How Many Fragment Pipelines Does G70 Really Have?

Won't be of any use since the card is already starved for bandwidth.

 

I agree. I think Nvidia may have been giving a politicians answer. What they may be saying is "that we have not announced an Ultra" without saying so. They are simply saying "these are the announces models and the GTX is the top of model." Speaking of which, have there been any statements or speculations about the specifications of the 7800 GT and 7800? I would really like to know what the difference is between the GT and GTX.

 

Wouldn't they need the BIOS for the full pipelined version of the chip first to do the trick?

 

Probably. When we were messing about with X800 pros, enabling the pipes didn't help (except for the upgraded clockspeeds). You needed a BIOS that was aware of and could make use of the extra pipes (scheduling, etc).

 

They aren't exactly faulty as the parts are working as advertised.

I have to admit though, this could cheese off a lot of people who buy 7800 GTXs because they want 'the best' a few months down the line.

 

Depends on the few months. If it's more than 3-4 i don't think people should get mad at it. Since that's almost refresh time in the old gens. But i think it's quite unlikely that nv will announce something faster in 2-3 months. Would halt the sales of their current new cards :/

 

As with most home electronics - the product you buy today is already out of date when you bring it home.

There is nothing wrong with what Nvidia is doing - "IF" they are doing I should clear up. Most chip makers do this already, no? Intel and AMD - Celerons and Durons. Memory speeds, etc.

Are we looking at faults now because Nvidia released something that performs well for a second time? Is there no trust? Just wondering if we are looking too much into this and just want to Crucify ...

 

the die size of nv40 is 303mm^2, not 287 mm2 , according a paper that nvidia shown on HOT CHIPS 16. And, AFAIK,the gate density of IBM is higher than TSMC.

 

Despite starvation, it's still twice as fast as a 6800U in some benchmarks.

 

Yeah I know. What I'm saying is that unlocking the extra pipes probably won't add much to that 2x increase since the 24 pipes are probably choking at that point.

 

Could a BIOS hax0rs do similar to try to 'enable' extra pipes on a 7800, such that we might tell if any were present or not? Not necessarily to make them work properly.

ERK

 

ATI should have by now X-Rayed the thing and gotten a good look at the core and knows if its really a 32-pipe or not. They wouldn't of course come out and tell the world that. They would try their best, if its even possible at this stage, to make sure that they have something that can compete with a 32-pipe Ultra. At the same time they need to make sure that whatever they come out with when they launch R520 is just a bit faster than 7800 GTX. So now they are probably finalizing pipelines and core speeds, assuming THEIR core isn't having yield issues at TSMC.

 

They wouldn't want to disable pipes to fill a "lower-end" if the suggested yield values are correct, they'd be saving those for the Ultra/7900/whatever as the article mentions. Despite the whole "availability at launch" deal. In the coming months if yields improve and/or if the R520 (and thus the Ultra/7900/whatever) doesn't come soon-ish, they may start doing that though.

 

Dave, what you are implying is that nvidia not only

a) released their product before r520 is released so they do not know what they are up against

but

b) released it at mild clocks rather than the full clocks of say 450-480Mhz

and

c) released it with 24 internal shader "pipes" instead of 32 to flood the market and still have something in reserve.

THATS MORE CUNNING THAN A FOX WHO HAS GOT A DEGREE IN CUNNING AND MACHIVALIAN ARTS FROM OXFORD UNIVERSITY !

To badly quote Blackadder ( which was nice ).

 

I saw a review somewhere that stated even at the bigger size, nV was getting as many G70 dies on a wafer as they did NV40, which would argue against this.

The stated size of 300m, as someone said upstream, would also --but then it could be absolutely accurate. . .for a 24 ps part.

Could they have a quad of ROPs turned off too?

How about our old friends who do tweaker programs that turn these things on --wouldnt it be evident to them? Somewhere in the chip that info does reside, right?

Edit: Btw, I've always believed this was ATI's plan for R420 last time. . .until they saw NV40.

 

1: Nvidia G70_core is 0.11m which is impossible to have 32pipelines, you have to have 0.09m in order to activate 8 more pipes + another vertex shader.

2: Based on 0.11m you could push a core maybe around ~520MHz.

3: ATI is at least 3 months ahead working on 0.09m tech then Nvidia.

4: (Based on 0.09m+low-k tech) ATI could push a core around ~600MHz easily.

5: ATI R520 based on 0.09m which has 32 pipelines + 10 vertex shaders could easily outperform G70_GTX then G70_Ultra version which also has 24pipe 8vertex running core around ~500MHz.

6: At first ATI may only introduce 24pipe 8 vertex plus running core ~500MHz. But if ATI delay their 32pipe version core, then Nvidia may not be behind after all.

 

[/quote]
ATI is at least 3 months ahead working on 0.09m tech then Nvidia.
[/quote]

your forgetting ps3, sonys 90nm.

Anyone notice anything about device id's? It may mean nothing but then again it may mean something.

From Forceware 77.72

NVIDIA_NV35.DEV_0330.1 = "NVIDIA GeForce FX 5900 Ultra"
NVIDIA_NV35.DEV_0331.1 = "NVIDIA GeForce FX 5900"
NVIDIA_NV35.DEV_0332.1 = "NVIDIA GeForce FX 5900XT"
NVIDIA_NV35.DEV_0333.1 = "NVIDIA GeForce FX 5950 Ultra"
NVIDIA_NV35.DEV_0334.1 = "NVIDIA GeForce FX 5900ZT"

NVIDIA_NV43.DEV_0140.1 = "NVIDIA GeForce 6600 GT"
NVIDIA_NV43.DEV_0141.1 = "NVIDIA GeForce 6600"
NVIDIA_NV43.DEV_0142.1 = "NVIDIA GeForce 6600 LE"
NVIDIA_NV43.DEV_0145.1 = "NVIDIA GeForce 6610 XL"

NVIDIA_NV40.DEV_0040.1 = "NVIDIA GeForce 6800 Ultra"
NVIDIA_NV40.DEV_0041.1 = "NVIDIA GeForce 6800"
NVIDIA_NV40.DEV_0042.1 = "NVIDIA GeForce 6800 LE"
NVIDIA_NV40.DEV_0045.1 = "NVIDIA GeForce 6800 GT"

NVIDIA_G70.DEV_0091.1 = "NVIDIA GeForce 7800 GTX"


All top of the line cards end with 0.1, 7800gtx is 1.1.

From Leaked Encrypted Forceware 75.90

NVIDIA_G70.DEV_0090.1 = "NVIDIA G70"
NVIDIA_G70.DEV_0091.1 = "NVIDIA G70"
NVIDIA_G70.DEV_0092.1 = "NVIDIA G70"
NVIDIA_G70.DEV_0093.1 = "NVIDIA G70"
NVIDIA_G70.DEV_0094.1 = "NVIDIA G70"
NVIDIA_G70.DEV_009D.1 = "NVIDIA G70G"
NVIDIA_G70.DEV_009E.1 = "NVIDIA G70GL"

 

It'll be quite a long time until ATI has good enough yields for a 32 pipe R520.

 

ATI is at least 3 months ahead working on 0.09m tech then Nvidia.
[/quote]

your forgetting ps3, sonys 90nm.

[/quote]

Yes. Which is why ps3 will be availaible sometime in 2006 and xbox2(360) this year or before ps3.

Since Nvidia mistake with NV30 chip gave ATI little_bit ahead in time frame.

 

well if your bining the 32pipe for a real launch this fall, it would take that long to get a good supply. As for the OT on 90nmLowk, Ati has been on it for over ayear... and is most likly the issue with the RXS/PS3 being a 2006 part.
Then again maybe ATI is doing the same thing.. but with the new node its more of a issue untill fall when they have enough of the good chips for a real launch I speculate with the extra fabs at 90nm,TSMC are trying a few differnt chips...16,24 32....