips.

 

[Inane_Dork]

To put that number in perspective, at 500Mhz that would be the quivalent of 6Billion verts/second. To me that doesn't deserve qualifiers like "just".

Good point. I did a quick calculation, and I don't think that will be the bottleneck. 720p, 4x AA and 2x overdraw with one polygon per fragment still got about 100 fps.

 

[kyetech]

To put that number in perspective, at 500Mhz that would be the quivalent of 6Billion verts/second. To me that doesn't deserve qualifiers like "just".

Good point. I did a quick calculation, and I don't think that will be the bottleneck. 720p, 4x AA and 2x overdraw with one polygon per fragment still got about 100 fps.

but if the architecture is 64 pipe uniform. then it means if you dedicated all the pipes to geometry you would have none left for pixels.

or if you dedicated all to pixels youd have no geometry. So all things being even. it would end up looking more like 21 pixel pipes / 21 texture pipes / 22 vertex pipes.

edit: you could call this card a 21x1 pipeline architecture. with 22 vertex engines !

its more like 2 billion vertices.

and more like 1.5 times the pixel power of x800

and 1.5 times the power of texel rate

????

dunno if thats right. I might be talking shit ??

 

[Mintmaster]

or if you dedicated all to pixels youd have no geometry. So all things being even. it would end up looking more like 21 pixel pipes / 21 texture pipes / 22 vertex pipes.

edit: you could call this card a 21x1 pipeline architecture. with 22 vertex engines !

its more like 2 billion vertices.

and more like 1.5 times the pixel power of x800

and 1.5 times the power of texel rate

????

dunno if thats right. I might be talking shit ??

Not talking shit, but you're not considering realistic loads. The X800 would rarely be using all it's power at once. In areas of dense geometry, the vertex pipes would be the bottleneck and the pixel pipes are unfed. R500 could use just a few units for pixel processing and the rest for vertex processing, and you'd get a big improvement.

The same goes for large polygons like on walls or floors. The X800 would have it's vertex units mostly sitting idle, whereas R500 could use nearly all its units for pixel shading. Then it'd be much more than 1.5 times. I would expect this case to happen quite often. Even now we can see that vertex power is rarely the limiting factor, so those transistors have low utilization.

By the way, does texturing take up a unit? I would think they'd do it similar to their past architectures, where a texture lookup is parallel. Actually, what seems more realistic is if there are a limited number of texturing units (say 16 or 24), and they are separate from the 48 ALU's. Not actually separate, but don't impede arithmetic calculation ability.

The final thing to consider is that I doubt R500 will be able to do triangle set up at a rate of 12 per clock. The vastly increased geometry power will mostly come into play during vertex shading rather than simple transformation.

 

[kyetech]

do agree with u.

my disclaimer was 'all things being equal' when I know in real life they wont be.m

any how would make sense for the texturing to be seperate though i confess I dont know enough about it. What I will say is that the 10x and 4x statement is pure hyperbole. but even so, a chip thats about 2 - 2.5 times the power of an X800 with a nice piece of edram will be awsome.

I cant wait till Jan 5th, to see just how powerful this thing is gona be.

 

[Megadrive1988]

thanks for all the replies so far, got more info than i expected. seems i may not be too far off thinking what i did.

 

[Arun]

The "48 ALUs for 64 pipelines" approach is interesting. It definitively has a bunch of advantages, if the implementation is done correctly, including:
- Better branching prediction; you can basically add latency to giving the next instruction as you see fit in such a system, which can dramatically reduce bad branch prediction if the controller logic is smart enough.
- No more loss - or at least a much lower loss - of pixel power for very small triangles. Since I assume XBox2 devs will be very aggressive on the polygon count, this is relatively important, much more so than in the current PC space.
- It helps to sanitize AA and potentially alpha testing. Although I kinda doubt ATI would bother about that last one.

And of course, it also helps for texturing in the VS, which was subpar in a few POVs in NVIDIA's implementation of SM3.0. - I'm sure there are other advantages, and perhaps disadvantages (besides the transistor overhead), but I can't think of much else for now

Uttar

 

[davepermen]

having today 16 + 6 shader units (or so?!), means it will double in a year possibly. will lead to theoretically 32 + 12 units. if they unify, it will be more around 32 units at all.. if you SLI then, you get 64..

it's not far away (considering 3 cpu's for the xbox.. so you could have 2 gpu's as well.. or both plugged into one, a.k.a. 64x1..).

it is definitely doable. and it sounds great