PDA

View Full Version : R423 will have 16 pipes(16x1) and will own nv40


Pages : [1] 2 3 4 5

lokeshray
03-23-04, 03:01 PM
pretty obvious here
r420 core needs a few more revisions before it can use all 16 pipes
till then it will be 12 pipes...

r423 will be introduced soon after nv40...
nv40 will have a lead..albeit shortlived..
kinda similar to the short lived lead the ati radeon 64mb vivo had over the geforce 2 gts........

the german site did mention r423 will have 16 pipes.

THINGS WILL GET REALLY INTERESTING WHEN THE NV40 REFRESH IS OUT.

the german site did say r500 project is progressed very well

Wenzula
03-23-04, 03:12 PM
Which site? Do you have a link? But i heard the rumor of a 16x1 r423 too...

Geforce4ti4200
03-23-04, 04:04 PM
the nv40 would never own the r420 if it doesnt have 16 real pipes and its 8x2 accroding to the most recent rumors. sure it might own in select games with heavy optimizing, but the r420 has real power, no optimizing BS. also if its true about clocks as high as 600/1100, ati will be miles ahead in the same way a 9700 pro was over a ti4200

Nutty
03-23-04, 04:23 PM
but the r420 has real power, no optimizing BS. also if its true about clocks as high as 600/1100, ati will be miles ahead in the same way a 9700 pro was over a ti4200


LOL!! no optimizing eh? Brute force inefficiency then?

Why is it when ppl said NV40 will be 16 pipes at 500, its like "Woaa, I'd like to see how they'll do that!" But now its R420 with 16 pipes at 600Mhz, its like "Yeah thats easy, its gonna own everything." :rolleyes2

Of course the 9700 was miles ahead of the 4200, the 4200 was the low-end of the GF4 TI range..

and its 8x2 accroding to the most recent rumors.
Most recent? What about Baron and Vegetta's deductions that it was definitly 16 pipes from fillrate tests? Fillrate tests which use textures, so if their deductions are correct, its a 16x1 minimum.

I guess the real problem is trying to describe these gpu's in such simple terms as pipes x texture units, as its probably nothing like that now.

We'll all see soon enuff tho eh..

GalaxyFX
03-23-04, 05:07 PM
That is the great thing about rumors. Everybody can read into them what they want.

Fanatics say Ati ownz, while Nvidiots are even more eager that Nvidia regains the much desired performance crown...

Nvidia ownz...

Geforce4ti4200
03-23-04, 05:17 PM
"Most recent? What about Baron and Vegetta's deductions that it was definitly 16 pipes from fillrate tests? Fillrate tests which use textures, so if their deductions are correct, its a 16x1 minimum."


If I stand correct, 8x2 can act like 16x1 in certain situations....

jimmyjames123
03-23-04, 05:26 PM
That's not exactly what they are saying though. They are saying 16x1/32x0

Geforce4ti4200
03-23-04, 05:29 PM
that will be very good if its 16 real pipes, im sure they know for real, but alot of people are thinking its 8x2

Wenzula
03-23-04, 05:30 PM
What matters most to me is, that even the ati aibs said, that the nv40 will be faster this time. I am not a nividia fanboy (i own a readeon 9800 pro). Maybe this will change with the r423, but the real deal will be the r500, which is nearly complete in terms of the chipdesign.

Moreover my soruces said, that the r420 and 423 will be "just" a wider r300. So, ati did not change much. Whereas nvidia did a whole redesign.

Hope to see real benchmarks too. At least after april the 13th :-)

GlowStick
03-23-04, 06:24 PM
What matters most to me is, that even the ati aibs said, that the nv40 will be faster this time. I am not a nividia fanboy (i own a readeon 9800 pro). Maybe this will change with the r423, but the real deal will be the r500, which is nearly complete in terms of the chipdesign.

Moreover my soruces said, that the r420 and 423 will be "just" a wider r300. So, ati did not change much. Whereas nvidia did a whole redesign.

Hope to see real benchmarks too. At least after april the 13th :-)
Benches are gaurenteed on the 13th because around 7 will be released to the general public via contest winners.

Geforce4ti4200
03-23-04, 07:00 PM
depends if any of the 7 winners choose to benchmark and realese info, also they may be biased towards nvidia so we best wait for review sites. and the nv40 will not own the r420 unless its 16 real pipes

Fiver
03-23-04, 07:09 PM
that will be very good if its 16 real pipes, im sure they know for real, but alot of people are thinking its 8x2
So you're listening to theinq on this one :wtf: ?

Fiver
03-23-04, 07:13 PM
If they are both 16 pipe(and nvidia still has 1 alu and two mini's) then Nv40 will have a moderate PS 2.0 lead on 420(assuming the 420 still has the same setup of one full alu and one mini and reg problems are fixed). How is this getting "owned" :screwy: ?

Geforce4ti4200
03-23-04, 07:15 PM
So you're listening to theinq on this one :wtf: ?




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:

GlowStick
03-23-04, 08:00 PM
depends if any of the 7 winners choose to benchmark and realese info, also they may be biased towards nvidia so we best wait for review sites. and the nv40 will not own the r420 unless its 16 real pipes
Well i am sure if any GAMERS win the cards their will be benches oh yes.

I honestly dont think that they would 'lie' about the numbers, but i do forsee this happeing

After the contest forums will be flooded with threads like

"OMG, i just won a nv40' and "OMG my best friend just won a NV40" etc
and oviously they just post BS

Lezmaka
03-23-04, 08:13 PM
You know what I heard? R525.21 will have 29.5932 pipes and run at 24.57Thz!

lokeshray
03-23-04, 09:15 PM
someone did post this before,
this is the german site i was talking about,
the original german link is here

The German Site (http://www.withingames.net/?show=articles&type=showarticle&id=83&site=7)

i translated this, with babelfish and here is the translation.

http://picserver.org/view_image.php/RR19S66G4RL0

Nv40
03-23-04, 09:32 PM
What matters most to me is, that even the ati aibs said, that the nv40 will be faster this time. I am not a nividia fanboy (i own a readeon 9800 pro). Maybe this will change with the r423, but the real deal will be the r500, which is nearly complete in terms of the chipdesign.

Moreover my soruces said, that the r420 and 423 will be "just" a wider r300. So, ati did not change much. Whereas nvidia did a whole redesign.

Hope to see real benchmarks too. At least after april the 13th :-)

but the real deal will be the r500, which is nearly complete in terms of the chipdesign.
----------------------------------------------------------
not surpsingly to me.. see my other post.. ;)

a -wider- R300?

512 bit bus Gpu ?

when the R423 will be released?

EMunEeE
03-23-04, 09:51 PM
LOL!! no optimizing eh? Brute force inefficiency then?

Why is it when ppl said NV40 will be 16 pipes at 500, its like "Woaa, I'd like to see how they'll do that!" But now its R420 with 16 pipes at 600Mhz, its like "Yeah thats easy, its gonna own everything." :rolleyes2

Of course the 9700 was miles ahead of the 4200, the 4200 was the low-end of the GF4 TI range..


Most recent? What about Baron and Vegetta's deductions that it was definitly 16 pipes from fillrate tests? Fillrate tests which use textures, so if their deductions are correct, its a 16x1 minimum.

I guess the real problem is trying to describe these gpu's in such simple terms as pipes x texture units, as its probably nothing like that now.

We'll all see soon enuff tho eh..

Because, look at the engineering masterpiece that was the R300/R350/R360, all on the .15u process. Where as NVidia had a.13 process, but stuggled with heat and power problems on the NV30. Going back, ATI has engineered some great chips...(R100, R200, the Rage 128 was also a good piece of work, software however held ATI back in the past).

Geforce4ti4200
03-23-04, 10:51 PM
people will pretend they won the card or their friend's friend won one and make up numbers. well one thing for sure, the nda probably will expire by then and we will know for sure if its really 16 pipes. and in late may itll be in stores for sale and reviews

nelg
03-24-04, 01:39 AM
Digitimes says the nV40 is eight pipes.
LINK (http://www.digitimes.com/NewsShow/Article.asp?datePublish=2004/03/24&pages=A1&seq=2)

Geforce4ti4200
03-24-04, 02:13 AM
my god, poor nvidia. atis gonna have *16* real pipes either on their r423 or r450. Itll be like the fx5900 losing to the 9800 pro all over again. I guess the nv40 is gonna perform a bit ahead of a 9800xt now in games that cant make 8x2 act like 16x1. doom3 may be the few game that both cards will be equal

zakelwe
03-24-04, 03:19 AM
my god, poor nvidia. atis gonna have *16* real pipes either on their r423 or r450. Itll be like the fx5900 losing to the 9800 pro all over again. I guess the nv40 is gonna perform a bit ahead of a 9800xt now in games that cant make 8x2 act like 16x1. doom3 may be the few game that both cards will be equal

GF4 you're starting to jibber hysterically, I think you have got rumour shell shock. I'm going to have to slap you out of it

:ORDER:


That's better. Stop keep thinking 16x1 is better than 8x2, theoretically it might be, but it also depends on what you are trying to display and other limitations in the system, ie memory bandwidth ?? I don't know enough about it but you just can't take the number of pipelines on who will be faster , there's too many ways of using the pipelines it seems.

As an example

I was testing my 5800 last night and it is 4x2 at 475Mhz and 1GHz RAM. Theoretical pixel / texel fill rate at 475 is 1900/3800 Mpt/s but i was only getting 1500/3300 ( texels being slightly better % of theoretical). If I reduced the GPU speed to half 475 then I got 950 MP/s which is actually the theoretical .. what was happening was that my memory was not running fast enough to provide the theoretical max fill rate. It turns out that the max theoretical fill rate is only reached up to 320Mhz or so gpu for 1Ghz memory.

Strangely Ati seems to be able to get theoretical fillrate with lot lower bandwidth, so some inefficency with nv30 architecture it seems.

To get 1900Mp/sec I would need my memory to run at 1600Mhz, or better still 800Mhz with 256 bit bus :)

I've come to the conclusion that it is impossible for us laymen to work out the actual fillrates ( not theoretical ) of nv40 on just the pipeline numbers, never mind all the other more compex issues that go towards the final fps in Dx games etc.



Regards

Andy

Demirug
03-24-04, 03:55 AM
GF4 you're starting to jibber hysterically, I think you have got rumour shell shock. I'm going to have to slap you out of it

:ORDER:


That's better. Stop keep thinking 16x1 is better than 8x2, theoretically it might be, but it also depends on what you are trying to display and other limitations in the system, ie memory bandwidth ?? I don't know enough about it but you just can't take the number of pipelines on who will be faster , there's too many ways of using the pipelines it seems.

As an example

I was testing my 5800 last night and it is 4x2 at 475Mhz and 1GHz RAM. Theoretical pixel / texel fill rate at 475 is 1900/3800 Mpt/s but i was only getting 1500/3300 ( texels being slightly better % of theoretical). If I reduced the GPU speed to half 475 then I got 950 MP/s which is actually the theoretical .. what was happening was that my memory was not running fast enough to provide the theoretical max fill rate. It turns out that the max theoretical fill rate is only reached up to 320Mhz or so gpu for 1Ghz memory.

Strangely Ati seems to be able to get theoretical fillrate with lot lower bandwidth, so some inefficency with nv30 architecture it seems.

To get 1900Mp/sec I would need my memory to run at 1600Mhz, or better still 800Mhz with 256 bit bus :)

I've come to the conclusion that it is impossible for us laymen to work out the actual fillrates ( not theoretical ) of nv40 on just the pipeline numbers, never mind all the other more compex issues that go towards the final fps in Dx games etc.



Regards

Andy

If you have the speed of the Core and the number of TMUs you can calculate the texelfillrate very easy. But as you say only the theoretical value. But you can not calculate the shaderfillrate for different shaders with this little information.

A ?x1 configuration make it easier for the driver developer to get the maximum speed out of it.

As a general rule you can say that with a ?x1 configuration there are more things that can use all the shadingpower. But a ?x2 configuration and everthing else is the same need less transistors.

If you build a shader 3 hardware ?x1 is much better because you can have shorter branch blocks.

Nv40
03-24-04, 04:07 AM
Geforce4ti4200: I guess the nv40 is gonna perform a bit ahead of a 9800xt now in games

hehe... me too .. but just a bit.. after all the Nv40 was meant to compete with the 9800xt in performance ,dont you think? :wonder:

hehe sorry couldnt resist. :)

IMHO when the Nv40 is launched , something tells me , that there will be many people surprised as much if not more as when the R300 was first released. :smoking: