NVIO specifications?

H

hkultala

Regular
What are the actual nvio specifications?

(and do they differ between generations?)

die size?
mfg process?
pad count?
width and MHz of the bus to the GPU?
what are all the busses and io links it "offers"?
 
The one with G80 boards was on 130nm bulk process, if I remember right.
It hosts all the RAMDACs, digital transmitters (that is DVI) I/Os and probably the SLi interlink.
 
I actually have related question.
If nVidia had included NVIO-features into G80 and GT200 instead of separate chips, how much would it have grown the chip in terms of transistors and/or diesize?
 
I think the difference would be minor (a few mm²?). One of the key benefits of NVIO isn't die-size reduction (I think), but yields and manufacturing price. Defect in I/O section would kill entire 500mm² GPU. SPs, TMUs, ROPs and MC can be switched of and GPU can be sold as a cheaper part, but I/O abilities are essential.

I'm curious, if the yields and price benefits of the NVIO are still present during current aggressive price war - e.g. it can be good to invest a few more $ to rescue some GPUs, which can be sold for price X, but is it still profitable, if you must sell these GPUs for price X/2 or X/3?
 
Kaotik said:
I actually have related question.
If nVidia had included NVIO-features into G80 and GT200 instead of separate chips, how much would it have grown the chip in terms of transistors and/or diesize?
Click to expand...

Not much, BUT

they already have huge number of pins because of wide memory bus, and replacing the few pins of the "nvio link" with all the connections that nvio offers would have increased the pin count quite much, and made the PCB even more expensive.

G92 got rid of separate nvio bus not because of mfg process/die size, but because of anrrower 256-bit memory bus.
 
I agree with what hkultala says, it's all about the complexity of the GPU design when interfaced with memory using an 512-bit Interface.

btw, off topic question, what's exactly the die size if the G92b ?
 
Thanks for the reply, but I am now confused !

I've read these numbers till now: 231mm², 264mm², 270mm², which is true ?

GTS 250 is based on the G92b, true ?
 

I was told, that on the GPU edges was layer of the resin, and because of that the dimensions should be slightly lower (maybe -0,25mm for each value)... about 16.4 * 16.7.
 
From our excel sheet

NVIO 1 6,2 6,1 37,82mm²
NVIO 2 5,85 6,17 36,0945mm²
(measured, son including some margin of error due to packaging)
 
I think hkultala is talking about something different. It's likely, that only a fraction of the die is used for RAMDACs and TMDS - the rest is present only to reach dimensions necessarry for I/O pads. Notice, that despite different manufacturing proces, both dice versions are about the same size.
 
You must log in or register to reply here.

Similar threads

A
Understanding game performance through profiler trace graphs *spawn
2 3 4
Replies
66
Views
3K
XboxKING
X
R
Let's discuss the ways the future console architectures could potentially improve in efficiency
2 3
Replies
52
Views
6K
Miniature Kaiju
M
T
GPU Stability testing for Nvidia cards - PowerSweep
Replies
0
Views
445
T2098
T
B
How do "lead platforms" work?
Replies
13
Views
2K
Shifty Geezer
Shifty Geezer
I
Possible rasterization pop-in/too fast frame delivery on LG OLED A2?
Replies
18
Views
2K
inlimbo
I
Back
Top