R4XX to be in the HUNDREDS of millions of transistors

[suburbanguy]

ATI's R4XX VPUs are apparently going to at least double the R300/R350 VPUs in transistor count and be in the hundreds of millions of transistors.

well that means AT LEAST 200 million transistors, would it not?

which suggests to me that R400/R420/R450
(or just R420 and R450) will be 16 pipeline VPUs with more depth.

source
http://www.rage3d.com/#1060676768

discussion
http://www.beyond3d.com/forum/viewtopic.php?t=7376

[CaptNKILL]

SO... what your saying is, its gonna be fast?

[GlowStick]

Originally posted by suburbanguy
ATI's R4XX VPUs are apparently going to at least double the R300/R350 VPUs in transistor count and be in the hundreds of millions of transistors.

well that means AT LEAST 200 million transistors, would it not?

which suggests to me that R400/R420/R450
(or just R420 and R450) will be 16 pipeline VPUs with more depth.

source
http://www.rage3d.com/#1060676768

discussion
http://www.beyond3d.com/forum/viewtopic.php?t=7376

um duble r300 transisitors would be about 160million, while the nv35 weighs in at like 130m give or take a few million so no big deal.

[nvnews-reader]

Has anyone heard how many the nv40 is supposed to have?

[GlowStick]

Originally posted by nvnews-reader
Has anyone heard how many the nv40 is supposed to have? The info hasent leaked yet, but people sure speculate, may i recomend browsing http://www.notforidiots.com/GPURW.php

[StealthHawk]

Originally posted by GlowStick
um duble r300 transisitors would be about 160million, while the nv35 weighs in at like 130m give or take a few million so no big deal.

Um, R300 has over 100 million transistors. Double R300 would be over 200 million.

[GlowStick]

Originally posted by StealthHawk
Um, R300 has over 100 million transistors. Double R300 would be over 200 million. Hm crap, i was thinking of the rv350 whops.

[Nebuchadnezzar]

220 to be the exact double :)

[suburbanguy]

um duble r300 transisitors would be about 160million, while the nv35 weighs in at like 130m give or take a few million so no big deal.


no, R300 is 107 or 110 million transistors, therefore double that would be 214 or 220 million, so I'd expect R420 to be about that plus or minus a few tens of millions.


Nv35 is indeed 130m, and Nv40 should be 250~300m

[Morrow]

Originally posted by suburbanguy
no, R300 is 107 or 110 million transistors, therefore double that would be 214 or 220 million, so I'd expect R420 to be about that plus or minus a few tens of millions.


Nv35 is indeed 130m, and Nv40 should be 250~300m

nv35 will never be 250-300m since it is based on 0.13 micron and 300m trannies won't be possible before 0.09 based graphiccards ship.

So most likely the nv40 will have 150-200m transistors...

[Geforce4ti4200]

Originally posted by CaptNKILL
SO... what your saying is, its gonna be fast?


if it has 16 pipelines......err subprocessing units whatever you wanna call it. so itll be 16x2 after all and not 12x2 like rumored? wow then yes it should and probably could be double a 9700 pro, maybe even double a 9800 pro depending on clock speeds. 500/500 sounds reasonable though

[Morrow]

Originally posted by Geforce4ti4200
if it has 16 pipelines......err subprocessing units whatever you wanna call it. so itll be 16x2 after all and not 12x2 like rumored? wow then yes it should and probably could be double a 9700 pro, maybe even double a 9800 pro depending on clock speeds. 500/500 sounds reasonable though

The current number is still 12x1, not 12x2 or 16x2. Bandwidth currently is not enough to feed anything like 12x2 or above...

[Geforce4ti4200]

then Nvidia's 8x2 will outperform ati's 12x1 :mad:

[Nebuchadnezzar]

Originally posted by Geforce4ti4200
then Nvidia's 8x2 will outperform ati's 12x1 :mad:

Don't think so..

[-=DVS=-]

Originally posted by Geforce4ti4200
then Nvidia's 8x2 will outperform ati's 12x1 :mad:

Nvidia's 8x2 Might push more texels per cloack but not pixels like ati's 12x1

Core 500
8x2 = PixelFilrate 4000 TexelFillrate 8000

12x1 = PixelFilrate 6000 TexelFillrate 6000

Well i personally would take 12x1 more raw power in fillrate :p

Current 9800Pro lets say have 400 Core that gives it 3200 Fillrate , so new card from ATI will have almoust twice the power :drooling:

[Uttar]

We should worry about GFLops, not fillrate, IMO.


Uttar

[-=DVS=-]

Originally posted by Uttar
We should worry about GFLops, not fillrate, IMO.


Uttar

Only problem noone is telling us spec how much gigaflops these cards can calculate :(

[Demirug]

Originally posted by Uttar
We should worry about GFLops, not fillrate, IMO.


Uttar

as always it depends. Most games today and in the near future will more limited by fillrate than by shaderpower at least if played with triAF. This will change if time goes by.

But even in this case GFlops do not tell the whole truth. You need 4 multiplications and one addition of the 4 result to calculate a vector4 dot product. That are 5 flops. But I can say that my addition works in an other way. It first calculate the sum of the first two multiplications. After this the sum of the other two multiplications ist calculated and this all ends with the sum of the two sums. Now I have 7 flops. GFlops means nothing if you do not know how the chip use the power it have.

I strong believe that MPSI/s (PSI = Pixel Shader Instructions) is a good way to go. Because todays chips use different FPUs for different jobs (R300), use the same FPU for all jobs but have different executions rates for diffrent kind of jobs (NV30) or mixed this all up (NV35) we need additional informations:

MPSI(tex)/s for textureinstrutions. Instructions not samples
MPSI(vec3)/s for vector3 operations
MPSI(vec4)/s for vector4 operations
MPSI(scalar)/s for scalar operations

and an large list (if necessary) with the conditions that make it possible or impossible to use all the available shaderpower.

But I have a feeling that still this will not work with complex FPU designs like nVidia use. Simply too much crossbars (mux) and small functions units to really tell how much useable work is done per clock.

[lukar]

For long time I'v got full spec for loci, but nobody believes me. The spec of loci will be

0.13 500Mhz design 1TMU 16Pipes, 256bit mem, 128/256 DDR II 500Mhz
~180millions trans, full support for dx 9.1 and open gl 2.0
Smartshader 3.0, AA 8x mode will be introduced
Filtering remains the same 2x 4x 8x 16x performance/quality
Trueform 3. Now fully suppot by hardware and also support for displacement mapping found in Matrox cards.
Videshader II
HyperZIV compression
The card will not need additional power, It will support AGP 4x, 8x. Also the card will support PCI expres, but those cards will not be shipped yet. There will be no adapter for agp 8x/ PCI express.

[GlowStick]

Originally posted by lukar
For long time I'v got full spec for loci, but nobody believes me. The spec of loci will be

0.13 500Mhz design 1TMU 16Pipes, 256bit mem, 128/256 DDR II 500Mhz
~180millions trans, full support for dx 9.1 and open gl 2.0
Smartshader 3.0, AA 8x mode will be introduced
Filtering remains the same 2x 4x 8x 16x performance/quality
Trueform 3. Now fully suppot by hardware and also support for displacement mapping found in Matrox cards.
Videshader II
HyperZIV compression
The card will not need additional power, It will support AGP 4x, 8x. Also the card will support PCI expres, but those cards will not be shipped yet. There will be no adapter for agp 8x/ PCI express. Hm sounds nice, and when will it be released?

[Geforce4ti4200]

I believe you but got questions:


"0.13 500Mhz design 1TMU 16Pipes, 256bit mem, 128/256 DDR II 500Mhz"


500/500 clock speeds, right? Thats not too bad, any chance the core could be 550MHz though? also 16x1? I thought it would be 12x1 cause 16x1 takes too many transistors


"The card will not need additional power"


theres no way the agp slot can supply the power. even the 9500 pro has a floppy header to get additional power from the psu, if you dont give it more juice, I heard itll either be underclocked or crash.



"Also the card will support PCI expres, but those cards will not be shipped yet."


seeing pci express mobos would ship around the same time, not many people would get a new mobo for that. I heard the nv40 and r420 will be the last cards to be agp, the rest of them now on will be only pci express, however I heard pci express mobos will still have an agp slot for older cards, kinda like isa was still around when agp came

[Paul]

Those specs aren't entirely correct.

[StealthHawk]

Originally posted by Paul
Those specs aren't entirely correct.

No surprise there :p

[Nebuchadnezzar]

Originally posted by lukar
For long time I'v got full spec for loci, but nobody believes me.

And that's for a good reason. :rolleyes: :rolleyes:

[ClyssaN]

loki