what will the released NV45 improve over NV40?

[Megadrive1988]

*faster clockspeed, say 500~600 Mhz
*more shader units or shader ops per clock
*more Vertex Shaders, maybe 8 instead of 6
*240~250 million transistors


I can't think too much right now, haven't had my coffee.

 

[PaulS]

I'd be surprised if it was anything much above higher clocked memory (and possibly core), since that will probably be the main area of trouble for NV40 - a lack of bandwidth. The cooling system will likely be revised too, since they'll have some time to work on the layout and improve the process a bit.

 

[John Reynolds]

Yep, the best thing for a fall refresh would be faster RAM.

 

[Joe DeFuria]

Yep, the best thing for a fall refresh would be faster RAM.

Although to be honest, with the high end card supposedly shipping in June or even July, I'm starting to wonder if there will be a "traditional" fall refresh at the high end at all this year...for either company.

 

[Dave Baumann]

remember, the lateness of of the introduction of PCI-Express means that this can, at least, be considered an extra spin over the previous versions. It might even afford you the opportunity to move to a different fab and a slightly different process.

Curious thing is though, the leaked driver indicates that there will be a bridged NV40 board.

 

[LeStoffer]

Although to be honest, with the high end card supposedly shipping in June or even July, I'm starting to wonder if there will be a "traditional" fall refresh at the high end at all this year...for either company.

But Joe, Santa gotta have something new to bring to the Christmas tree! You can bet they will release something new no matter how puny or far fetched the improvements may look.

 

[DemoCoder]

How about a low-K PCI-E version, slightly tweaked for clocking, plus higher RAM?

 

[nobie]

How many high end FX cards were there? 5800 Ultra, 5900 Ultra, 5950 Ultra, 5900 XT. Maybe there will be 6800, 6900, 6950, 6999, 6999.9, etc.

 

[LeStoffer]

How about a low-K PCI-E version, slightly tweaked for clocking, plus higher RAM?

As long as the version number is higher than 6800 Ultra (like 6850 Ultra) its a sure winner!

 

[Joe DeFuria]

remember, the lateness of of the introduction of PCI-Express means that this can, at least, be considered an extra spin over the previous versions. It might even afford you the opportunity to move to a different fab and a slightly different process.

Good point.

Curious thing is though, the leaked driver indicates that there will be a bridged NV40 board.

Why curious? Wasn't that the plan all along? (I know that NV45 is supposedly PCI-E Native, but I always understood that NV45 wouldn't be the first high-end PCI-E part.) I thought nVidia was going to have an entire line-up (initially) of non PCI-E native parts...

 

[Joe DeFuria]

Oh...and Dave...does your presence mean you've touched down in Toronto?

 

[Evildeus]

Oh...and Dave...does your presence mean you've touched down in Toronto?

I would say so

 

[hstewarth]

How about a low-K PCI-E version, slightly tweaked for clocking, plus higher RAM?

Thats is my guess on the NV45.. Also a lot of bandwidth optimizations for PCI-Express. Maybe unlocking stuff in the architexture that they can't handle in AGP versions..

Big guestion after NV40/6800, will there ever be any more AGP versions? or is the NV40/6800 the last one? If so then its a must to get one in my opinion. For fastest possible GPU on AGP type platforms.. Even when I build a PCI-Express machine, I desired my older systems to have the fastest Video card possible.

 

[Nebuchadnezzar]

How many high end FX cards were there? 5800 Ultra, 5900 Ultra, 5950 Ultra, 5900 XT. Maybe there will be 6800, 6900, 6950, 6999, 6999.9, etc.

5900 XT is low end..

 

[Tim Murray]

Low-k, if it's done at TSMC. .11u is right out, unless IBM is a bunch of Supermen (because it's not ready at TSMC) or SOI is ready to roll.

 

[Mendel]

Looking at Nvidia's past track record, this is my speculation:

There will be something broken / some serious design flaw in NV40 and NV45 will fix that.

 

[ZenThought]

Looking at Nvidia's past track record, this is my speculation:

There will be something broken / some serious design flaw in NV40 and NV45 will fix that.

Give me a break. Nv30 was only major fiasco that Nvidia had.
Before then, they had several success generation of chips.

 

[Maintank]

What track record is that? Outside of the NV35 I cant think of a time when they fixed anything major in the GPU on the fall refresh. Most of the time it was simply a core + memory speed increase.

Chances are the NV45 may have some minor tweaks but I doubt it will be a revamp. NV50 will be the next "big" thing IMO.

 

[SsP45]

Oh...and Dave...does your presence mean you've touched down in Toronto?

Toronto? That's only a 4 hour drive for me being in Ottawa. Hey, I could meet Dave!
________
New Mexico Dispensary

 

[Dave Baumann]

Looking at Nvidia's past track record, this is my speculation:

There will be something broken / some serious design flaw in NV40 and NV45 will fix that.

Give me a break. Nv30 was only major fiasco that Nvidia had.
Before then, they had several success generation of chips.

Before running odd on one, think a little. He didn't say that chips weren't successful.

Releasing chips with limited functionlity in some areas is common - its all about your time to market vs key feature risks. Chips often ship with feature not working 100% effectively but these features are glossed over and eventually get fixed in revisions. Two examples in NVIDIA's case are NV10 (Trilinear samplers probably should have been dual texturing samplers, and NSR capabilites were present but du to a bug only some of the functionality exposed in OpenGL - both fixed in NV15) and probably NV20 (Likelyhood is that NV25's "PS1.3" was in NV20 but didn't work 100% effetively, but wasn't worth delaying NV20 in order to get it running).