Latest nV info

Depends on whether this means 4 texels with trilinear filtering per clock or not.

 

A bout the 100 000 first VC built, it seems that the builder will be.... Selectron according to this "scoop" (in french sorry )

http://www.clubic.com/n/n7675.html

 

No surprise. I posted back during COMDEX that I sat next to an NVidia senior product manager carrying a box from Selectron.

 

Ok

 

When you see someone working for ATI hoping the NV31 is a good card, you know nVidia is in trouble

It's indeed very unlikely the NV31 will be able to beat the GF4 Ti4600 at 300Mhz... It barely got any advantage over it, and it got the huge texel disadvantages.
MuFu said somewhere that the NV31 original target clock was of about 400Mhz. The more efficient architecture could have given it equal performance even in multitexturing cases, but there's no chances of that at 300Mhz...
So now, the only performance advantage over the NV25 is Adaptive Aniso & Color Compression which will increase AA Speed. So in games which are too intensive to let you enable either, I guess nVidia is out of luck...


Uttar

 

From the latest commits by an NVidia employee to the XFree CVS, one can see a pretty long list products, current and future.

Interesting additions lately (PCI ID and the description):

{ 0x10DE0186, "GeForce4 448 Go" },
{ 0x10DE0187, "GeForce4 488 Go" },

AGP 8x versions of GF4 Go?

{ 0x10DE0300, "0x0300" },
{ 0x10DE0301, "GeForce FX 5800 Ultra" },
{ 0x10DE0302, "GeForce FX 5800" },

So there is an ultra version, but what is the 0x0300...

{ 0x10DE0308, "Quadro FX 2000" },
{ 0x10DE0309, "Quadro FX 1000" },

No surprises here as they have been launched already.

{ 0x10DE0311, "0x0311" },
{ 0x10DE0312, "0x0312" },
{ 0x10DE0316, "0x0316" },
{ 0x10DE0317, "0x0317" },
{ 0x10DE0318, "0x0318" },
{ 0x10DE0319, "0x0319" },
{ 0x10DE031A, "0x031A" },
{ 0x10DE031B, "0x031B" },
{ 0x10DE031C, "0x031C" },
{ 0x10DE031D, "0x031D" },
{ 0x10DE031E, "0x031E" },
{ 0x10DE031F, "0x031F" },

Quite a few versions of NV31. Note that not all numbers are there, probably only ones they really have. From the code it seems the 0x316-0x31f are mobile chipsets.

{ 0x10DE0321, "0x0321" },
{ 0x10DE0322, "0x0322" },
{ 0x10DE0323, "0x0323" },
{ 0x10DE0326, "0x0326" },
{ 0x10DE032A, "0x032A" },
{ 0x10DE032B, "0x032B" },
{ 0x10DE032E, "0x032E" },

No one has been talking about NV32 so far? 0x326 and 0x32e seem to be again mobile.

Also mentions of NV33 are in the code, but nothing on the NVKnownChipsets list from which the above are from. But where is NV34?

Just food for thought

 

Just "GeForceFX" if I remember rightly.

Hmm... maybe they are just arbitary assignments or placeholders (?). I have never heard of NV32. NV31, NV33 and NV34, yes, although it NV33 is a bit of a mystery at the moment.

MuFu.

 

Makes sense. Geforce2 Go chipsets in laptops were always AGP, so it's another checkbox for Dells and whatnot to have "AGP 8X" on the side of the box for integrated mobility video chips.

Between the newer Radeon Mobility and Geforce4 Go chips, it looks like laptops are finally going to have somewhat acceptable 3D game performance. This is a good thing.

 

I can't see why they would add placeholders and actual code even (from which I deduced the mobile chip part) unless they had actual chips themselves and were bringing them to market sooner or later. Supporting imaginary chips in such a marginal (so far) product as XFree would be, well... interesting. As I said, the additions were made by NVidia employee.

 

I suppose so. Those long lists of IDs (like the 12 for NV31) could be for ES's, in which case I'm sure they'll end up as only 2-3 entries once the product is launched. Not entirely sure that is what they are for but it seems to be a given pattern for products in developement, i.e. a large range of Device_IDs are allocated during RnD and the only ones that end up in final code are the ones that correspond to qualified reference samples. Anybody know for sure?

NV35 and NV36 should pop up on there soon - they aren't as far off as most people seem to think. No idea what NV36 is, but NV35 sounds extremely

MuFu.

 

The following can be found in the 42.70 drivers:

NVidia.Nv30.1 = "NVIDIA NV30"
NVidia.Nv30.2 = "NVIDIA GeForce FX 5800 Ultra"
NVidia.Nv30.3 = "NVIDIA GeForce FX 5800"
NVidia.Nv30GL.1 = "NVIDIA Quadro FX 2000"
NVidia.Nv30GL.2 = "NVIDIA Quadro FX 1000"

NVidia.Nv31.1 = "NVIDIA NV31"
NVidia.Nv31.2 = "NVIDIA NV31 "
NVidia.Nv31GL.1 = "NVIDIA NV31GL"
NVidia.Nv31GL.2 = "NVIDIA NV31GL "

NVidia.Nv34.2 = "NVIDIA NV34"
NVidia.Nv34.3 = "NVIDIA NV34 "
NVidia.Nv34.4 = "NVIDIA NV34 "
NVidia.Nv34GL.3 = "NVIDIA NV34GL"
NVidia.Nv34GL.4 = "NVIDIA NV34GL "

%NVidia.Nv30.1% = nv4, PCIVEN_10DE&DEV_0300
%NVidia.Nv30.2% = nv4, PCIVEN_10DE&DEV_0301
%NVidia.Nv30.3% = nv4, PCIVEN_10DE&DEV_0302 %NVidia.Nv30GL.1% = nv4_WSApps, PCIVEN_10DE&DEV_0308
%NVidia.Nv30GL.2% = nv4_WSApps, PCIVEN_10DE&DEV_0309

%NVidia.Nv31.1% = nv4, PCIVEN_10DE&DEV_0311
%NVidia.Nv31.2% = nv4, PCIVEN_10DE&DEV_0312
%NVidia.Nv31GL.1% = nv4_WSApps, PCIVEN_10DE&DEV_0318
%NVidia.Nv31GL.2% = nv4_WSApps, PCIVEN_10DE&DEV_0319

%NVidia.Nv34.2% = nv4, PCIVEN_10DE&DEV_0321
%NVidia.Nv34.3% = nv4, PCIVEN_10DE&DEV_0322
%NVidia.Nv34.4% = nv4, PCIVEN_10DE&DEV_0323
%NVidia.Nv34GL.3% = nv4_WSApps, PCIVEN_10DE&DEV_032A
%NVidia.Nv34GL.4% = nv4_WSApps, PCIVEN_10DE&DEV_032B

So that "32" is nothing more than the NV34.

It would seem a lot more logical for the NV30.1 to be a slightly faster GFFX to counter the R350. I'd guess the GFFX 5800 Ultra will be at $399 and that one will be at $499 ( but it'll be released slightly later )
But then again, we've already seen numbers in the drivers never being released.

The NV34 is a lot more complex the understand... If all numbers were filled, we'd be getting 8 models!
However, there IS an explanation to this. Notice how it's always the high-end NV34s which are missing ( the lowest numbers ) ?
And multiple rumors are saying us that nVidia is going to have to clock the NV31 lower than originally expected. So, the high-end NV34s might have been as fast as the low-end end NV31s! And nVidia obviously couldn't accept that. The easy solution? Discontinue the high-end NV34s.

Obviously, those drivers don't give any info on mobile solutions. And, sadly, there's nothing about NV33... Ah well, we're still in the dark about that one


Uttar

 

The reviewers got their review boards yesterday and today, monday is probably the big day.

 

Thanks.

Will TSMC's low-K process be used to fab NV35? If it won't, then the numbers I have don't make any sense. :?

MuFu.

 

Please, give us the numbers and let us decide if they do make sense or don't

ciao,
Marco