ATI Northern Islands (6xxx) also at 40nm?

[Daedalus685]

5890

The 5890 will have to be out by the summer for it to matter at all, not a chance in hell anything less than 40nm will be ready by then.

I think the most likely situation is the 6000 series coming out at 40nm, shrinking part or all of it to 32/28 a bit later, and the 7000 being the first to leave 40nm behind. Though they 'might' delay the 6000 a tad, giving us some 5790 like card at the new node first. Given how they seem to love talking about keeping to the schedule that seems less likely.

 

[lifeblood]

Given what happened with Fermi and the fact they are going direct from 40nm to 28nm, totally bypassing 32nm, a current gen architecture on a new gen process is pretty much mandatory (a tick tock approach, as Intel calls it). For ATI I'm just not sure if they will create the 5000 series refresh architecture on 28nm, or bring out Northern Islands at 40nm then port it to 28nm. Timing of Northern Islands versus availability says it will come out 40nm then transition.

 

[extra]

Fermi being 6 months late is not a "good" thing by any measure.

Yeah...though the damage control is amusing lol

 

[lavaheadache]

Yeah...though the damage control is amusing lol

I couldn't agree more. I have the upmost respect for Keysplayr as a forum participater and fellow gpu enthusiast but you really can't sugar coat a shitty situation and pretend like it's candy.

I will be the first to buy fermi if it is an upgrade from my trifire setup, which sadly i do not feel like a 5870 is. 2 would be but I'm not looking to spend $800 on them or the almost $700 on a 5970.

Honestly, even if Fermi is a golden performer I think that the price for entry will be astronomical and hardly justifiable for someone in my position.

 

[cbn]

Selling HD6xxxx at 40nm is going to be tough unless we are talking a complete architecture overhaul or a 500+ mm 2 die?

I personally am hoping for an architecture change accompanied by "sideport" (to synchronize GPUs).

 

[Munky]

I don't see why having the 6xxx series on 40nm would be a huge problem unless it was designed to use a different node that's not ready. The 5870 is not exactly a huge chip, if Nvidia pulled off a gigantic g80 on 90nm, AMD could also do something similar on 40nm.

On the other hand, who says they absolutely need the 6xxx series next fall? AMD is in a comfortable spot right now, they got a complete top to bottom DX11 lineup, the 5970 will most likely keep the crown until Nvidia releases a dual-chip Fermi, and they will probably release a higher-clocked 5xxx series refresh in a few months which they would ride until next year.

The thing to remember is Nvidia is using the same foundry to build their already-gigantic Fermi on 40nm, so if the smaller node gets delayed, they would end up in an even tougher spot than AMD.

 

[cbn]

On the other hand, who says they absolutely need the 6xxx series next fall? AMD is in a comfortable spot right now, they got a complete top to bottom DX11 lineup, the 5970 will most likely keep the crown until Nvidia releases a dual-chip Fermi, and they will probably release a higher-clocked 5xxx series refresh in a few months which they would ride until next year.

Maybe ATI wants better tessellation performance?

GPGPU?

 

[MagickMan]

Fermi being 6 months late is not a "good" thing by any measure.

Agreed. I don't think the AMD fanbois truly realize that.

 

[LtGoonRush]

I think based on the RV870 Story article, we're going to get another R4800-like GPU that focuses on power efficiency and value. They could perhaps redesign the GPU to eliminate dedicated hardware in exchange for running more things as compute operations. I'm not sure how much fixed-function hardware is left on a modern GPU, but I know at least the video decode hardware are in dedicated units. Some of that hardware may have made sense back when available GPU computing power was relatively limited, and be removable if you have 1600 1Ghz shaders sitting mostly idle. We may even see something nifty like a radical redesign with much better clock/power gating, like how the Intel Lynnfield processors improved over Nehalem.

 

[Plimogz]

A: Why wouldn't ATi, especially if they're working on a whole new architecture, try to stick with the 40nm TSMC tech which is yielding nicely and which they must be familiar with by now? I mean, isn't it pretty much common knowledge that simultaneously moving to a new arch and a new tech is looking for trouble? Plus it isn't exactly as if nicely optimized 40nm is going to be obsolete going towards the end of 2010.

B: You can be sure that the GPU portion of Llano and future AMD Fuzion products, although generally based on ATi dedicated GPU design, will have to be at least tweaked to perform optimally alongside a AMD CPU on SOI. So there has to be a sizeable portion of DAAMIT's resources figuring out how best to get an r800 derivative working on a manufacturing tech which was until recently solely geared to making AMD's CPUs. And, let's face it, this ain't Intel we're talking about here, you can only take on so much at the same time if you happen to be in the dire straits which AMD has been in these last couple of years, and still hope to deliver product to market in a timely, competitive fashion.

C: I'm pretty sure that whoever it was within AMD/ATi that came up with the idea of fabbing rv740 on the then newfangled 40nm TSMC process was given many extremely enthusiastic pats on the back once it became clear how much of an advantage the expertise acquired fabbing the relatively small HD4770 would be with regards to manufacturing the following, much more complicated 5xxx series chips.

D: If the above holds some water, it could be that, at this very moment, DAAMIT is simultaneously working on finalizing their next architecture on TSMC's by now well controlled 40nm tech while working with Global Foundries to iron out the kinks inherent to adapting their dx11 R800 design to AMD's new 32nm manufacturing process, and at the same time preparing a cut-down 5xxx series chip (à la rv740) to maybe go into some risk production on TSMC's upcoming bulk 28nm.

You know, all of this is pure speculation on my part, but I feel pretty confident in saying that, manufacturing process wise, ATi is setting us up for a repeat of their last, very succesful, migration. Except this time, they're working on two fronts: Dedicated GPU (R900) and integrated GPU (Llano). -- And of course, they must have some pretty nice extra resources in the form of AMD engineers whose job it surely is to help the Graphics division guys nail down the integrated GPU portion of Fusion.

 

[toyota]

5890

with 1500 SP?

I just dont see how they can release a new series on the 40nm. how would they increase anything without being outside of reasonable power usage. I mean they had to reduce the clocks on the 5970 so how could thy possible make it any better within the same power envelope.

 

[cbn]

I just dont see how they can release a new series on the 40nm. how would they increase anything without being outside of reasonable power usage. I mean they had to reduce the clocks on the 5970 so how could thy possible make it any better within the same power envelope.

http://www.brightsideofnews.com/new...an-and-northern-islands-use-32nm-process.aspx

These slides from the end of 2009 say 32nm.

 

[toyota]

http://www.brightsideofnews.com/new...an-and-northern-islands-use-32nm-process.aspx

These slides from the end of 2009 say 32nm.

yeah I think most of us know about that. the point of this thread is that now the next gen may just stick with 40nm which seems a little odd to me.

 

[gorobei]

So there has to be a sizeable portion of DAAMIT's resources figuring out how best to get an r800 derivative working on a manufacturing tech which was until recently solely geared to making AMD's CPUs.
.......
D: If the above holds some water, it could be that, at this very moment, DAAMIT is simultaneously working on finalizing their next architecture on TSMC's

What is DAAMIT?

 

[toyota]

What is DAAMIT?

http://www.google.com/search?q=DAAM...s=org.mozilla:en-US:official&client=firefox-a

very first entry: AMD, ATI Merge to Form DAAMIT

 

[gorobei]

pretty weak anagram. (aadmit or something that keeps the "a" at the start would seem more appropriate. plus any word that starts with 2 vowels is automatically funny. i.e. aardvark, aaron burr)

hadn't heard it till today. seems weird to be using it still. the implied pejorative is based on the idea that the merger was a bad idea at the time. given that it seems to be working out, the user is kinda identifying themself as one of those shorter sighted individuals. admittedly i still call them ati every so often.

 

[Lonbjerg]

Fermi being 6 months late is not a "good" thing by any measure.

With $1.3 billion in the bank it is nothing "doomsday" like.
NVIDIA is posting healthy profits...unlike the competition, with losses and a big debt.

In the big picture only fanboys has been affected.

 

[biostud]

the 40nm process will have matured, so it will probably be easier to produce a larger GPU in the end of the year than now.

 

[tommo123]

Given what happened with Fermi and the fact they are going direct from 40nm to 28nm, totally bypassing 32nm, a current gen architecture on a new gen process is pretty much mandatory (a tick tock approach, as Intel calls it). For ATI I'm just not sure if they will create the 5000 series refresh architecture on 28nm, or bring out Northern Islands at 40nm then port it to 28nm. Timing of Northern Islands versus availability says it will come out 40nm then transition.

wouldn't ATi just release a 5k series card (tiny one) on 28nm to test the waters like they did with the 4k series on 40nm? that gave them a lot of feedback on the process and they used that to prepare themselves for it. Afaik, nvidia didn't do that, jumped right in and had no idea what they were getting into.

 

[toyota]

wouldn't ATi just release a 5k series card (tiny one) on 28nm to test the waters like they did with the 4k series on 40nm? that gave them a lot of feedback on the process and they used that to prepare themselves for it. Afaik, nvidia didn't do that, jumped right in and had no idea what they were getting into.

Nvidia has had 40nm gpus for a while now.

 

[tommo123]

was that before the fermi debacle started though ? i was under the impression that they jumped in without testing anything on the process 1st.

 

[Meghan54]

was that before the fermi debacle started though ? i was under the impression that they jumped in without testing anything on the process 1st.

From AT's AMD Showing up to the Fight article:

For its first 40nm GPU, ATI chose the biggest die that made sense in its roadmap. That was the RV740 (Radeon HD 4770).

NVIDIA however picked a smaller die. While the RV740 was a 137mm2 GPU, NVIDIA’s first 40nm parts were the G210 and GT220 which measured 57mm2 and 100mm2. The G210 and GT220 were OEM-only for the first months of their life, and I’m guessing the G210 made up a good percentage of those orders. Note that it wasn’t until the release of the GeForce GT 240 that NVIDIA made a 40nm die equal in size to the RV740. The GT 240 came out in November 2009, while the Radeon HD 4770 (RV740) debuted in April 2009 - 7 months earlier.

 

[Lonyo]

with 1500 SP?

I just dont see how they can release a new series on the 40nm. how would they increase anything without being outside of reasonable power usage. I mean they had to reduce the clocks on the 5970 so how could thy possible make it any better within the same power envelope.

Well, here's an easy question to answer.
AMD has a ~2 billion transistor chip in the HD58xx cards/

NV are making a ~3billion transistor chip called Fermi on the same process as that ~2bn transistor chip. It uses more power.

That means AMD have a theoretical 50% increase in transistors to play with. However will they manage to improve performance?! They can also increase power consumption for a single chip.

 

[Kuzi]

Selling HD6xxxx at 40nm is going to be tough unless we are talking a complete architecture overhaul or a 500+ mm 2 die?

Not necessarily. Remember Anand's excellent RV870 article? He mentioned how the 5000 series had to be "trimmed" to get to the current die sizes, and some performance was lost. That's why the 5000 series is actually 15-20% slower than the 4000 series.

For Northern Islands, ATI could insert whatever they had to cut back (cache?), tweak the architecture a bit and/or add a bit more SP's/TMU's (2200SP?/100TMU?). The end result would be a GPU that is ~40% faster than Cypress, the same performance difference between 4890--->5870. And the size of such a GPU @40nm could be around ~460mm^2.

Of course I'd rather ATI release a completely new architecture on a smaller process, but that is not possible this year.

 

[Kuzi]

the 40nm process will have matured, so it will probably be easier to produce a larger GPU in the end of the year than now.

Exactly. A more mature process and it would be ATI's 3rd GPU architecture @ 40nm, so it would be much easier to produce. But it would end up on the large side, something ATI have been trying to avoid for a while now.