Printable View
Quote:
Originally Posted by largon
Not only that, but I think it would be necessary to maintain a 512bit external interface. Otherwise that circuit would be much too insane for anyone to do anything to.
I remember with old school cards, they required chips on both sides to use a 256bit interface, but that was taken care of by a more complicated 1d circuit. My guess is that the first 512 will also be a 2d circuit.
Will it be named the X2800XTX?:slobber: X2800GTO FTW!Quote:
Originally Posted by theteamaqua
Frankly if the INQ is correct (which they seldom are) and the 16(64) specs are correct, they are going to need 512bit RAM to compete with the G80.
Judging by the size of the chip the INQ may be correct.
oh man i hope it will be a bit more intelligent than that..........it just does sound good at allQuote:
Originally Posted by Kingcarcas
well since ATI is owned by AMD now. im pretty sure AMD will want to come up with a good name
yeah and for 16 ROP, and 64 unifiied shaders, it doesnt seem enough to compete with GeForce 8800GTX
but it does have 512bit interface, and remored 114GB/sec to 140GB/sec mem bandwidth
i wonder what it's pixel, texture fillrates are
btw, does anyone know how nvidia come up with 36.8 Giga-texels/ sec for GeForce 8800GTX 's fillrate?? cuz 32 ROP x 575MHz does equal 36.8Gt/sec
way too many numbers and abbreviations there for me
i just want to see the card go :D
nvm i got something for pixel fillrate:
128 (shaders) / 2 x 575MHz = 36.8 billion/sec
but why divide by 2 ? lol
samething with GTS:
96 (shaders) / 2 x 500MHz = 24 billion/sec
EDIT:
http://www.motherboards.org/reviews/...re/1671_6.html
Quote:
The GeForce 8800GTX has fully decoupled texturing units from the new stream processors. The 8800GTX can deliver up to 64 pixels per clock of raw texture filtering horsepower, versus the 24 on the GeForce 7900 GTX), 32 Pixels per clock of texture addressing, 32 pixels per clock of 2x anisotropic filtering, and 32-bilinear filtered pixels per clock. What this means is that bilinear anisotropic filtering is almost free on the GeForce 8 (nothing is free in 3D). FP16 bilinear filtering output is 32 pixels per clock, and FP16 2:1 AF is done at 16 pixels per clock.
so for GTX its 18.4 Giga-texels/sec and 36.8Giga-pixels/sec, which is shown here:Quote:
The texture units are clocked at a fast 575MHz. With 32 pixels per clock of bilinear filtered texels that gives a texel fill rate of 18.4 Gigatexels a second which is the standard way to derive texel fill rate scores. However, when 2:1 bilinear anisotropic filtering is applied, two bilinear texels are used, giving the card an effective 38.4 gigatexel fill rate. The GeForce 8800GTX can output 24 pixels (ROPs) in a single clock with color and Z-processing. The 24 ROPs are divided into 6 partitions of 4 (16 sub-pixel samples). There is a new mode called Z-only processing that allows 192 samples per clock. If 4X MSAA is applied then 48 Pixels Z-only processing is possible. Again, in terms of pure pixel fillrate, the 8800GTX can output 24x575MHz=13.2 Gigapixels a second theoretical, more than enough to fill a 2560x1600 screen many
http://en.wikipedia.org/wiki/Compari...Force_8_Series
Sentential,
According to the INQ the 64 shaders on R600 are not comparable to ATi's previous gen shaders nor those on G80:Quote:
Originally Posted by the INQ
wait so 4 way SIMD thats 4x64= 256 shaders!!! but it depends on its shader clock as G80's shader clocks runs at 2.4 , 2.34 times of ROP, and TMU
this reminds me of Xenos, 48 shaders but 2 ALU, so 96 total +"16 texture fetch + 32 control flow + 16 vertex fetch"(dont know what thsoe 3 are ..) and total of 48billion shader operation/sec @ 500MHz core
It is highly likely that shaders in R600 are similar to those in R500 chip (xbox360), so if anyone have details how Xenos is organized then we can speculate more accurately. Other way we should wait for a card :D .Quote:
Originally Posted by theteamaqua
Edit: I was digging a little and I found some info:
http://news.teamxbox.com/xbox/9487/A...nology-Papers/
Xenos - 232milion transistors, way less than R600 and still 48 unified shaders
On the other hand extra 32 pixel shaders in R580 eaten only 65milions transistors so it is highly likely that those in R600 are way more complex.
We can't forget that R600 is more functional than R500 so at least 100-200 milions transistors are eaten by mem controller, composite engine, more robust GPGPU functions, etc.
I fathom real world performance won't be as good as with 256 "real" simple scalar shaders due to less-optimal efficiency of the wider shaders.
Yes, but still 64(256) will be faster in 98% cases than 128 scalar pipes, especially in games where complex scalar operations are very often. ;)Quote:
Originally Posted by largon
well i think heres the situation:
G80 has more ROP, TMU
R600 has more shaders
so games like Oblivion, heavy shader games will play insanely well on R600. ie 1 R600 > G80 SLI!!
but in 3dmark06, BF 2142, BF 2 , it seems like G80 will win as it benefit a lot from more ROPs, TMUs
basically we covered that games make in the ATi way of doing things, gives ATi the edge. Games designed for nVidia, give nVidia the edge... No really :rolleyes:
We can't be sure what Ati eng. had done with R600 so it may be that we get 64 TMUs in this monster. With ROPs I'm opting for 16 thus it's enough for way over 100FPS in 2560x1920 :)
Can you explain why?Or are you just quoting what the Iquirer thinks?Quote:
Originally Posted by Lightman
Everything is about how fast you can calculate (math speed). It is almost same as A64 vs P4 clock wise :)Quote:
Originally Posted by schtuga
ATI one complex ALU is supposed to do 4 math results in 1 cycle, nVidia simple ALU is doing 1 result in 1 cycle.
Before nVidia G70 GPU had 2 alus per pixel pie, one simple and one complex. ATI R5xx on the other hand had only 1 complex ALU per pixel pipe. That's why in Shader Mark equally clocked G70 was faster than R520 in simpler shaders. R520 was superior only in highly branched shaders thanks to much better general purpose architecture (RingBus, good branch predictors, etc.).
Why don't you just say it, nVidia's design is superior for simple math
and ATi's is superior for complex math
Ok ,that makes sense.I just thought that clock speed and other hardware/features,might also factor in.256 @ 700mhz vs 128 @ 1350mhz,as an example.And seeing as through most of history,they have always traded blows,I thought it would happen again.More like an amd 60% nv 40% kind of thing. Thnx for the response.There I have maintaned my post per month count.:)
this is just like last gen,
G71 has more ROP , higher texture, pixel fillrate
while R580 has more shaders
thats why G71 performs better on 3dmark06
soem games like BF 2142 performs better on G71 while Oblivion performs better on R580
Because I'm improving my written English :comp10:Quote:
Originally Posted by nn_step
BTW It's sunday midnight so plenty of time and so little active threads on forums :dammit:
Greets and keep posting ;)
Looks hot - needs a good cooling system :)
Hence the reason why it does better for F@H than the nvidia cards?Quote:
Originally Posted by nn_step
I really want to see what ATI - uh, AMD - comes up with in the final product :cool:
Hmmm, I dunno why but I always thought this way
Multimedia (Video, high resolution/definition, etc.) = ATI FTW
Gaming = Nvidia FTW
In which I liked ATI better, since its RED :p: , but really
because it seemed like a multi-tasker in a sense that it will
be good for gaming as semi-professional video editing :)
Anyways...
WHEN THE BLOOD HLL ARE THEY GONNA RELEASE THE BLOODY
RD600 BOARDS!!! BAAAAAAAH!
"Its gonna be release in October. No scratch that. November. Um...
maybe Q4 2006? December...?"
Fudgicals! come out with it already :p:
well G80 was delayed by 2,3 months
Q2 this year they were saying how G80 will be launch on Aguest, then September, then ... November