Will we get a .plan update from Carmack on Radeon 9x00?

[Joe DeFuria]

I'm curious to know what Carmack's thoughts are, now that the 9700 has matured to it's shipping form, in terms of drivers and relative performance to the GeForce4.

And actually, I'm even more interested to know how the 9000 handles Doom. We know that Carmack wasn't particularly impressed with the GeForce4 MX, what about the newest "budget" chip?

We know that the Radeon 8500 is a "fine card" for Doom3. The 9000 doesn't seem to handle "DX7" type games quite as well as the 8500, but in synthetic DX8 tests (and in a flexible texture engine like Serious Sam), the 9000 tends to exceed the 8500.

I would think lots of people would like to know how the 9000 stacks up against the 8500, and the GeForce4 MX in Doom3...

I know B3D has gotten some answers from Carmack before...try again?

 

[KimB]

Well, obviously, the Radeon 9700 trumps the GeForce4 in DOOM3, and by a good margin. Personally, I'd be more interested in JC's thoughts on driver quality...that's a bit more interesting at the moment. What will be even more interesting is the comparison between the Radeon 9700 and the NV30 when it arrives.

As for the Radeon 9000 vs. GeForce4 MX, again, the MX will obviously lose, and again by a good margin.

What will be more interesting are the upcoming NV18 and NV28 cards...hopefully they're not just the GF4 MX and GF4 Ti with AGP8x (Although, unfortunately, that's still a very good possibility).[/url]

 

[Joe DeFuria]

Personally, I'd be more interested in JC's thoughts on driver quality...

Well, along those lines, I'd like to know if JC has plans for a specific R300 path (or if one is needed or could be beneficial?) To date, I believe Doom3 has been running on the R300 using the R200 path.

As for the Radeon 9000 vs. GeForce4 MX, again, the MX will obviously lose, and again by a good margin.

Yes, but I'm more interested in the 9000 vs. the 8500. Somewhere between the GeForce4 MX (not good for Doom) and the 8500 (good for Doom), lies the "this is the minimal card you'll want to run Doom." Where does the 9000 fit? Is it actually better than the 8500 for Doom?

What will be more interesting are the upcoming NV18 and NV28 cards...

Actually, I think they will be the most uninteresting.

 

[KimB]

The NV18 and NV28 will only be uninteresting if those extra transistors were just a typo.

 

[Reverend]

JC has already stated his opinions on the R300 (as well as vis-a-vis the best NVIDIA offering right now, the GF4 Ti).

He may comment on any R300 bugs that are solved via drivers or he may not care.

 

[Joe DeFuria]

JC has only really said "it's faster".

That we could have all guessed. How much faster is what we'd like to know.

The Doom3 engine is a much different beast than everything before it, so relative scores on DX7 titles that proliferate today aren't necessarily reflective of Doom3's relative performance.

 

[Sage]

JC likes the R300, its very simple. but just wait til he gets actual NV30 hardware in his hands right now everyone is stuck useing NV30 emulators, which accounts for the R300 being faster.

 

[Reverend]

JC has only really said "it's faster".

That we could have all guessed. How much faster is what we'd like to know.

The Doom3 engine is a much different beast than everything before it, so relative scores on DX7 titles that proliferate today aren't necessarily reflective of Doom3's relative performance.

Well, the way I read his words is that he seems to be relatively quite pleased by the R300's faster performance. That usually should be quite enough if it comes from JC. Remember, he said that D3 is GPU-bound (I assume at least up to 1024x768) and we all know how much faster the R300 is compared to the Ti4600 in all the games tested by various reviewers (even without AA, aniso).

 

[pascal]

I would like to see more generic comments like:
- The improved image quality, how good is it?
- How much faster with uncompressed textures?
- How good is the new DX9 level cards compared to previous cards?
- How he could improve the current engine and something about the future engines?

Something more technicall than just "faster".
Also I would like to know more about the Doom3 CPU scalling.

The Radeon 9000 should be at least twice as fast as the GF4MX because it will need a single pass and the GF4MX is like the GF1/GF2 requiring at least 4~5 passes.

The best cheapest nvidia 3Dcard still available (not for OEM) on the market is the GF3Ti200.

Xmas is coming and people need some help understand the scenario of future games.

 

[BoddoZerg]

I think it's awfully sad that Nvidia is releasing a "nv18", whose numerical designation implies that it's a DX7 part based on nv15. PC game developers can only implement DX8 features once the cheapest new computers have DX8 videocards, but if everyone slaps nv18s into their budget boxes just like nv17 and geforce2mx(nv11?), it will limit popular games to DX7 for a long time.

 

[RussSchultz]

NV18 (Crush) is supposed to be the clawhammer motherboard, isn't it?

 

[Joe DeFuria]

I don't know, at nVidia's last conference call when asked about the NV18/28 all they basically said they don't talk about unannounced products.

But, uh, NV30 isn't announced yet either and that didn't seem to stop them from talking about it in the same call.

 

[Sage]

NV18 = Clawhammer

 

[kid_crisis]

NV18 (crush) has 81 million transistors and is the Northbridge of the Nforce-2 chipset. There may be a variant of it intended for the Clawhammer, but it will be drastically different, because Clawhammer has a big chunk of the Northbridge built into the chip already (i.e. the memory controller).

 

[Sage]

right right, NV18 = NForce2. what was i thinking!?

 

[Geo]

There's a para from JC about R300 in the new Maximum PC (got it yesterday).

"Here's what John Carmack. . .told Maximum PC about the ATI R300:"


"The R200 already had very good feature specs, providing all the the texture units and frament operations that Doom III needs for optimal implementation; it just didn't perform as well as the competitive Nvidia cards, due to low-level hardware details. The only new R300 feature that is explicitly coded for is two-sided stencil, which cuts in half the number of stencil shadow triangles that need to be processed, for a worthwhile speedup. The big win is just the raw bandwidth and the eight pixel pipelines, which make the R300 by far the fastest card I have tested for our applications."

"There are a couple minor improvements that I would like to try out on the upcoming generation of cards that have more fragment processing capabilities. But that will probably be done through the OpenGL 2.0 rendering path, which won't be available for a while yet."

Pg 13, lower left, Max PC Sep. 2002.

That sounds to me like he's completed his code path for R300, otherwise he couldn't identify "only new is explicitly coded for".

 

[Geo]

Oh yeah, "by far the fastest", while lacking the quantitative exactitude that we'd all like to see, is still, I think, qualitatively more definitive than "it's faster", which is why I bothered to post this in the hopes y'all hadn't already seen it.

 

[Joe DeFuria]

Thanks, geo!

I doubt I'll get a response, but I did e-mail Carmack today asking about Radeon 9000 performance. Hopefully, he'll make a .plan update with another round of "state of the video cards" type thing in the near future.

 

[Bjorn]

Thanks, geo!

I doubt I'll get a response, but I did e-mail Carmack today asking about Radeon 9000 performance. Hopefully, he'll make a .plan update with another round of "state of the video cards" type thing in the near future.

You're not the only one interested in that.
But somehow i think that Nvidia might not want something like that at this moment

 

[no_way]

has JC

commented on why D3 in-game models are still _so_ low-poly ? Don't current cards have enough vertex-processing power or, more likely, would shadow volume calculations would become prohibitively expensive ( why couldnt he do shadow calcs on lower-lod model ) ?
I know they are making it up with Polybump-like bumpmaps derived from high-res models, but still .. characters do look blocky. recent Quakecon screenshots show it quite clearly.
the quoted 1500 polys per character isn't much.[/img]