G70 & G80 Specs to come! Help need more Info...

[ToxicTaZ]

So now we know that NV-## code is all moved around like so:

NV-40 = 130nm 6800 AGP
NV-45 = Is 130nm NV-40 with PCI-E and slower clocks
NV-47 = Is now the G70 a Spring Refresh @Booth 346 E3 http://www.nvidia.com/object/IO_18245.html
NV-48 = Is 130nm NV-40 & NV-45 now both with 512MB Ram.
NV-50 = Is now the G80 GeForce 7 (7800) ? Some time this fall we all hope? :rolleyes:

Now that we all know the code and the spring refresh is now at E3 http://www.e3expo.com/e3expo/ Nvidia Booth 346 http://www.nvidia.com/object/IO_18245.html

G70 We expect it to be based on a high end GeForce 6 tech using 110 nanometre and have more pipes than NV-40. :cool:

- GeForce 6900 Ultra AGP, GeForce 6900 GT PCI-E, (G70)
- 256-bit GPU core
- 0.11-micron process made @ TSMC
- 24 pixel pipelines
- 500MHz core speed
- Under clocked RAM @ 750MHz from 256-bit 800MHz GDDR 3 Ram (SAMSUNG) K4J55323QF
- 512MB of on board Video Ram
- 8 Hardware Vertex Shaders
- 260 million Transistors
- DirectX 9c PS/VS 3.0 (shader models 3) and OpenGL 1.5 Support
- Dual 500MHz RAMDAC
- Dual DVI for HDTV in & out
- SLI
- a working hardware video decoder this time. MPEG-2
- All other GeForce 6 specs are the same.

There all so mite be a cut down verson of this card like a 6700 to come still too? :confused:

And we wait to see the real specs of the G80 GeForce 7 to come? :rolleyes:

- GeForce 7800 (G80)
- 0.09-micron process made @ TSMC or IBM
- 512-bit GPU core
- 32 pixel pipelines (maybe a Dual Core 2x16 pixel piplines GPU)
- 600MHz core speed
- 256-bit 1000MHz GDDR3 1ns (SAMSUNG) 64GBps, Total of 2000MHz of video RAM or (SAMSUNG) 512-bit 500MHz GDDR3 2ns
- 512MB and 256MB Models of Video Ram
- 12 Hardware Vertex Shaders
- 325 million Transistors
- UltraShadow III Technology
- Intellisample 4.0 HCT Technology
- CineFX 4.0 Engine & WGF DirectX 10a PS/VS 4.0 (shader models 4) and OpenGL 2.0 Support
- Dual 600MHz RAMDACs display resolutions up to 2048x1536 @120Hz
- Dual DVI for HDTV in & out
- DVD and HDTV-ready MPEG-4 decoding up to 1920x1080 ( 1080p/i ) resolutions.
- PCI Express and AGP 8x Support
- SLI
And as for XDR™ DRAM (Rambus) some could say it may be used this year? not very fast this year...
http://www.samsung.com/Products/Semiconductor/News/DRAM/DRAM_20050125_0000098594.htm

If we put all are heads together from all the threads we read in the last 6 months theres a lot of info put in to one thread! HARD CORE HARDWARE SPECULATING SPECS!! :angel:

[retsam]

well this is a good start .... i do think there is more to come in the g70 core then what you are saying there ... do i hear sound storm ... or even maybe a physics engine

[MUYA]

G70 is almost spot on with what I hear, only the PS part is much beefier than the nv40 to nv45 range.

Nothing on G80

[Mr_LoL]

Whats the obession with dual core parts these days?

[MUYA]

For CPUs, dual core is somewhat hazy almost relying on optimized software.

On a GPU it's quite different. It's probably suggesting two new (ie different atchitecture/tweaked architecture) 4 quad block GPU being slapped together as one. All NV4X from NV40 to NV45 are modular, ie NV can chop or lob on a quad block to suit their need. Like the nv40/nv45 has 4 quad block, NV41 has 3 quad blocks, nv43...has 2 etc. So essentially you are slapping 2 4 quads to make it 8 quad GPU :) I think the emphasis is on modularity! Of course the G80 and G70 architectures are different from nv40-NV45...G80 even more so than G70.

[Mr_LoL]

For CPUs, dual core is somewhat hazy almost relying on optimized software.

On a GPU it's quite different. It's probably suggesting two new (ie different atchitecture/tweaked architecture) 4 quad block GPU being slapped together as one. All NV4X from NV40 to NV45 are modular, ie NV can chop or lob on a quad block to suit their need. Like the nv40/nv45 has 4 quad block, NV41 has 3 quad blocks, nv43...has 2 etc. So essentially you are slapping 2 4 quads to make it 8 quad GPU :) I think the emphasis is on modularity! Of course the G80 and G70 architectures are different from nv40-NV45...G80 even more so than G70.

Thanks for that :confused: :D

[halduemilauno]

So now we know that NV-## code is all moved around like so:

NV-40 = 130nm 6800 AGP
NV-45 = Is 130nm NV-40 with PCI-E and slower clocks
NV-47 = Is now the G70 a Spring Refresh @Booth 346 E3 http://www.nvidia.com/object/IO_18245.html
NV-48 = Is 130nm NV-40 & NV-45 now both with 512MB Ram.
NV-50 = Is now the G80 GeForce 7 (7800) ? Some time this fall we all hope? :rolleyes:

Now that we all know the code and the spring refresh is now at E3 http://www.e3expo.com/e3expo/ Nvidia Booth 346 http://www.nvidia.com/object/IO_18245.html

G70 We expect it to be based on a high end GeForce 6 tech using 110 nanometre and have more pipes than NV-40. :cool:

- GeForce 6900 Ultra AGP, GeForce 6900 GT PCI-E, (G70)
- 256-bit GPU core
- 0.11-micron process made @ TSMC
- 24 pixel pipelines
- 500MHz core speed
- Under clocked RAM @ 750MHz from 256-bit 800MHz GDDR 3 Ram (SAMSUNG) K4J55323QF
- 512MB of on board Video Ram
- 8 Hardware Vertex Shaders
- 260 million Transistors
- DirectX 9c PS/VS 3.0 (shader models 3) and OpenGL 1.5 Support
- Dual 500MHz RAMDAC
- Dual DVI for HDTV in & out
- SLI
- a working hardware video decoder this time. MPEG-2
- All other GeForce 6 specs are the same.

There all so mite be a cut down verson of this card like a 6700 to come still too? :confused:

And we wait to see the real specs of the G80 GeForce 7 to come? :rolleyes:

- GeForce 7800 (G80)
- 0.09-micron process made @ TSMC or IBM
- 512-bit GPU core
- 32 pixel pipelines (maybe a Dual Core 2x16 pixel piplines GPU)
- 600MHz core speed
- 256-bit 1000MHz GDDR3 1ns (SAMSUNG) 64GBps, Total of 2000MHz of video RAM or (SAMSUNG) 512-bit 500MHz GDDR3 2ns
- 512MB and 256MB Models of Video Ram
- 12 Hardware Vertex Shaders
- 325 million Transistors
- UltraShadow III Technology
- Intellisample 4.0 HCT Technology
- CineFX 4.0 Engine & WGF DirectX 10a PS/VS 4.0 (shader models 4) and OpenGL 2.0 Support
- Dual 600MHz RAMDACs display resolutions up to 2048x1536 @120Hz
- Dual DVI for HDTV in & out
- DVD and HDTV-ready MPEG-4 decoding up to 1920x1080 ( 1080p/i ) resolutions.
- PCI Express and AGP 8x Support
- SLI
And as for XDR™ DRAM (Rambus) some could say it may be used this year? not very fast this year...
http://www.samsung.com/Products/Semiconductor/News/DRAM/DRAM_20050125_0000098594.htm

If we put all are heads together from all the threads we read in the last 6 months theres a lot of info put in to one thread! HARD CORE HARDWARE SPECULATING SPECS!! :angel:

no.
G70 7800 ultra.
G75 7900 ultra.
G80 8800 ultra.
IMHO
(sofahide)

[JoKeRr]

man, the G70 clock, 500mhz, is that the GT clock or Ultra clock?? 24 pipe monsters, that's huggggee

[ToxicTaZ]

man, the G70 clock, 500mhz, is that the GT clock or Ultra clock?? 24 pipe monsters, that's huggggee


What? whats wrong with these specs? :(

If you don't think these are close so put your idea up plz :o

[JoKeRr]

What? whats wrong with these specs? :(

If you don't think these are close so put your idea up plz :o

wowowow, don't flame me man, I'm just asking a question, and saying it'll be great to have 24 pipe card clocked at 500. :angel:

[InqWoN1776]

well, joker does have a point with the clock speed. It doesn't seem like nV should be able to hit that kind of clock on that process but who knows. They did get the 6600gt up to 500 core and it overclocks very well. However, ATI's x800xl is at 400 core and it doesn't overclock ver well but has 16 pipes instead of 8. So, we really don't know what nV can do with this process and atleast 16 pipes. I would guess nV goes with more pipes instead of core clock but maybe they do a mixture of both.

[bkswaney]

WOW... Sounds great!
But this should be the spring release for sure. "G70"
The 6800U carried them all of 04 and well into 05.
It's time for a refresh for sure. :)

[ToxicTaZ]

wowowow, don't flame me man, I'm just asking a question, and saying it'll be great to have 24 pipe card clocked at 500. :angel:

Sorry man did not mean to flame but your right :) It will be great to see if Nvidia can pull this off some thing close my put together specs :eek:

:angel:

[MUYA]

I think the 260 Million transistors maybe a lil off. Just speculating, 2 extra quad blocks with tweaked pipelines and 2 tweak VS units and maybe new optimizations for color compressions and new AA unit...hmm I dunno.. may cross 300 million :)

Just a wild stab

[MUYA]

BTW I really dunno if it's g70 or g80.... the codenames ...could be nv47, nv48 and nv50 really.

[wEEt]

I guess you expect too much from the new Nvidia Card: The clock speeds are too high. If we see 24 pipes, which I think is possible, we won't have the core clocked @ 500MHz. I hope, it's clocked @ 400MHz, which is already much.
And I think, 512MB will be possible, but we'll see 256MB in the beginning.

Your specs would make the 6900U be nearly twice as fast as a 6800U. I guess that's simply too much. It's going to be a refresh, maybe a good one, but this is not going to be wonder card.

[JoKeRr]

I guess you expect too much from the new Nvidia Card: The clock speeds are too high. If we see 24 pipes, which I think is possible, we won't have the core clocked @ 500MHz. I hope, it's clocked @ 400MHz, which is already much.
And I think, 512MB will be possible, but we'll see 256MB in the beginning.

Your specs would make the 6900U be nearly twice as fast as a 6800U. I guess that's simply too much. It's going to be a refresh, maybe a good one, but this is not going to be wonder card.

I think 500 mhz is actually quite possible. Let's do some math: the shrink from 130nm to 110nm rough reduces the die size by 30%, increase the transistor count in increase 2 addition quad to 300million is rough an increase of 36%, so the final die size is roughly the same (and a little big less, about 95%) as NV40.

Since it's made with 110nm process, which means it will switch faster, add in some revision tweaks and a big heatsink, boom-->you should be able to have it at 500mhz (if NV wants to). Yield at the 110nm process for NV has been good (think 6600s and 6200s and nv41). So I don't see why they couldn't get it to 500mhz core.

Again think about the example of NV35/38 Vs. NV41M, 130nm Vs. 110nm, I guess for the die size, NV41M is prob. still a bit bigger, but b/c of the shrunk in transistor size, even for a mobile part, it's clocked at 450mhz and only making just over 40W. suggestions?? :D

[retsam]

you forgot something... the smaller you go the more leakage there tends to be .. so we need to factor that in also .. i think 500 mhz is gonna be hard for nvidia to hit i say its gonna be around 450 with highly efficiant pipes.

[MUYA]

Well to counterpoint that, NV have had more time to get those things sorted out...ie it's been almost 1 year since the nv40 was launched. They have had plenty of time to tweak the design to prevent as much leaks and get thermal profile right. Remember ATi are oing 90nm with 300 million plus GPU, that will also require heavy design tweaking to get teh thermal.electrical aspect just right.

:)

Maybe u all knew this anyway but my 2 cents... :D

[wEEt]

But I think, ATi is going to be the winner in performance. They wait until Nvidia shows their new card and can adjust the clock speeds of the R520 so that it's fast enough to beat Nvidias Refresh Chip.
That already happend...

[MUYA]

Depends on how much they have tweaked their pixel pipelines. It's all hearsay ...we will have to wait till we get in depth technological previews and benches etc

[MUYA]

Hmm U cannot just ease anotehr maybe 50 mhz as you seem to suggest ATi can do in a few short weeks especially on 90nm process which has been problematic as Intel, IBM and AMD to certain degree have shown.

[retsam]

ok riddle me this batman.. if ati's part is 300 million transisters. and they are adding everything that nvidia has at 160 million transisters, plus another quad. and lets say that a quad takes up 5 million transisters.. what is the other oh 150 million for ? and please dont tell me its wfg 1.0.. also is wfg 1.0 done yet ? or even close to being done.

[Subtestube]

Actually,the NV40 has 222 million transistors, not 160. The NV43 has 143 million transistors or so.

I could be wrong, but I believe that as the complexity of the chip, and in particular the shader cores, gets higher, those quads/units actually require more transistors. In other words - it's not just 5M transistors per quad - it'll actually be much more for chips of this complexity. As I say - I don't actually know - just surmising as it were. In any case, MUYA is (apparently) correct - the ATi part (the R520), is rumoured to have between 300 and 350 mill transistors, but that's not that inconcievable. If you consider that the NV40 has twice the shading capacity of the NV43, let's look at the difference in transistor count:

222 - 143 = 79. Let's assume, though I'm absolutely CERTAIN this is an oversimplification and is probably wrong, that for the same addition we'll get the same improvement in shader core numbers. So, we add another 2 quads, and 2 vertex shader cores. Well, if it were 2 and 3, rather than 2 and 2, that would be exactly the same as the increase from NV43 -> NV40. So, again oversimplifying, to get the NV40 to 24pp 8vp, we'll need to add approx 75 million transistors. That puts the G70 at *just* under 300 million transistors. Does that make sense?

[MUYA]

ok riddle me this batman..

.../begin Taxi Mode
...U tahkin tuh me?

/end Taxi mode

:p