Xbox GPU and PC-side

[cthellis42]

I was recently reading a GameSpy article which I'd missed before regarding the upcoming PC version of Halo, and something on the top of page 2 just didn't quite sit right.

The interviewed is Randy Pitchford, president of Gearbox Software, and when he starts talking about the graphical side of the PC port, he begins with:

"Halo arguably used the graphics processor of the Xbox better than any other Xbox launch title. And that graphics processor is very custom and very much ahead of the PC -- relative to when it debuted. In fact, we're just now seeing video cards released by the major manufacturers that are capable of doing what the Xbox's GPU can do. This was both a blessing and problem for us."

Now I know that the Xbox GPU isn't a total clone of the PC-side GF3, but just what might he be talking about in here that doesn't qualify is shifty propagandizing to make things sound more exciting than they are?

 

[Paul]

I declare bullshit.

XGPU is a mix between a geforce 3 TI 500 and a Geforce 4 ti 4200. It's nice stuff, however I just don't think this guy knows what he's talking about.

 

[Fox5]

Well, I think the fastest card out at the time of the xbox launch was geforce 3 ti 500 or radeon 8000. Anyhow, just change what he said to "we're just now seeing affordable video cards released" and it becomes correct, since the average person isn't going to plop down $400 for the latest and greatest. You could also change it to just seeing pcs that come with integrated video good enough.

 

[DeathKnight]

It's not quite bullshit.

The XGPU is really not a GF3. It's actually closer to a GF4 (XGPU has relatively the same functions, same expanded shader capability, and the speed). At the time of the Xbox release the GF4 was not yet released. The XGPU is also said to include a couple functionalities that were found in the much later NV30 (supposedly some special geometry features).

 

[Steve Dave Part Deux]

DN- I believe you're thinking of the NV2A's ability to write 8 stencil values per clock(like the NV30). Though, I'm not entirely sure if that is true.

 

[Fox5]

Xgpu is speed of a geforce4?! I thought it was more like a mid or low end geforce 3, with geforce 4 level shader capability.(perhaps a bit less)

 

[Paul]

NV2A is between the NV20 and NV25.

NV2A has 4 pixel pipes and two vertex shaders, the same as the NV25(geforce 4).

In fact, we're just now seeing video cards released by the major manufacturers that are capable of doing what the Xbox's GPU can do.

Let's see, this quote was made in May. And a 9700 PRO which was released how long ago? Destroys the XGPU.

Like I said, bullshit.[/quote]

 

[Fox5]

Radeon 9700 wasn't cheap though, they could care less if 10,000 people have something that blows away the xgpu, they need a market big enough that can play their game. Maybe it's just now that computer manufactuors are shipping computers with video cards powerful enough to beat out the xgpu. Certainly the average computer wasn't shipping with a radeon 9700 when it was new. Of course, it's also bs.

 

[Paul]

Doesn't matter if it wasn't in many PC's, look at the quote.

we're just now seeing video cards released by the major manufacturers that are capable of doing what the Xbox's GPU can do.

He is obviously wrong.

 

[DeathKnight]

Xgpu is speed of a geforce4?! I thought it was more like a mid or low end geforce 3, with geforce 4 level shader capability.(perhaps a bit less)

Yes. The actual chip itself is the speed of a 233MHz GF4. The memory bandwidth available to it is more like a mid or low-end GF3.

 

[Fox5]

Well, there is one last possiblity for him not to be wrong....maybe when he said major manufacturers he was referring to dell and gateway(or even the board makers like hercules and creative)....I wouldn't be surprised if dell dwarfs ati and nvidia combined.

What's the point of the xgpu being clocked higher, when the memory isn't fast enough?

 

[Sonic]

Not to mention that the memory controller for the GF4 is different than the NV2A. It's actually improved somewhat according to Nvidia.

I think the key differences between a GF3 and NV2A are the extra vertex shader, better memory utilization, and maybe a different cache setup. And I'm not sure on this but I thought that there was support, albeit very limited, for 3d textures as well.

 

[Fox5]

Geforce 3 can do 3d textures, or at least that's what my geforce 3 box said....

 

[DeathKnight]

Not to mention that the memory controller for the GF4 is different than the NV2A. It's actually improved somewhat according to Nvidia.

I think the key differences between a GF3 and NV2A are the extra vertex shader, better memory utilization, and maybe a different cache setup. And I'm not sure on this but I thought that there was support, albeit very limited, for 3d textures as well.

The NV2A lacks the GF4's more advanced memory controller. The NV2A includes the quad-cache architecture of the GF4. I'm wondering if the NV2A also includes support for z-corrected bump-mapping (haven't seen a case in any game where this would be put to good use though).

 

[zurich]

I'm halfway through writing a vertex shader system on Xbox, one of the things thats often lost about Xbox is how unlike a PC it can be. Using the Xbox at a lower level allows multiple vertex shader to be held resident and also gives you exact control over the vertex stream, makes the cost of changing vertex shader much cheaper.

Sure you can use an almost identical API to DirectX 8 but if you want to get good performance out of it, you have to go to a lower level. I'm using the so called 'Direct' functions (there is actually an even lower level), its really quite nice to actually know that you not going through some complex driver mapping but are driving the hardware relatively directly.

With Xbox its largely a question of approach, do you treat it just as another PC video card or do you treat it in a similar fashion to PS2. While the first is cheaper and makes mult-platform titles easier, the later allows so much more power. Things that are impossible or at least quite slow on PC becomes almost trivial on Xbox, due to you knowing EXACTLY what the hardware is doing. E.g. You can fiddle with the stencil buffer via standard rendering commands on Xbox (you simple pretend the depth stencil buffer is an ARGB render target, do the renderng and then put it back, this makes things like blurring the stencil buffer trivial, yet try and do a similar thing on a PC) or change a vertex buffer almost as soon as the GPU has finished with that part.

Perhaps what the Gearbox guy meant was that PC hardware today allows things previously only possible on the NV2A via low level hardware coding.

 

[Fox5]

Well, I don't think gearbox is known for being technical wizzes, maybe he just didn't know what was going on in the market, or assumed the average person reading the interview wouldn't. Or he was just bsing to make himself look better. The only previous things I can remember from them are opposing force(which was pretty good, certainly better than blue shift and a few free halflife mods) and the tony hawk's pro skater 3 port, which I remember them talking about how advanced it was with internet and keyboard support(but not full keyboard support, it only supported the letters and numbers the ps2 version had), which while good compared to the other tony hawk ports, certainly wasn't on the level of a game made for the pc.(I believe it had a few glitches and no duh things missing...not sure, I didn't play it much, but could you change the res?)

 

[cthellis42]

Or perhaps he was just saying whatever he felt like, assuming most people wouldn't know or wouldn't care, and otherwise knew it would make their efforts (and the Xbox in general) sound all the better.

 

[megadrive0088]

NV2A, I believe, also lacks the new (then new) FSAA unit that the NV25 got.

Isn't the NV2A ment to have slightly difference/longer Vertex Shader instructions than either the NV20 (GeForce3) or NV25 (GeForce4) ?

Overall, the NV25 has higher performance and more features than NV2A.


When MS annouced XBox, it was going to have a 300 Mhz NV25, or even something a bit beefier than NV25. something inbetween NV25 and NV30, but obviously that turned out not to be the case. XGPU was scaled back conciderably. or MS was lying all along....

I wonder if NV30 or NV35 could do a near-exact reproduction of the CGI Raven-Robot render that MS (falsely) used to show what XBox would be able to do.

now don't get me wrong, I think XBox is a decent console, I own one and plan to get another for modding (emulators and stuff) but XBox did not turn out to be as powerful as it could have been. though XBox is more powerful than where the mid-1999 reports placed it. that is, having a 500 Mhz Athlon with NV10 (GF1)

 

[Proteus]

What about the framebuffer access etc for neato effects like edge tracing emboss etc, like the filters in wreckless. Depth of Field etc (Tho you can do that lots of different ways, like most effects.. having framebuffer access just makes some effects simpler).

That's my understanding anyway, but I'm not the most technically minded person..

 

[a4164]

I have a question about the NV2A. Does if feature 2TMUs per pipe (4x2) or is it a 4x1 part?