• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Instead of taking sides in Green vs Red cards

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Idontcare said:
Consider the improvements AMD was able to make between the time they debuted the Deneb-based Phenom II X4 940 (Jan 8, 2009: 3GHz, 4 cores, 125W TDP, $275) versus the Thuban-based Phenom II X6 1090T (Apr 27, 2010: 3.2GHz, 6 cores, 125W TDP, $295).

That's 50% more cores, and at higher clockspeed (plus that 3.6GHz turbo-clock for 3 cores), all within the same power-envelope and nearly the same price-point.

You can do a lot when you have more time to become all the more acquainted with the process tech.
Click to expand...


Sure thing, I understand that much. 🙂 I have heard of a B stepping Fermi in the works, but it hasn't seem to materialized yet that I know of. I'm sure Nvidia can do a lot with what they have, I just can't see a bigger, potentially more power hungry part coming from them on 40nm in the form of Fermi 2. But with Nvidia only releasing harvested Fermi parts, I imagine that says things still aren't going so great for them on 40nm?

As of now I could see a full enabled GF100 that is probably not too exciting to us or a fully enabled GF104 that does get a lot of attention. I'd bet that a fully enabled GF104 that can give a 5870 a run for it's money and is priced $100 less would sell quite well.
 
Last edited:
RubyBG.jpg
 
SlowSpyder said:
Sure thing, I understand that much. 🙂 I have heard of a B stepping Fermi in the works, but it hasn't seem to materialized yet that I know of. I'm sure Nvidia can do a lot with what they have, I just can't see a bigger, potentially more power hungry part coming from them on 40nm in the form of Fermi 2. But with Nvidia only releasing harvested Fermi parts, I imagine that says things still aren't going so great for them on 40nm?

As of now I could see a full enabled GF100 that is probably not too exciting to us or a fully enabled GF104 that does get a lot of attention. I'd bet that a fully enabled GF104 that can give a 5870 a run for it's money and is priced $100 less would sell quite well.
Click to expand...

Definitely there's two rather plausible outcomes here. There's the Fermi B-Stepping respin, along the lines of what Intel did with their Q6600 B3 versus G0 stepping. And then there's the Fermi architecture refresh, along the performance/watt improvements Intel did between Penryn and Nehalem, both on the same 45nm process.

I agree whatever more Nvidia does on TSMC's 40nm the resultant product can't really consume more power nor can the die itself be much larger as they are almost at the physical limits of the shotmap for modern litho gear.

So definitely we aren't going to see a chip that has 50% more performance because it is 50% larger or because it is clocked higher while consuming 50% more power. But we could see a chip that has 30% more performance because the layout has been all the further optimized and the xtor density gets a boost along with a bit of a clockspeed bump as the power-efficiency gets a boost as well.

Same for AMD and Evergreen B-stepping and SI. The opportunity is there for both NV and AMD, only time will tell which manages to carpe the heck out of the diem though.
 
Idontcare said:
Definitely there's two rather plausible outcomes here. There's the Fermi B-Stepping respin, along the lines of what Intel did with their Q6600 B3 versus G0 stepping. And then there's the Fermi architecture refresh, along the performance/watt improvements Intel did between Penryn and Nehalem, both on the same 45nm process.

I agree whatever more Nvidia does on TSMC's 40nm the resultant product can't really consume more power nor can the die itself be much larger as they are almost at the physical limits of the shotmap for modern litho gear.

So definitely we aren't going to see a chip that has 50% more performance because it is 50% larger or because it is clocked higher while consuming 50% more power. But we could see a chip that has 30% more performance because the layout has been all the further optimized and the xtor density gets a boost along with a bit of a clockspeed bump as the power-efficiency gets a boost as well.

Same for AMD and Evergreen B-stepping and SI. The opportunity is there for both NV and AMD, only time will tell which manages to carpe the heck out of the diem though.
Click to expand...

You are probably the smartest, most neutral person in these forums. And Nvidia pretty much did exatly what you are describing when they moved from NV30 to NV35.
 
Idontcare said:
The difference in the balance of the equation is that the SI engineers have had an extra year to implement and optimize their architecture over that which the Evergreen designers had.

The Evergreen team had much much less time to get acquainted with the nuances of TSMC's 40nm process tech in practice versus the models whereas the SI team had access to the same understanding of the process nuances at a much earlier phase of the SI design timeline.

It's very much the reasoning why cpu steppings can dramatically improve power-consumption. Experience with the fab reality can feed into the design cycle (not much, but still enough to make a difference).

Nothing trumps the 2x xtor budget that comes with a node shrink, but designers can do a lot of optimizations and make better use of their existing xtor budget once a node matures and a bevy of unknowns have since become knowns. (and of course this works for everyone, Nvidia included, so the advantage for SI over Evergreen can be expected for Fermi+ as well)
Click to expand...

Sure but how much? 15% 20% at best? This is a minor refresh. NI is the one that will be handling the beatings and ruining toy drums.

EDIT: I see you said 30%, is that really possible with layout and process optimizations?
 
Last edited:
sxr7171 said:
EDIT: I see you said 30%, is that really possible with layout and process optimizations?
Click to expand...

Definitely possible but you don't have to take my word for it, take your pick of recent real-world examples of what AMD and Intel have done with performance and power-consumption within the same node over time:

Idontcare said:
Consider the improvements AMD was able to make between the time they debuted the Deneb-based Phenom II X4 940 (Jan 8, 2009: 3GHz, 4 cores, 125W TDP, $275) versus the Thuban-based Phenom II X6 1090T (Apr 27, 2010: 3.2GHz, 6 cores, 125W TDP, $295).

That's 50% more cores, and at higher clockspeed (plus that 3.6GHz turbo-clock for 3 cores), all within the same power-envelope and nearly the same price-point.
Click to expand...

Idontcare said:
There's the Fermi B-Stepping respin, along the lines of what Intel did with their Q6600 B3 versus G0 stepping.

And then there's the Fermi architecture refresh, along the performance/watt improvements Intel did between Penryn and Nehalem, both on the same 45nm process.
Click to expand...
 
Vesku said:
Why not talk about when we are going to see price cuts and next gen cards? I think the computer philosophy of "only upgrade when you have to" is very relevant right now. Personally, I'm trying to hold out for the 2nd gen DX11 cards.

I've heard that the next Radeons will actually be a half step because of manufacturing issues. What does nVidia have in the pipe for around the same time frame?
Click to expand...

You just saw an effective price-cut with the GTX 260. If you want performance for <$200, there is not substitute. For the last 10 months, that performance was >$250.00 and probably closer to $300.00.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top