What will be Nvidia's first 28nm part?

I'd bet on a 28nm Notebook shrink of Fermi (eventually being released on the desktop). As I said in another thread, a GF116 with a 128-bit GDDR5 bus makes sense but we'll see what happens.
 
I'd bet on a 28nm Notebook shrink of Fermi (eventually being released on the desktop). As I said in another thread, a GF116 with a 128-bit GDDR5 bus makes sense but we'll see what happens.
I've considered this "GF117" the most likely chip if NVIDIA goes 28 nm Fermi before Kepler. I'm guessing it will replace the GT 545's and the GT 55xM's (and maybe more).

But if it's 28 nm, would it still be a GF11x, or would it be a GF12x?
 
NVIDIA GeForce 600 Series: 28nm, OEM-only Fermi Refresh?

From BSN

by: Theo Valich

According to our sources, the company will repeat the strategy from previous architectures (G80-G9x, GT200-GT21X) and introduce a limited number of die-shrunk Fermi-based parts which are to be used for process optimization.

As such, the low-numbered GeForce 600 parts will target OEM and SI vendors, but we expect some parts to appear in e-tail/retail as well, as the box-business (Add-In-Board Vendors) companies usually get their share of parts which are worth seeing in discrete end of business.

http://www.brightsideofnews.com/new...600-series-28nm2c-oem-only-fermi-refresh.aspx
 
Interesting. The wording makes it seem like there will be more than 1 Fermi chip on 28 nm.

Let me guess:

GF114 (384, 256) » GF125? (384, 192?)
GF116 (192, 192) » GF127? (192, 128?)
GF108 (96, 128) » GF129? (96, 64?)

I doubt there will be 3 chips though (more like 1-2) unless Kepler won't be here for quite a while. In the 3 chips case, I wonder if someone will make a dual "GF125." That would easily fit under 300 W, should beat the 580, and depending on clocks (the 300's all have modest clocks) could approach the 590. (Although I think high-end Kepler would give similar if not better performance once it shows up.)

Maybe high-end Kepler would succeed GF110 around Q2 and then trickle down to the **4 and below a few months later, giving time for the GF12*'s.
 
A GTX560Ti -> GTX650 with only 120W TDP would be pretty nice for me. (And price slightly below 200$)

A simple shrink of GF114 would probably be too small for a 256-bit bus.

Given that the price of GDDR5 is finally coming down, I'd expect that the sanest chip they can make is something with (roughly) the flops horsepower of a GTX460 with a 128 bit GDDR5 bus. That would be cheap, and sell like hotcakes to the "high end" gaming OEM crowd. Also it would probably rock in a laptop.
 
Yeah but it also looks like it's going to be at least 5 months late compared to its Southern Islands counterpart, so we'll see. But on paper it does look quite good.

1. How reliable is this chart?
2. How reliable are SI rumors?
 
The bit that doesn't make sense is the sea change in perceptions about AMD. Previously they were the pragmatic, sensible guys who tested the waters before diving in. Now they're plunging head first into 28nm with their biggest architecture change in 5 years? Where's the 28nm pipe cleaner?
 
You can't do a pipe cleaner, if the pipes aren't ready yet. IF TSMC doesn't deliver, AMD will slip with their launch date too.
 

how far along 32nm did they get? maybe they feel comfortable with a "die shrink" with what they learn't from 32nm? maybe they knew how bad 40nm was looking @ TSMC and they feel 28nm is looking a lot better etc.

i guess we will just have to wait and see what we see
 
Yeah but it also looks like it's going to be at least 5 months late compared to its Southern Islands counterpart, so we'll see. But on paper it does look quite good.

1. How reliable is this chart?
2. How reliable are SI rumors?

Also both of you are forgetting the fact that this supposed chart is talking about Nvidia's mobile parts, and you're comparing this with AMD's desktop parts. The mobile parts typically lag desktop by a few months or more so take that into account as well.

The bit that doesn't make sense is the sea change in perceptions about AMD. Previously they were the pragmatic, sensible guys who tested the waters before diving in. Now they're plunging head first into 28nm with their biggest architecture change in 5 years? Where's the 28nm pipe cleaner?

Some of the rumours say that not all the SI parts will be based on GCN. One or more of the low end cards are possibly still based on VLIW4, and that could very well be the pipecleaner.
 
Back
Top