Latest nV info

[CMKRNL]

NV31 A01 samples available now, A02 samples early February, volume production expected early March.

I noticed some question about NV33/NV34 configuration earlier: it is 4x1.

As expected, next high end part is NV35, next consumer part is NV36.

 

[DemoCoder]

What's the specs on NV31?

 

[CMKRNL]

NV31

0.13u
DX9
4x1 pipeline
128 bit memory interface, DDR-1
300Mhz core
Z culling, Z compression, Color compression
HW IDCT, iQuant, MC for DVD
Dual integrated 400Mhz Ramdacs
Integrated TV encoder, TMDS transmitters

 

[Mr.huang]

NV31

0.13u
DX9
4x1 pipeline
128 bit memory interface, DDR-1
300Mhz core
Z culling, Z compression, Color compression
HW IDCT, iQuant, MC for DVD
Dual integrated 400Mhz Ramdacs
Integrated TV encoder, TMDS transmitters

so the biggest difference between NV31 and NV34 is NV34 lack some bandwidth saving features , do i mean right ?

 

[Typedef Enum]

What have you heard on NV35? (as far as features and availability, of course)

 

[KimB]

so the biggest difference between NV31 and NV34 is NV34 lack some bandwidth saving features , do i mean right ?

That would be a strange difference. My first guess would be that the NV34 is to have a 64-bit memory interface.

 

[megadrive0088]

specs on NV35 WANTED!

 

[nggalai]

NV31

0.13u
DX9
4x1 pipeline
128 bit memory interface, DDR-1
300Mhz core
Z culling, Z compression, Color compression
HW IDCT, iQuant, MC for DVD
Dual integrated 400Mhz Ramdacs
Integrated TV encoder, TMDS transmitters

so the biggest difference between NV31 and NV34 is NV34 lack some bandwidth saving features , do i mean right ?

Or perhaps NV31 that failed some cache tests will be relabeled NV34--Celeron-style.

Other possibility: NV34 = NV31 minus TV and DVD stuff and at a lower clock.

ta,
-Sascha.rb

 

[KimB]

Minus TV and DVD stuff? Very unlikely. These have been nVidia's primary features in their lower-range parts since the GeForce2 MX.

 

[Dave Baumann]

I should imagine the video processing is done via the shaders anyway.

 

[nggalai]

I should imagine the video processing is done via the shaders anyway.

Good point, Wavey. You're right. Wouldn't make much sense cutting these features.

So what else could it be? 64bit DDR interface, anybody?

ta,
-Sascha.rb

 

[Randell]

NV31

0.13u
DX9
4x1 pipeline
128 bit memory interface, DDR-1
300Mhz core
Z culling, Z compression, Color compression
HW IDCT, iQuant, MC for DVD
Dual integrated 400Mhz Ramdacs
Integrated TV encoder, TMDS transmitters

wonder how it will stack up price, performance and IQ wise with either the 9500 (very similar specs) 9500Pro (more pipes) or 9700 non-pro (more pipes plus advantage of 256bit bus).

If it's priced at Gf4MX levels then it shouldl be a great card for the money.

TBH its got to be priced at 9500/Pro levels hasnt it?

 

[Dave Baumann]

I'm at a bit of a los to see exactly what market NV34 is aimed at, if NV31 is 4x1 - surely differences in clock speed would allow NV31 to scale to many different markets.

However, if NV34 is lower-end than NV31, as a few roadmaps we've seen suggest, then, yeah, a 64-bit interface may be one region where its cut down. VS may go as well as this is a good fit for the CPU.

 

[breez]

64-bit memory interface? Wouldn't that kill the performance totally?

 

[LeStoffer]

I'm at a bit of a los to see exactly what market NV34 is aimed at, if NV31 is 4x1 - surely differences in clock speed would allow NV31 to scale to many different markets.

However, if NV34 is lower-end than NV31, as a few roadmaps we've seen suggest, then, yeah, a 64-bit interface may be one region where its cut down. VS may go as well as this is a good fit for the CPU.

I wonder the same about the NV34. The NV3x imply a DX9 level part so at least we should be avoiding the GF4MX farce. I don't know whether it makes sense to scrap 2 VS when 4 DX9 pixel pipelines cost so much in terms of silicon already. Full 32 bit FPU precision should be gone with 16 bit FPU instead though.

The main point is that they move to the 0.13 process from a cost perspective. I don't know whether the move from 128 bit to 64 bit is a huge jump in order of cost anymore?

 

[Hyp-X]

Given the spec NV31 should be between 9500 and 9500Pro in performance (most likely closer to 9500).
In this case it's not much space to fit a new chip between NV31 and GF4MX.
(They will surely have NV31 clocked at different speeds.)

On the other hand a 4x1 chip @ 300 MHz and a 8x1 chip @ 500 MHz leaves a big gaping hole in the middle.
Kind of like at the end of last summer for ATI with nothing between the 9000Pro and the 9700Pro.

It would make much more sense if NV34 is a mainstream product, or maybe an integrated part.
But, not everything nVidia does makes sense...

 

[mr]

I was wondering about the following in Dave Baumanns GeforceFX launch Interview.

Q: I assume that anything available currently using the the 32-bit format will be run in FP16 mode?

A: Actually, no. We have native support for 32-bit integer, which is how we get the performance on the older apps. If we were to run them as FP16 then they wouldn't run as fast. So we have dedicated hardware with native support for 32-bit per pixel integer, 64-bit per pixel floating and 128-bit per pixel floating.

I assume that NV31 and/or NV34 will contain the FP path throughout the pipeline to get DX9 compliancy.
Could nvidia drop their native support for 32-bit INT and run them in FP for NV 31/34 ?
How many transistors could be saved by this method?
How much performance would be lost?

Any answer would be helpful.

 

[Bigus Dickus]

Given the spec NV31 should be between 9500 and 9500Pro in performance (most likely closer to 9500).
In this case it's not much space to fit a new chip between NV31 and GF4MX.
(They will surely have NV31 clocked at different speeds.)

On the other hand a 4x1 chip @ 300 MHz and a 8x1 chip @ 500 MHz leaves a big gaping hole in the middle.
Kind of like at the end of last summer for ATI with nothing between the 9000Pro and the 9700Pro.

It would make much more sense if NV34 is a mainstream product, or maybe an integrated part.
But, not everything nVidia does makes sense...

I was wondering the same thing Hyp-X. I would have thought there would be an 8 pipe DDR-I with lower clocks or 4 pipe with DDR-II and relatively high clocks somewhere in nVidia's mix to fill in that hole. This seems a bit odd.

ATi's lineup:

8 pipe 256 bit DX9
8 pipe 128 bit DX9
4 pipe 128 bit DX9
4 pipe 128 bit DX8

Seems like a good well rounded lineup. And nVidia's with this new information is looking like:

8 pipe 128 bit DX9 (500/500)
4 pipe 128 bit DX9 (300/400?)
4 pipe 64 bit DX8/9 (no VS?)

That is assuming the rumors about only 128/256 MB versions of the 500/500 NV30, with no "Ultra" version and regular versions based on clock speeds. If they did have a slower NV30 version (400/400) this would fill out their lineup a bit better IMO.

 

[Joe DeFuria]

8 pipe 256 bit DX9
8 pipe 128 bit DX9
4 pipe 128 bit DX9
4 pipe 128 bit DX8

Seems like a good well rounded lineup.

The thing that has really enabled ATI to do that, is by using a SINGLE CHIP (R-300), they were able to produce products with significant performance differentiation. This was achieved not really by clock speed differences as is the tradition, but by changing external bandwidth, and by the "disabling" pixel pipes. That was a very shrewd move by ATI.

The disadvantage of that method, is that on the lower end boards, you reduce your margins significantly because you are essentially "wasting silicon". I suspect ATI is getting very poor margins on the 9500 non-pro.

I expect ATI to rectify that with the RV-350. (Which I expect to be extremely similar to the NV31, specification wise.)

 

[RussSchultz]

Its a tough call as to whether NVIDIA's lineup is too sparse, or ATI's is too crowded.

The general concensus seems to be "why buy a 9500 (regular) with a 9500 pro just slightly more expensive and it performs very close to the 9700". This basically makes ATI have 2 valid products (9500 pro and 9700 pro).

I think if I were in control, when the RV350 shows up, I'd axe the 9500/9500 pro and replace it with a single product using the RV350. The segment currently served by the 9500 pro would have to choose between lower performing RV350 or the higher performing 9700/9700 pro.

I can't make any armchair assesments with the NVIDIA lineup since we have no idea how they'll stack up amongst themselves, much less with their competitors products. (But damn, did they screw that launch up. March? Feh!)