NV35 to have all that was wrong in NV30 fixed?

[megadrive0088]

skimming through the various forums i notice that a number of people including Hellbinder think that NV35 will have a true 8:1 configuration
in addition to the other widely expected improvements such as a 256-bit bus. that overall, NV35 will have everything fixed that was wrong with NV30.

how likely is it that NV35 will have a true 8:1 config, and wouldn't that alone dramatically improve the situation, provided it has the memory bandwidth to back it up?

 

[shaderman]

no way jose

it's a 256-bit bus and it taped out.

- shaderman

 

[megadrive0088]

alrady planning a R350 purchase (9800 Pro or the later 9900 Pro)

id just want to see both companies putting out kickass cards, with neither dying out. with both companies putting out their absolute best, which will benefit us the most

 

[KimB]

According to a statement I read recently, the NV30 is highly-configurable, and could be run in full 8x1 mode if the driver writers so chose. Supposedly it is not run in this mode due to performance reasons. While I will claim there won't be any performance difference in 99.9% of all single-textured scenarios (memory bandwidth), there will be mild performance penalties for some multitextured scenarios. Assuming that all pixel pipelines must operate on the same triangle, then it seems plausible that increasing the size of the current working area (by increasing the number of pixels per pipe) could actually decrease performance.

As for the "problems of the NV30," I think the biggest problem is its FSAA technique. The performance certainly is not bad, and don't forget that all performance figures released to date have been on pre-release drivers. Unfortunately, I have serious doubts that nVidia will fix the FSAA deficiencies of the NV3x core by the time the NV35 rolls around. It will be interesting to see how aggressive ATI is going to be throughout the rest of the year.

And shaderman, this just sounds silly:

NV is dead dead dead (at least for this year).

If NV is "dead" then the company is dead. You don't come back from the dead.

Regardless of what happens at the high end, nVidia really needs to put out some very good low-mid range products. If the NV31 and NV34 are good, then nVidia will be in very good shape as far as their video chip portfolio for the rest of the year. Regardless of the performance of these chips, however, it really is looking like nVidia's motherboard chipsets are really going to be a success this year, so I doubt that they'll have much problem coming back (assuming the corporate culture isn't all shot to hell...).

 

[kyleb]

You don't come back from the dead.

there are a lot of crhistans who would disagree with that.

 

[Joe DeFuria]

According to a statement I read recently, the NV30 is highly-configurable, and could be run in full 8x1 mode if the driver writers so chose.

Even if true...who cares. Even according to you, the NV30 doesn't have the bandwidth to support 8 "legit" pipes. So nVidia has a choice: "enable" these mystery pipes and legitimately claim 8 pipes, or stick to 4x2 and don't claim 8 pipes. Can't have it both ways.

As for the "problems of the NV30," I think the biggest problem is its FSAA technique.

I see3 major problems:

1) FSAA
2) potentially poor shader instruction throughput with floating point math...still yet to be borne out
3) Power Consumption. To get comparable performance to R300 architecture, NV30's power consumption is much too high.

and don't forget that all performance figures released to date have been on pre-release drivers.

How could we forget that? This thing's been on "pre-release" status for eons now. Having said that, nVidia CLAIMS to have shipped the first actual products....I presume they ship with a driver of some form or another?

Regardless of what happens at the high end, nVidia really needs to put out some very good low-mid range products. If the NV31 and NV34 are good, then nVidia will be in very good shape as far as their video chip portfolio for the rest of the year.

Not necesarily.

Don't undersestimate the mindshare impact of having the best video card on the planet. nVidia doesn't...otherwise we'd never see such a ridiculous product like the FX Ultra...which is obviously purely an attempt to capture the "performance crown", rather than actually be a viable product.

The NV31/34 could even be BETTER than the equivalnt offerings from ATI, but the "customers" coud still be asking for "Radeon / ATI" boards due to how the web sites "review" the high end products.

Regardless of the performance of these chips, however, it really is looking like nVidia's motherboard chipsets are really going to be a success this year....

It's debatable how successful they could be being limited to AMD systems only. That being said, ATI has a very tough market to try and crack into this summer with their own desktop chipsets (for btoh AMD and Intel platfors.) ATI certainly has lots to prove in that area.

 

[megadrive0088]

As for the "problems of the NV30," I think the biggest problem is its FSAA technique. The performance certainly is not bad, and don't forget that all performance figures released to date have been on pre-release drivers. Unfortunately, I have serious doubts that nVidia will fix the FSAA deficiencies of the NV3x core by the time the NV35 rolls around. It will be interesting to see how aggressive ATI is going to be throughout the rest of the year.

I agree and disagree with you here. I agree that one of the main problem with NV30-GFFX is its FSAA technique. from what i understand, its the same technique as the NV25-GF4Ti, only with more samples of it.

however I think there is a strong possibility that Nvidia will have a new AA unit in NV35, in a similar way that NVidia put a new AA unit in the
NV25-GF4Ti that the NV20-GF3 did not have.

 

[Althornin]

I agree and disagree with you here. I agree that one of the main problem with NV30-GFFX is its FSAA technique. from what i understand, its the same technique as the NV25-GF4Ti, only with more samples of it.

Until the AA gets better on the NV3x series, it cant even be a consideration for me.

Also, Chalnoth, if 8x1 would be memory limited with the GFFX's bandwidth, then why are 9500 pro/NP scores so different? I think its fairly clear that while they would be SOMEWHAT memory limited, there would be a noticeable gain in "enabling" 8x1 in the drivers, which i think is impossible, or they'd be doing it.

 

[Mulciber]

As for the "problems of the NV30," I think the biggest problem is its FSAA technique. The performance certainly is not bad, and don't forget that all performance figures released to date have been on pre-release drivers. Unfortunately, I have serious doubts that nVidia will fix the FSAA deficiencies of the NV3x core by the time the NV35 rolls around. It will be interesting to see how aggressive ATI is going to be throughout the rest of the year.

I agree and disagree with you here. I agree that one of the main problem with NV30-GFFX is its FSAA technique. from what i understand, its the same technique as the NV25-GF4Ti, only with more samples of it.

however I think there is a strong possibility that Nvidia will have a new AA unit in NV35, in a similar way that NVidia put a new AA unit in the
NV25-GF4Ti that the NV20-GF3 did not have.

right, but remember they had a year turn-around between the gf3 and gf4. if we're to believe the rumors the nv35 will be out less than 6 months after the nv30.

I think that the AA issue is a valid one, but I also think a much bigger problem is the disasterous FP and PS2 performance of the chip.

 

[Socos]

I think we are forgetting a couple of things here. If the NV35 still has that leaf blower attached its going to put a lot of people off. If they somehow improve the .13 process (which I believe is totally out of Nivida's control) and make it produce less heat then Nivida can say they have something.
To me the NV30 is just another re-fresh with changes from the NV25. I believe we can see the results of this with the light shinning on the 4x2 arcitecture. I don't understand why nivida would throw all this FUD out there if they didn't realise that the 4x2 is inferrior to a 8x1 setup.
This is the reason they are doing the PR two step.

My two cents.

 

[megadrive0088]

right, but remember they had a year turn-around between the gf3 and gf4. if we're to believe the rumors the nv35 will be out less than 6 months after the nv30.

true, but it actually might be similar this time also. the NV30 design was finished in early 2002, while the NV35 design has probably just been finished now in early 2003. thats a whole year.

I think that the AA issue is a valid one, but I also think a much bigger problem is the disasterous FP and PS2 performance of the chip.

I don't deny that at all, its just im not nearly as familar with its PS2 performance. also forgot to mention that its FP performance is also seriously lacking.

 

[Joe DeFuria]

Also, Chalnoth, if 8x1 would be memory limited with the GFFX's bandwidth, then why are 9500 pro/NP scores so different? I think its fairly clear that while they would be SOMEWHAT memory limited, there would be a noticeable gain in "enabling" 8x1 in the drivers, which i think is impossible, or they'd be doing it.

Well, 9500 Pro vs/ 9500 is 8x1 vs. 4x1....not 4x2.

That being said, I do agree that there would be some tangible gains for an 8x1 vs. 4x2....particularly with so-called "adaptive" anisotropic techniques enabled.

 

[NitroX]

Well I think the NV31 and NV34 could be very good.
The Ti4200 is hte best selling GF4 and the 9500 pro is the best selling of the R300 series

for obvious reasons that you get decent performance but do not empty your pockets so there is potential


Socos I think it was a 3 weeks ago they changed the fan on reference cards
too many people with the leaf blower/hair dryer comments
they found a solution that still cools well and does not slowly push your comptuer away from the wall that the back is facing

All in all I think IMO that Ati has the upperhand with the price performance ratio especially with the 9500 pro
and with the next release they will be doing even better for price/performance....... I hope

 

[Reverend]

Not having a 8x1 pipeline architecture is not "wrong".

Not having a 256-bit bus is not "wrong".

Unless what you mean by "wrong" is "not having the latest features/architectures that the competition has", i.e. "behind the curve".

 

[MfA]

They misjudged the competition, they were wrong. This would have been true even if the NV30 had been on time.

 

[KimB]

You don't come back from the dead.

there are a lot of crhistans who would disagree with that.

"come back" being the operative words there. I don't think there's a religion out there that believes that happens (except in very specific, rare cases...).

 

[KimB]

Also, Chalnoth, if 8x1 would be memory limited with the GFFX's bandwidth, then why are 9500 pro/NP scores so different?

Because the texel rate of the 9500 non-Pro is half that of the Pro.

 

[KimB]

Not necesarily.

Don't undersestimate the mindshare impact of having the best video card on the planet. nVidia doesn't...otherwise we'd never see such a ridiculous product like the FX Ultra...which is obviously purely an attempt to capture the "performance crown", rather than actually be a viable product.

The NV31/34 could even be BETTER than the equivalnt offerings from ATI, but the "customers" coud still be asking for "Radeon / ATI" boards due to how the web sites "review" the high end products.

Remember that nVidia took the performance crown from 3dfx in an undisputed fashion with the release of the GeForce DDR. Many people purchased the Voodoo5 cards just because they were made by 3dfx. Most people are slower to change their perceptions than you seem to think.

It's debatable how successful they could be being limited to AMD systems only. That being said, ATI has a very tough market to try and crack into this summer with their own desktop chipsets (for btoh AMD and Intel platfors.) ATI certainly has lots to prove in that area.

The only thing is, Intel really is a much harder competitor than SiS or VIA in the chipset market. I don't know how much of a chance ATI has yet.

 

[Reverend]

They misjudged the competition, they were wrong. This would have been true even if the NV30 had been on time.

Misjudgements would have been an excuse and (possibly) a reason... unless they didn't know what the competition was coming out with. Even in such a case as yours, misjudging or (for, perhaps, a better word) overlooking the competition doesn't mean they're wrong in the context of the original poster, IMO - IMO, it means, like I proposed as as a possibility above, a simple case of being behing the curve.

In any case, "misjudgement" of the competition does not invalidate what I said... it is an entirely separate issue wrt to pipeline architecture and adoption of a bus tech everybody knows existed for ages.

 

[kyleb]

but Chalnoth, you should not overlook the fact that many other people purchsed the voodoo5 because it had superior image quality and acceptable performace, myself included. also, just to dispell any notion that i was simply loyal to 3dfx i should point out that i uprgaded from an orignal gefroce 256.