NV45 spotted @ Anand

[Joe DeFuria]

Well, if we all knew that we should also have known that there were not much reason to be a real refresh....

The only exception could be from memory.

If Samsung or Micron can get 800 Mhz GDDR-3 out in time for a "fall line-up", just pairing that memory with the same 6800 Ultra and especially the X800 XT-PE, which is more bandwidth starved, would be a meaningful "refresh" product.

 

[Dave Baumann]

Maybe the product cycles really are slowing. (Good job too IMO, if and when it happens).

IMO, this is an innevitability, purely because of chip sizes - its not just cycles slowing, but "refreshes" will begin to look more and more like clock bumped refinements and true refreshes will take longer to achieve (as with R300 and R420).

I still say the rule of thumb from Tape-out to production is somewhere in the region of 6 months, so technically they are taping out their refreshes at the time they are still trying to finish up qualification and ramp to production their previous generation. Silicon cycle times are just getting longer and longer.

 

[Bjorn]

If Samsung or Micron can get 800 Mhz GDDR-3 out in time for a "fall line-up", just pairing that memory with the same 6800 Ultra and especially the X800 XT-PE, which is more bandwidth starved, would be a meaningful "refresh" product.

Remains to be seen if bandwidth is the biggest problem at that time though since Half Life 2, Doom3 and perhaps other more shader limited games will be out by then.

 

[Evildeus]

If Samsung or Micron can get 800 Mhz GDDR-3 out in time for a "fall line-up", just pairing that memory with the same 6800 Ultra and especially the X800 XT-PE, which is more bandwidth starved, would be a meaningful "refresh" product.

Remains to be seen if bandwidth is the biggest problem at that time though since Half Life 2, Doom3 and perhaps other more shader limited games will be out by then.

Yeah, even on 6*AA on X800XT, BW doesn't seem to be the limiting factor. Shader power seems to be the most demanding ATM.

 

[Joe DeFuria]

Remains to be seen if bandwidth is the biggest problem at that time though since Half Life 2, Doom3 and perhaps other more shader limited games will be out by then.

Yeah, even on 6*AA on X800XT, BW doesn't seem to be the limiting factor. Shader power seems to be the most demanding ATM.[/quote]

It all depends on the application, and the AA setting.

Where did you see that at 6X AA, bandwidth isn't the limiting factor on the X800 XT?

 

[Bjorn]

It all depends on the application, and the AA setting.

That's true. But it's going to be hard to justify the extra cost of the memory if there's no benefits in the games that need the performance boost the most.

 

[Joe DeFuria]

That's true. But it's going to be hard to justify the extra cost of the memory if there's no benefits in the games that need the performance boost the most.

And you've just summed up the entire "Why ATI stuck with PS 2.0 and didn't go to PS 3.0 argument." Just replace "memory" with "pixel shade pipeline".

 

[Bjorn]

That's true. But it's going to be hard to justify the extra cost of the memory if there's no benefits in the games that need the performance boost the most.

And you've just summed up the entire "Why ATI stuck with PS 2.0 and didn't go to PS 3.0 argument." Just replace "memory" with "pixel shade pipeline".

That's a bit of a different thing though. In the SM3.0/FP blending case, it's possible that you might get some benefits of the extra features. Especially FP blending (perhaps it'll even be supported in Half Life 2). If the boost in memory speed doesn't give you any benefits in games like f.e Half Life 2 then it's likely that it'll be even more of a moot point in later games.

 

[Joe DeFuria]

That's a bit of a different thing though. In the SM3.0/FP blending case, it's possible that you might get some benefits of the extra features.

It's no different.

It's certainly possible to have some benefits from extra bandwidth.

 

[Evildeus]

Well, many bench didn't provide much differencies between 4* and 6* AA. For exemple:
http://www.xbitlabs.com/articles/video/display/r420_21.html
http://www.beyond3d.com/reviews/ati/r420_x800/index.php?p=26

 

[Bjorn]

That's a bit of a different thing though. In the SM3.0/FP blending case, it's possible that you might get some benefits of the extra features.

It's no different.

It's certainly possible to have some benefits from extra bandwidth.

Of course, i was thinking more along the lines of "benefits in future games". Not that i belive that all games released after Half Life 2 will be that shader limited but i doubt that there'll be that much interest in any super expensive 800 MHz mem cards if there are no benefits in games like that. The other thing is the "1 Gb of video mem for full detail" in Unreal Engine 3 games. Makes me believe that it might actually be more interesting with additional RAM then higher speed. Though i guess it should be much more interesting with high sped RAM when it comes to the GeForce 6800 since it supports FP blending.

 

[Joe DeFuria]

Look at the previous page, particularly, the last chart.

http://www.beyond3d.com/reviews/ati/r420_x800/index.php?p=25

 

[Bjorn]

Look at the previous page, particularly, the last chart.

http://www.beyond3d.com/reviews/ati/r420_x800/index.php?p=25

Could indicate that Half Life 2 won't need it but games with heavy stencil shadowing and not so much shader usage, like Doom 3 f.e might gain a lot of performance with more banchwidth. Though Doom3 will now use materials which might change the requirements a bit.

Anyway, we'll see what happens. Doom 3 and Half Life 2 will supposedly be out this summer

 

[Evildeus]

Look at the previous page, particularly, the last chart.

http://www.beyond3d.com/reviews/ati/r420_x800/index.php?p=25

Yeah, but on the same page, you can easily find out that it's more shader limited than BW limited.

 

[Tim Murray]

Hm. Does the NV45 chip itself look small to anyone else?

 

[Joe DeFuria]

The fact of the matter is, in some cases bandwidth will be a benefit, and in other cases it won't.

What's the big deal?

 

[Joe DeFuria]

Hm. Does the NV45 chip itself look small to anyone else?

Hard to tell. Really need some reference. (Even if the chip looks smaller relative to the package, which I'm not sure it does, the NV45 package may be larger than the NV40 one...)

 

[Evildeus]

The fact of the matter is, in some cases bandwidth will be a benefit, and in other cases it won't.

What's the big deal?

The big deal is that's it's no more a big deal to have much more BW ATM

 

[Evildeus]

Hm. Does the NV45 chip itself look small to anyone else?

It does seem small, but i can't find any pic to compare

 

[anaqer]

Regarding bandwith issues... the XTs and Pros I've seen so far all had 1.6ns and 2.0ns modules, respectively. Now if BW was such an issue, why would reference cards be underclocked like this (1120MHz instead of 1250, 900MHz instead of 1000MHz)?