am i off base?

quest55720

Regular
I was hoping a few of you veterans could help me with some speculation. This is not intended to start a flamewar.
1. After seeing the NV40 previews I see ATI in real trouble. The reason is that it seems almost everthing is CPU or bandwith limited. There is nothing ATI can do about the CPU. For the bandwith I just don't see them sitting on a stock pile of 700+GDR3. It would be a shock to me if ATI had significantly better bandwith saving technology than the NV40. So in the end you have 2 parts with in a few percent of each other. If this is the case the tie must got to the NV40 with the PS3.0 and VS3.0.
2. I can envision the x800 pro being neck and neck with the NV40 in all but the most extreme situations ie 1600x1200 4xAA 8AF. Between the CPU and bandwith limits will even out the playing field. I could be wrong but the only reason for the 800xt is for the numbers game. The extra pipelines will almost go to waste in many situations because of the lack of bandwith.
3. Waiting for the refreshes in 6 months might be the best idea this time around. My guess is in 6 months the ram situation will probably be a lot better. I can see atleast one of the refreshes with 800+ GDR3 ram making a monster difference in getting rid if bandwith problems.

I am i totally off base?
 
i dont think you are. is faster gdd3 really 6 months off? i was hoping for faster mem with the pci-express release :?
 
I know Micron is pumping out GDDR3 up to 833/1666 now to ATI and there board makers. I have a friend in micron R&D and he did a chunk of the GDDR3 IC layout. He told last week that they are selling all GDDR3 up to 833 they can make now.
 
You're completely overlooking the better AA quality the ATI series will offer; gamma-corrected, 6x modes, and perhaps an 8x mode in R42x. Not to mention it potentially running cooler with lower power consumption.
 
BRiT said:
You're completely overlooking the better AA quality the ATI series will offer; gamma-corrected, 6x modes, and perhaps an 8x mode in R42x. Not to mention it potentially running cooler with lower power consumption.

Yes i love the AA of my 9700 pro. Honestly I would guess with both cards running at 1600x1200 with 4x AA the difference will be so small that only a few people would be able to tell the difference. I would almost bet the house that the 420 will run cooler and use less power. The only thing is it does not matter for a 499$ card. Those cards are there to establish your line as the top dog. If performance and IQ are close then it comes to features and ps3.0 and vs3.0 win.
 
You're not off base, but I don't think you're seeing the entire picture either. :)

I would break down the "categories" for consideration like this....at any given price point (and in no partocular order)

1) Performance in "traditional" (non DX9 shader) games
2) Performance in DX9 shader games
3) Anti-Alias quality (includes AA and anisotropic filtering)
4) "Future-proofness"
5) Compatibility
6) Stability

So lets make some gross assumptions about R420...I think these are reasonable, but to be clear, I don't know any of this with any degree of certainty:

A) Same number of rendering pipes as NV40 at a given price point
B) Similar performance per pipe per clock as NV40 on any given app
C) Does not support VS/PS 3.0
D) Same aniso as R300
E) At least as good (6X sparse grid AA)
F) A bit better available bandwidth (600 Mhz for XT, for example?)
G) Higher in clock rate of the core. (500 Mhz?)
H) Lower power consumption than NV40

Now if we go back to my list of "things to consider", it would work out like this:

1) Tie (but slight advantage to R420 because of slight memory bandwidth increase)
2) Advantage R420 - bandwidth limitations may not come into play in DX9 shader heavy apps. So if the R420 core is clocked 25% higher, that can be a significant increase in DX9 shader limited apps
3) Advantage R420 - can be slight to significant depending on the AA that R420 supports. (NV40 could have a slight Aniso quality advantage if the option for the "old" aniso quality is made available, at a performance penalty though.)
4) Advantge NV40, due to advanced shader support
5) Tie (both of these architectures will be supported by developers)
6) Slight advantage to R420...depending on power consumption difference.

I think the major thing you're missing is number 2. If R420 ships at 500 Mhz or higher, and If it's DX9 shader performance is on par, clock for clock, with NV40, that means that R420 has a good chance at significant performance increases in PS 2.0 bound apps.

That, plus factoring in things like power consumption / stability, increased image quality (AA), can outweigh SM 3.0 support of NV40.
 
It is impossible to jump to any conclusions at the moment. We can only wait for the reviews to show up, with both cards using more mature drivers.
 
I have to disagree with point 4 there Joe. That entirely depends on developers, if they should all move to PS 3.0 within the life-span of these card, then sure, you're right. If they don't, then the most "future-proof" card is the one that pulls away once you've dealt with the CPU bottleneck, and that looks to be the R420, at least from clocks alone.
 
Reverend said:
Joe DeFuria said:
4) "Future-proofness"
4) Advantge NV40, due to advanced shader support
Does anyone know how useful is its SM 3.0 support (i.e. performance)? Or are you simply talking about a checkbox feature (which wouldn't be like you)?

I'm talking more or less from a checkbox standpoint. In other words, "all else being equal", if one card has support for PS 3.0, and another card doesn't, the PS 3.0 one is preferable.

We have no way of truly evaluating if PS 3.0 will or won't be "useful" over the life of the card, but that doesn't mean that if everything else about the 2 cards is "very close", that PS 3.0 support isn't a plus and can't sway the decision.

As we know though, things are rarely ever "equal" in other aspects. So you have to look at the whole picutre, which I've tried to define above. I would certainly put "significant performance improvement" of PS 2.0 shaders as more important than PS 3.0 support. But of course, we don't know anything about relative performance yet.
 
Dio said:
We sure do like to grasp the nettle by the horns and mix our metaphors here.
Hey, don't go changing your horses midstream like that or you'll end up more skittish than a long-tailed cat in a room full of rocking chairs.
 
Joe DeFuria said:
As we know though, things are rarely ever "equal" in other aspects. So you have to look at the whole picutre, which I've tried to define above. I would certainly put "significant performance improvement" of PS 2.0 shaders as more important than PS 3.0 support. But of course, we don't know anything about relative performance yet.

What about "shadow" performance in f.e Doom3 ? (Ultrashadow 2)

We don't know if the R420 will have something like that of course and it might be that the only engine to take advantage of it will be Doom3. But it could potentially be an advantage for the NV40.
 
Joe DeFuria said:
Reverend said:
Joe DeFuria said:
4) "Future-proofness"
4) Advantge NV40, due to advanced shader support
Does anyone know how useful is its SM 3.0 support (i.e. performance)? Or are you simply talking about a checkbox feature (which wouldn't be like you)?

I'm talking more or less from a checkbox standpoint. In other words, "all else being equal", if one card has support for PS 3.0, and another card doesn't, the PS 3.0 one is preferable.

We have no way of truly evaluating if PS 3.0 will or won't be "useful" over the life of the card, but that doesn't mean that if everything else about the 2 cards is "very close", that PS 3.0 support isn't a plus and can't sway the decision.
Agreed, and that was a big argument in favor of R300. NV40 now seems to have good SM2.0 performance, so anything above that should be gravy. Devs in particular should like the ability to use one card for both SM2.0 and 3.0 this early in the game, no? Even if the number of SM3.0 cards will be limited for the near future, when more appear, devs will be ready to take advantage of them.
 
Back
Top