R423 will have 16 pipes(16x1) and will own nv40

[Smokey]

God, I hope I can come back here in 2-4 weeks, and quote some of the crap people have posted :D

Fact is, you dont know which card is better, and what the performance is going to be of each card ;)

[nrdstrm]

I agree that we don't know wich is better yet. I am rooting for NV40 (I'm a bit of a fanboy I guess), but if Nvidia dosn't get their crap together with the NV40 and own the R420, I'm switching to ATI. I got burned twice with the 5800U and 5900U (Bought them before most benchmarks were released). Granted it's my fault for being an early adopter, so I'm gonna wait it out this time and see benchmarks first.
Nrdstrm

[MUYA]

I dun think...anyy card will own the other outright. Trade a few jabs here and there maybe :box:

[sxotty]

The only card I got burned by was a rage 128, and that made me a fan of Nv I guess...or really it just made me smart enough to wait till I saw reputable benchmarks about a piece of hardware, and that is why I got a 9500pro followed by a 9800pro :).

[Carbon Unit]

Im not going to be a fanboy this time around with Nvidia and ATI, wich ever is faster, the NV40 or 420 I will get after the initial drivers come out for both cards (aaa) :kill:

[zakelwe]

The nvidia card that most caught me out was the 5600U, I really thought it was going to be faster than my GF4 4200 but it turned out to be worse in any games nvidia has not optimised for. It was like 20% down in some things and in optimised stuff only 5% up, even though the clock speed was 450ish for the gpu and the clock speed for the 4200 was 320. At this point I realised the reduction to 4x1 pipelines was not a good thing :)

The 5700U was a lot better though.

I think I might go nv40 this time even if it is slower, assuming IQ is a lot better. I don't mind the card being slower than r420 as long as it is still x2 faster than current.

Regards

Andy

[Geforce4ti4200]

heh you still like nvidia eh? I doubt the iq will be better, nvidias whole point is to optimize by exchanging iq for performance while the r420 eats it and has better iq. I ditched nvidia once their fx5800 came. the ti4200 was the best card nvidia ever made

[zakelwe]

heh you still like nvidia eh? I doubt the iq will be better, nvidias whole point is to optimize by exchanging iq for performance while the r420 eats it and has better iq. I ditched nvidia once their fx5800 came. the ti4200 was the best card nvidia ever made


Yeah I still like nvidia ..sucka eh ? I do have 3 ati cards but they are doing duty where they belong, in my DF folding servers ;)

Yeah the 4200ti was a very good card, Ati has had 9700 / 9800 which were / are best since then, but I wait to see how the next battle goes.

Regards

Andy

[Geforce4ti4200]

heh those ati cards must be folding a storm indeed. ever had a radeon9700 and up? there is speculation the nv40 can break 40k easy overclocked with overclocked a64 cpu even if its 8x2 or 16x1 same thing. 3dmark 2001 is one of the benchmarks that is possible to run 4x2 as 8x1, etc. thats why my 4x2 ti4200 does so good against the 8x1 9500 pros, but it gets owned in newer games where my ti4200's 4x2 can NOT act as 8x1. seems like if the nv40 is 8x2 it will own the r420 in 3dmark 2001, doom3 and certain other games but the r420 will own in most benchmarks :D oh one more thing, the r420 might be clocked higher than the nv40 so even a 12x1 r420 could come very close to the nv40 in 3dmark2001, my fav benchmark. no matter, once the r423 or r450 comes with 16 pipes, itll finish off the nv40 uterely ;)

[Nutty]

no matter, once the r423 or r450 comes with 16 pipes, itll finish off the nv40 uterely
but NV40 already has 16 pipes, and by that time NV45 will be on the table to compete with R423, and R450.

[MUYA]

the inq also said itll be 16x1 in certain cases. It remains to be seen if its really 16 pipes or 8x2 acting as 16x1 in certain cases :spank: :kill:
The days GPU having individual pipes where only one pixel is prcoessed with X amount of textures applied/processed to it by a TMU is gone! u know AxB thing.

The NV3X is one big pipeline as 3dcenter showed it. R3XX is (i think) also one pipeline which works with 2 quads. All GPUs I think nowawdays work on quads 2x2 pixel blocks (regions) or something to that effect.

Again the NV40 is *rumoured* to push 16 pixels per pass therefore it seems like a 16X1 ie because it pushes 16 pixels per pass and is processed with one texture effect or whatever.

[Wenzula]

A friend of my told me that the "Serious Sam 2" Demo, which has been shown at GDC was running on a r420... By the way, it looks great!

[Evildeus]

Yeah it seems so (great and on R420 :))

[Fiver]

the reason why r420 is beating the nv40 in unreal 3.....

unreal 3 engine makes heavy use of directx 9..much more than half life 2
unreal 3 engine may need a directx 9 compliant card to run it, if the card is not directx 9 compliant it wont work...

r3xx ownz nv3x in half life 2, which uses a few directx 9 features....
so when u hav a game engine like unreal 3 that makes HEAVY HEAVY use of directx9, its obvious that the performance drop seen in unreal 3 would be a multiple of the performance drop seen in half life 2..

this also shows that the nv40 will not perform well with directx9 games...as compared to the r4xx series, and maybe even the r3xx series.

Your forgeting that NV40 is rumored to have 16 pipes. That theoretically means 4x the performance in Pixel Shaders . . . . . . Unless they lowered the pixel processing power per pipe to save trannies. They also supposedly fixed ( or at least lowered the effects of ) the register problem, which should mean less "bubbles", which should mean higher performance.

[-=DVS=-]

Your forgeting that NV40 is rumored to have 16 pipes. That theoretically means 4x the performance in Pixel Shaders . . . . . . Unless they lowered the pixel processing power per pipe to save trannies. They also supposedly fixed ( or at least lowered the effects of ) the register problem, which should mean less "bubbles", which should mean higher performance.

Yes but what if its not actually true 16 pipe :thinker: but 8 pipe with conditional 16x/32x with lower performance in one type of games but good in other :screwy: hmm...

[Geforce4ti4200]

if its 8 pipes, itll get owned by the r420, strait and simple. itll be comparable to a 9800xt

[zakelwe]

if its 8 pipes, itll get owned by the r420, strait and simple. itll be comparable to a 9800xt

Are you suggesting the 2x performance increase over the 5900/5950 is not going to happen and in fact the nv40 is going to increase in performance from 5900/5959 up to 9800XT speeds ?

Regards

Andy

[Geforce4ti4200]

a 9700 pro owns the fx5950 in dx9. so yea the nv40 would own the fx5950, be about twice as fast or a little ahead of a 9800xt

[The Hip]

Sorry, but I think the question ought to be: by HOW much is the NV40 gonna own R420/423....

Take a gander here:
http://www.beyond3d.com/forum/viewtopic.php?t=11153&postdays=0&postorder=asc&start=0

And you'll find a link to the whole presentation there as well.

Oops!

Snippet from there:

-------------------------------------------------
The 3.0 shader model

Requires 32 bit floats throughout the pipeline
But that’s not necessarily full IEEE 754...
With it’s -0.0s, NANs and INFINITYs etc

It’s generally agreed that SM3.0 requires a move to 32 bit floats (some still think it’s not)
But - The spec requires it...

Steer people away from flow control in ps3.0 because we expect it to hurt badly. [Also it’s the main extra feature on NV40 vs R420 so let’s discourage people from using it until R5xx shows up with decent performance...]
-----------------------------------------------------

[MUYA]

It is no way near comparable to a r360. Just get this pipeline thing out of your head folks. As the r360 or r3XX doesn't have 8 pipelines. It has one, and so does the nv3X series. In fact Go search for tech info at b3d forums. All modern GPUs that DX8 compatible (i am quoting DaveB here) are GPUs that process quads in one pipeline. Therefore just saying or quoting a certain GPU acts like it has so much pipes and so much TMUs are not going to facilitate a apples to apples comparision. Just because you are hearing these inkwell stories about certain GPUs having certain configuration of pipes and basing your info on that is not tthe way to discuss relative performance to each other. This is what I have gathered from reading various posta nd correspondence with ppl

The R360/350/300 posses 1 pipeline which processes 2 quads per pass but thats all in theory when certains OPs are required it processes less. The NV3X is one pipeline too but the amount of quads it process is also governed by what OPs are required to be applied to pxels in the quad. The performance were lower for NV3X because of bubbles appearing in the pipeline due to penalties arising from certain conditions and hence many ppl deduced that was the reason for nv3X less than stellar PS2 performance compared to ATIs line.

Now NV4X is (rumoured) to be a heavily modified and essentially (rumoured to) capable of pushing 16 pixels per pass...that should equate to 4 quads. Thats all in theory and of course there will be inherent performance lost due to certain penalties when certain OPs or I think filtering techniques are called for but the feeling around the boards are that the new nv40 is more efficient and less bubbles are likely and that will equate to faster performance in PS.

R420 is a step from r3XX, it is also evolutionary (it seems though not much is known) than revolutionary just like the NV40. but how many quads it's able to prcoess per pass I dunno. Maybe 3 or 4. But the penalties are less for PS for r420 inherited from r3XX series than the nv3X and maybe NV40 but the amounts of quads being pushed the pipeline per pass is still limited by what OPs are required or to be processed. Therefore without any direct benches of theoreticals we cannot assess which will be faster. Lets wait till launch and when the previews come out with both both cards benched and then discuss those.

Just my 2c

Edit : of course it's more complicated than that but thats I think basic unerlying principles :p

[Geforce4ti4200]

"And you'll find a link to the whole presentation there as well."


says itll own the nv40. also there will be a 512mb at the end of this year. wow!

[-=DVS=-]

Alright so talking pipelines is useless thx alot Muya :|

So lets talk Quads :D

R3xx = Efficient 1 quad per cycle right ?
NV3x = Inefficient 1 quad per cycle ?

and speculation

NV40 = 4 quads ? thats what 4x time performance of NV3x?
R420 = ??? 2 ? 3 ???4 ??????????? there is hardly any rumors on this baby

Megahertz for core and memory have influence here also , but heck speculate is getting harder and harder these days :banghead:

[zakelwe]

Alright so talking pipelines is useless thx alot Muya :|

So lets talk Quads :D

R3xx = Efficient 1 quad per cycle right ?
NV3x = Inefficient 1 quad per cycle ?

and speculation

NV40 = 4 quads ? thats what 4x time performance of NV3x?
R420 = ??? 2 ? 3 ???4 ??????????? there is hardly any rumors on this baby

Megahertz for core and memory have influence here also , but heck speculate is getting harder and harder these days :banghead:

Sure is, that is why it is best to wait until we have the actual results.

If they are quadroupling the gpu throughput but not even doubling the memory bandwidth will the gpu not become starved ? Maybe that is one of the reasons why it does times as many pixels but only has twice the performance ?

Regards

Andy

[MUYA]

"And you'll find a link to the whole presentation there as well."


says itll own the nv40. also there will be a 512mb at the end of this year. wow!
I find that highly speculative. There is a lot of talk back and forth of the validity of the statements povided.

[MUYA]

R3XX did 2 quads....most of the time of course it was said to vary with degrees of filtering and also conditions etc

NV3X dunno, no one has actually said anything about it but it is less efficient it was said than the r3XX