Fuad of The Inq, trying to explain NV40 ... AGAIN

[PaulS]

For what it's worth (no more than any other news source, i guess), here's a translated news piece from Within Games which continues to talk about a 16x1/32x0 architecture being confirmed.

 

[akira888]

16 textures per pass is congruent with an 8x2 pixel pipeline organization--that's 8 pixel pipelines which each have 2 attached TMUs, so although you could get 16 textures per clock, you are limited to a maximum of 8 pixels per clock.

The number of texture samples the chip can take in a single pass has really nothing to do with the number of TMUs. All the "textures per pass" number refers is how many samples the chip can take using loopback (ie multiple iterative cycles) before writing out the fragment (or rejecting it) to the FB. Textures "per cycle" on the other hand is what you're talking about.

R3XX and NV3X both had 16 textures per pass available, while having only 8 physical TMUs.[/quote]

 

[WaltC]

Uhh, WaltC, _you_ dont even seem to understand what you're talking about.

Why is that, exactly?...

 

[digitalwanderer]

For what it's worth (no more than any other news source, i guess), here's a translated news piece from Within Games which continues to talk about a 16x1/32x0 architecture being confirmed.

Some interesting R420 info there too!

- it will give two models of the R420: A pro map with 500Mhz chip and 475Mhz storing act and one?Nitro? map mentioned with 600Mhz chip and 550Mhz storing act.
- it is very probable that the R423 will have actually 16 pipelines, while the R420 with their 12 comes along. Possibly it concerns also the same chips with those dependent on the quality 4 pipelines (a Quad block) to deactivate itself or not.
- the R420 is as further R300-Refresh as well known a temporary solution. Accordingly ATi wants to rely all too no longer for a long time on architecture and bring soon the R500, whose development status progressed allegedly already very far. With the development of the R500 one can besides strongly on the support of Microsoft relies, since the technology is to be inserted use any form in the Xbox2.

 

[WaltC]

The number of texture samples the chip can take in a single pass has really nothing to do with the number of TMUs. All the "textures per pass" number refers is how many samples the chip can take using loopback (ie multiple iterative cycles) before writing out the fragment (or rejecting it) to the FB. Textures "per cycle" on the other hand is what you're talking about.

R3XX and NV3X both had 16 textures per pass available, while having only 8 physical TMUs.

Yes, I understand the difference, but my point was the INQ keeps talking about "16x1" while mentioning 16-textures per pass, so I was assuming they meant "per clock" instead of "pass". I really don't know what they mean to say...

 

[madshi]

For what it's worth (no more than any other news source, i guess), here's a translated news piece from Within Games which continues to talk about a 16x1/32x0 architecture being confirmed.

Here's a better translation for ya non-german speakers:

Finishing this year's CeBIT report we will tell you about some interesting rumors, which we were hearing during those 4 days:

- The NV40 architecture (16x1 for textures / color values, 32x0 for Z values) was confirmed by the x-th source. That cip is going to be gigantic.

- The NV40 will be faster than the R420. It can be expected that the R420 is between 15% and 5% slower than its NVidia competitor.

- There will be two models of the R420: A "Pro" card with 500Mhz chip and 475Mhz memory clock and a card called „Nitro“ with 600Mhz chip and 550Mhz memory clock.

- It's very probable, that the R423 will really have 16 pipelines, while the R420 comes with 12. Possibly those 2 chips are identical, with 4 pipelines (one quad block) getting disabled depending on the quality.

- The R420 (as another R300 refresh) is known to be a temporary solution. So ATi will not rely on the architecture too long anymore and will soon bring the R500, whose development state is reportedly quite far. Additionally, when developing the R500 ATI can rely on Microsoft support, cause the technology is planned to be used for the Xbox2 in one form or another.

 

[Sabastian]

heh, considering that the Inq have suggested just about every sort of configuration that one could possibly fathom after all is revealed with regards to the NV40 or R420. Someone will point out in the future that the Inq was "spot on" about things in the past so that they can rationalize speculation about future generation hardware that the Inq has posted. It's a never ending cycle for them guys.

 

[PaulS]

- The R420 (as another R300 refresh) is known to be a temporary solution. So ATi will not rely on the architecture too long anymore and will soon bring the R500, whose development state is reportedly quite far. Additionally, when developing the R500 ATI can rely on Microsoft support, cause the technology is planned to be used for the Xbox2 in one form or another.

I've seen this elsewhere and I still don't buy it. There's no way MS are going to want comparable graphics hardware to be on the PC months before XB2 debuts, so either XB2 is a lot closer than we think or the tech is beyond R500 (similar to XB1 being inbetween nVidia PC parts tech-wise).

Or it's just false altogether .

EDIT: Fixed typo

 

[VtC]

I've seen this elsewhere and I still don't buy it. There's no way MS are going to want comparable graphics hardware to be on the PC months before XB2 debuts, so either XB2 is a lot closer than we think or the tech is beyond R500 (similar to XB1 not being inbetween nVidia PC parts tech-wise).

At the risk of sounding like digitalwanderer here, could someone explain to me what MS could offer ATI in the area of GPU design?

Or, would it be more of a "we'll incorporate something special for you into DX-next?" type of thing they're talking about?

 

[PaulS]

At the risk of sounding like digitalwanderer here, could someone explain to me what MS could offer ATI in the area of GPU design?

Or, would it be more of a "we'll incorporate something special for you into DX-next?" type of thing they're talking about?

Well, look at it this way - let's assume XB2 will be based around an R500 DX10 part. Let's also assume that ATi are going to want to transfer a lot of the XB2 tech to the highend PC space.

Because the ATi GPU will need to be fully DX10 compliant, MS will obviously work closely with them to ensue that's the case. It's not that they're going out of their way to make it difficult for nVidia - indeed, the process is as open as any other DX version. It's just the nature of console development that the firm providing the GPU will work closely with whoever is designing the API. Thus, the chances of nVidia coming out with as good a DX10 part as ATi are lessened.

By no means impossible, though.

 

[Joe DeFuria]

- The R420 (as another R300 refresh) is known to be a temporary solution. So ATi will not rely on the architecture too long anymore and will soon bring the R500, whose development state is reportedly quite far. Additionally, when developing the R500 ATI can rely on Microsoft support, cause the technology is planned to be used for the Xbox2 in one form or another.

I've seen this elsewhere and I still don't buy it. There's no way MS are going to want comparable graphics hardware to be on the PC months before XB2 debuts, so either XB2 is a lot closer than we think or the tech is beyond R500 (similar to XB1 being inbetween nVidia PC parts tech-wise).

Or it's just false altogether .

EDIT: Fixed typo

Well, I personally sit somewhere between these two views.

I certainly don't think there will be any new architecture from ATI for the "fall" cycle. However, it's not out of the question to me for ATI to have a R500 based PC part out for spring 2005. (To be clear, I lean toward fall 2005, but I wouldn't say Sring 2005 is out of the picture.)

The way I see it, R500 has been in development for a long time, given that it was scrapped as the original R400. Now, obviously,what we see as the actual R500 is going to be significantly different from the original R400, but I'm also guessing that there will be some re-use there. So, a one year lag from R420 doesn't seem unreasonable.

I don't think MS has any real concnern about what PC parts exist or not. While X-Box is presumably based on R-500 architecture, I'm betting it will likely be a bit more customized for x-box / console use.

 

[Luminescent]

If the Cebit rumors are correct and R420 is indeed 5-15% slower than NV40, assuming it has 12 pixel pipelines (in comparison to NV40's hypothetical 16), it'd be interesting to imagine how R423 would perform in comparison to NV40. R423 would theoretically gain 33% more pixel rendering performance than R420.

Edit: Sorry for the miscalculation, relative to R420, R423 would gain 33%, and not 25%, more pixel rendering performance. Thanks Evildus

 

[Evildeus]

R423 would theoretically gain 25% more pixel rendering performance than R420.

I think it's more 33%.

Well, 5-15% more FPS is great, but if the quality difference is as huge as between NV35/R350, then i think the R420 is still the winner.

 

[Joe DeFuria]

If the Cebit rumors are correct and R420 is indeed 5-15% slower than NV40, assuming it has 12 pixel pipelines (in comparison to NV40's hypothetical 16), it'd be interesting to imagine how R423 would perform in comparison to NV40. R423 would theoretically gain 25% more pixel rendering performance than R420.

I'll say it again, but I think it bears repeating. If NV40 turns out to be 5-10% faster on average (in legacy, non pixel shading apps), it'll likely be because it has more raw bandwidth, IMO. I tihnk either 12 or 16 pipes are going to be saturating the bandwidth at high res / AA.

I find it interesting that some of the "leaked" benchmarks have been found to be running in 16 bit mode. That is an indication to me that bandwidth limits are still going to be very much in play. that is, if NV40 is 16 pipes, and R420 is 12...we won't see all that much difference in high quality modes.

That being said, 32 "z only" operations per clock (assuming that's true), could be a significant advantage with Doom3. As the z-only pass should not be bandwidth limited.

Also, pixel shading performance in general is just a complete unknown, and shouldn't be tied to pixel pipes until we learn more about the respective architectures. (How many and what type of pixel shading instructions can be carried out "per clock" will likely vary between the architectures.)

 

[vb]

If the Cebit rumors are correct and R420 is indeed 5-15% slower than NV40, assuming it has 12 pixel pipelines (in comparison to NV40's hypothetical 16), it'd be interesting to imagine how R423 would perform in comparison to NV40. R423 would theoretically gain 25% more pixel rendering performance than R420.

I don't think we'll see a R42x with 16 pipes this early. If it was designed for 12 it should have logic only for 3 working quads. Plus if they considered 4 out of 4 quad units would be an awful yield and a nightmare to pick them up, i dont think this will change until a major respin. If NV40 is faster it will depend a lot on the price Nv will have to ask for it to be able to make a profit. They can give away the chips and i don't see AIB making much of a profit at 500$.

Plus who's buying Hardware in the summer? Real marketshare battle will be fought in the autumn when value and mid solutions will also show up.

 

[Evildeus]

for what it worth....

nv40

- the chip will be feature 32 bit throughout the pipe without loss of speed
- first test with half-life 2 showed that the chip is going to be 2 - 7 times faster than the geforce fx 5950
- the chip will feature RGAA
- the chip will need more power, but will use a single slot solution, as we were told

http://www.nvnews.net/vbulletin/showthread.php?p=288551#post288551

 

[T2k]

If the Cebit rumors are correct and R420 is indeed 5-15% slower than NV40,

Sorry, maybe I'm lost but where did you get this rumor? Link, pls...

edit: typo

 

[volt]

Yet we do not know whether or not the benchmarks in question were forced (via forceware hah) to run in brilinear -- but would NV really do that *if* NV40 has *that* much RAW power?

 

[Evildeus]

If the Cebit rumors are correct and R420 is indeed 5-15% slower than NV40,

Sorry, maybe I'm lost but where did you get this rumor? Link, pls...

http://www.withingames.net/?show=articles&type=showarticle&id=83&site=7 (translation just above by Madashi)

 

[Johnny Rotten]

Lol, I think I'm just going to avoid the various 3d card websites for the next 3 weeks completely until we have concrete inforamtion. This is getting silly.