Nvidia demonstrating the next gen

[Sxotty]

http://www.tomshardware.com/business/20030307/gdc_2003-04.html

In any case, if it follows the same timeline as the FX it could easily be out in July awesome, right November FX was demoed came out in March, March to july or august seems reasonable, I think it will be pretty cool this summer anyway.

 

[nutball]

Well done NVIDIA Public Relations!

"I know, lets put up four fairies, then everybody will think it's four times faster!".

 

[RussSchultz]

Now I know their editors are idiots.

"Not an empty seat in the house"

And their picture shows about a third the seats not occupied.

 

[MuFu]

Now I know their editors are idiots.

"Not an empty seat in the house"

And their picture shows about a third the seats not occupied.

Maybe that was after they started the presentation.

MuFu.

 

[BoardBonobo]

I have to wonder how well ATi could compete with nVidia if they, nv, chose to sustain an aggressive price war in the mainstream market. Especially if nvidia are using a cheaper, higher yield process for their chips.

 

[Dave Baumann]

Especially if nvidia are using a cheaper, higher yield process for their chips.

They are not though.

And, as far as die sizes go, ATI and NVIDIA are like this:

RV280 (.15u) is smaller than NV34 (.15u)
RV350 (.13u) is smaller than NV31 (.13u)
R350 (.15u) Is about the same as NV30 (.13u)

Performances I've heard (I don't have these inhouse yet) seems to indicate that NV34 is less than Ti4200 and NV31 (ultra) ~= Ti4200 without AA (and but much faster than with). RV350 on the otherhand has a target of Radeon9500 PRO to beat, which puts it well ahead than a Ti4200 in the first place. I'm wondering what NV are doing here - As MuFu states, they are using complex PCB's in the cheap markets, have larger cores and/or using more expensive processes for each of the segments and, apparantly, not stellar performance so far.

These performances I'm hearing though are in current titles, its going to be interesting to see how the two lines stack up with predomanently shader dominated benchmarks. For instance, where NV30 has '8' shader pipes, I heard NV31 has 6 which would make it a reasonable shadr performance chip - assuming they can sort out some the the shader compiler 'twitchiness' we hear about.

Still, the proof will be in the testing - I'm pretty sure we'll get the opportunity to see where the real world performance lies soon.

 

[chavvdarrr]

re

ATi's performance is OK, whats unknown is are they able to produce&sell enough units.
Maybe in USA is different, but in Europe R300s are not so easy to find, even 9500 is 180-200$.
Last I heard that Saphire and Gigabyte receive not more than 60-70% of requested chips from ATi.
And it seems that ATi is not increasing its market share....

 

[V3]

RV280 (.15u) is smaller than NV34 (.15u)
RV350 (.13u) is smaller than NV31 (.13u)
R350 (.15u) Is about the same as NV30 (.13u)

Do you have the measurements for individual chip ?

 

[KimB]

I somewhat doubt the RV350 will beat the Radeon 9500 Pro. I think the 9500 Pro is just going to be discontinued.

 

[Dave Baumann]

Vertex Shader performance is down (as will 'pure' pixel shader performance be [edit] Although, if Rv350 features similar Pixel Shader optimisations as R350 then that migh be a closer run thing as well), but in most gaming situations 9600 PRO should oust the 9500 PRO because of the increased bandwidth. I believe ATI have been testing this quite a lot since the 9600 PRO has to be seen as a good alternative to 9500 PRO.

 

[Sxotty]

Well I got a 9500pro so I am stoked, in any case I was not saying it was way faster, I was simply saying that if it is demoing now than it might actually come out this summer, instead of being announced and come out next march, capish.

 

[darkblu]

Now I know their editors are idiots.

"Not an empty seat in the house"

And their picture shows about a third the seats not occupied.

ATI likes to use popular movies for the themes to center their launches around and in what might be best described as a hat trick launch by ATI, the theme for the 9800/9600/9200 launch was based on the movie "Flight Club".

gosh, i'll never forgive myself omitting that cult movie Flight Club.. doh.

 

[Dave Baumann]

Do you have the measurements for individual chip ?

Not exact, but more or less yes - am I going to share them? Not yet.

 

[surfhurleydude]

According to MuFu, the NV35 tapped out on January 19th(?). I have no reason not to believe him now, because of this showing of the NV35. NV has made some record tape outs to launch, it took 90 days for the GeForce 2... but others have taken up to 120 days (aside from GeForce FX). If this is the case... a launch of the NV35 could be not so far away...and for nVidias sake, I hope it isn't.

 

[BoardBonobo]

They are not though.

And, as far as die sizes go, ATI and NVIDIA are like this:

RV280 (.15u) is smaller than NV34 (.15u)
RV350 (.13u) is smaller than NV31 (.13u)
R350 (.15u) Is about the same as NV30 (.13u)

Ahh, my bad!

 

[LeStoffer]

Dave, any details on the Cine FX implementation on the NV34? Those NV30 shader level pixel pipelines won't come small in silicon real estate, so what's the story?

 

[Mulciber]

Especially if nvidia are using a cheaper, higher yield process for their chips.

They are not though.

And, as far as die sizes go, ATI and NVIDIA are like this:

RV280 (.15u) is smaller than NV34 (.15u)
RV350 (.13u) is smaller than NV31 (.13u)
R350 (.15u) Is about the same as NV30 (.13u)

Performances I've heard (I don't have these inhouse yet) seems to indicate that NV34 is less than Ti4200 and NV31 (ultra) ~= Ti4200 without AA (and but much faster than with). RV350 on the otherhand has a target of Radeon9500 PRO to beat, which puts it well ahead than a Ti4200 in the first place. I'm wondering what NV are doing here - As MuFu states, they are using complex PCB's in the cheap markets, have larger cores and/or using more expensive processes for each of the segments and, apparantly, not stellar performance so far.

These performances I'm hearing though are in current titles, its going to be interesting to see how the two lines stack up with predomanently shader dominated benchmarks. For instance, where NV30 has '8' shader pipes, I heard NV31 has 6 which would make it a reasonable shadr performance chip - assuming they can sort out some the the shader compiler 'twitchiness' we hear about.

Still, the proof will be in the testing - I'm pretty sure we'll get the opportunity to see where the real world performance lies soon.

I'd be willing to bet nVidia will grudgingly lower profits slightly to increase marketshare and mindshare until they can get a respectible performance part back into the market. With any luck (for nVidias sake), these PCBs we're seeing now either a) wont be the ones used in the shipping products or b) will be revised shortly and picked up by the various IHVs.

Correct me if I'm wrong, but did not the NV17 and 18 have DX8 class vertex shaders in hardware, but lack pixel shaders? And now we're looking at their replacement for it having DX9 class pixel shaders but omitting hardware vertex shaders. Hehe, kinda ironic. I for one hope the nv34 performs respectably. It may almost redeam nVidia for what they did to us with the GF4MX....but not completely.

 

[Dave Baumann]

IIRC There were '4 shader pipes' on NV34. Not sure of the precision though.

 

[MuFu]

According to MuFu, the NV35 tapped out on January 19th(?).

The 9th. Can't really vouch for that date myself but JFYI, it comes from a certain *female* B3D-based informant who is *never wrong* and happens to be lying-low right now. I'd tell you her name if I could pronounce it...

Yeah - 4 "shader pipes" in NV34, 6 in NV31. As OpenGL guy mentioned, 6 is strange number considering the supposed, 4x1 configuration.

I'd be willing to bet nVidia will grudgingly lower profits slightly to increase marketshare and mindshare until they can get a respectible performance part back into the market.

My thoughts, exactly.

MuFu.

P.S. Ah yes... "Flight Club", a classic in aviation circles.

 

[Dave Baumann]

FYI, AFAIK NV35 is about the same die size as NV30, despite an increase in complexity.