Nvidia G71 - rumours, questions and whatnot

Lol, it seems we're veering way off topic here, but I'll look into it tonight and make some screenshots on my card.. i DO have some major moire, but that's mostly on distant objects etc.

 

Well the issues I've noticed are limited to racing games where inferior filtering is easily noticed.

 

It's been some time since I touched Live For Speed, but I recall it having pretty crappy MIPmapping.

Bilinear + MIPmapping are basic texture antialiasing; if you need in a game idiotically high SS sample densities to cure some of the texture aliasing, take a long guess what could be wrong with it.

 

Your other option would be to adjust the the global LoD back a few points on a driver level with rivatuner. I havent played the game but it might be worth trying as a temporary patch job fix to see if it helps. It will reduce the sharpness of your textures slightly though.

 

Agreed. I'd actually say that running without aniso is more of problem, ie much more apparent, at high res than low.

 

It's getting closer to the rumored launch date (cebit/early march) and the rumors have died down????

 

Well, considering CeBIT isn't until the second-third week of March, I wouldn't be worried just yet. If we don't see a flurry of rumors two weeks from now, then it does seem likely we won't have a launch, or at least not a hard launch, at CeBIT.

 

That is interesting, isn't it? We've at least seen pics of G7600. Assuming they are "hard launching" G71, then it should be showing up in volume at AIB's pretty quick, I'd think. . .

 

Yeah, I'd imagine so. It is obvious, of course, that as we get closer to launch, it becomes harder and harder to keep things under wraps, as more and more people are involved.

 

Good point there. Either NV is doing an extremely well in keeping the G71 under wraps or this is an indication of delay....

 

Highest resolution possible with the highest hybrid sampling mode possible. You shouldn't really need to result to any form of SSAA to cover for crappy MIPmapping, but it's the most effective sollution IMO.

 

Due to the ever shrinking product life cycles, the critical nature of competitive feature support and performance leap frogging... revealing final product specification is held CLOSE to the vest for as long as possible

By itself, I don't take this as any significant delay.

However, if their Fab House leaked to the industry that they we having respins with the G71 die, that would indicate a delay.

 

Also, now that nvidia uses a third party maker for their top cards ( Flextronics still ? ) it does not take too long to slap a sticker on it and burn a driver CD ( half joking ) rather than the old times when the board makers had to do a lot more.

There was not too much prior notice of the 512GTX as far as I recall.

 

I'd rather say that the final set of specs is in constant flow until the product is delivered. Some features get stripped or (less so) added in the process for lack of time/resources etc.

 

Get out your micrometers and fire up your Photoshops/Paint Shop Pros. . .

7600 die shot: http://www.vr-zone.com/index.php?i=3219

Edit: I make it 7% bigger than NV43. NV43 was 150mm2, 143M transistors. So roughly 160mm2. What's the rule of thumb for increase in transistor density from 110nm to 90nm? If we assume 50% (thanks, Rys), then we get 229M transistors. Whee. . . But I think that die size rules out 256-bit pretty conclusively (tho we've been thinking that anyway).

 

sounds about right for a 12 pipe part

 

G71 or whatever it is called internally:
32 pipelines at 700MHz to 750MHz and 1.1ns memory.
Maybe...sounds a bit unbelievable and unbalanced to me.

 

Well, with those numbers, 12-pipe 128-bit interface chip, maybe?

Edit: Gah, too late

 

RV530 is 157M transistors on a 156mm2, so they're basically the same size.

Which is a contrast in comparison of RV515 and G72, where G72 is notably smaller (77mm2 versus 100mm2).

It's sorta sobering to realise that 7600 is roughly the same number of transistors as NV40, but with prolly half the ROPs (8) and 75% of the fragment pipelines (12).

Jawed

 

How exactly is it that we know the package size is the same? I've seen people do this while comparing between ATI and NVidia products, which makes no sense to me.

229M transistors seems waaay to much. RV530 is about the same size, and it's 157M (granted the IHV's count differently, but not by that much). Comparing transistor densities is pretty tough. From the 130nm R420 to the 110nm R430, there was only a 20% increase in density.

Either way, I think ATI is boned. I've kept saying that ATI spent way too many transistors for it's small batch size in R5xx, and now they'll suffer. If the 7600 is indeed 12 shader pipes, and I have no reason to doubt it, then it will have three times the texturing rate of RV530. It's likely that there will be 4 ROPs, but there could be even 8 with double Z.

This is gonna be ugly.