ATI RV740 review/preview

Yes you have a point, it might take more for ATI to get there.

Now that I think of it, it would really be ironic if Nv were to sitch to GF faster.

They did stray from TSMC with the NV41 project at IBM, even though that didn't work out particularly well.
 
Because presumably they will switch to new processes faster than TSMC.

Besides that, I also had the impresion that because they will start producing bulk silicon, they will aim it at gpu makers. Who else would need it? Cell phone processors, maybe.. Xilinx and the like?

By the way, does GF still has MY or any other Asian fabs? I have never heard anything about them.
 
Global Foundries is made up of AMD's current and future fabs.
At present, that means Dresden.
In the future, it is planned to also include the Luther Forest fab in NY.

AMD's graphics wing has additional pressures to use GF.
First, the internal reorganization appears to have made the former ATI less of an independent division, and the GF deal is basically what is keeping AMD the CPU manufacturer alive.

GF going under means AMD goes under, CPU and GPUs together.
If forcing GPUs onto GF buys AMD more time by keeping Abu Dhabi happy, it might be considered.
 
ATi Graphics WILL use GF simply because they will get integrated in Fusion, they ARE AMD, AMD will use GF for their processors, including the graphics core. thus work on 28nm GF production has already started at ATI. That's the only way I can see it.

You might even see TSMC take UMC's position of old where main (sub)28nm production is at GF and the backup production runs at TSMC
 
As for nV's recent statements that it is considering a switch to GF, is probably just to put some pressure onto TSMC
 
There were rumours that 40nm yields would be improving significantly by August(?), though not suggested in this update. Bergman indicates that AMD won't be dropping TSMC anytime soon.
 
Last edited by a moderator:
During GF's development of its bulk process, wouldn't an ideal development partner be AMD's graphics group? Wouldn't this development have started before GF came into existence?

Jawed
 
I kind of figured AMD would use both initially. Start the bulk process with one of their lines at GF. Still continue with what they are producing at TSMC , keeping their "eggs" in multiple baskets.

So yes AMD would be a still be a TSMC customer until they have enough capacity to make use of Global. It would be a VERY bad thing to try and sell your capacity to others when you say we will only use TSMC for our own graphics cards. But everyone understands gradual weening off an existing supplier to the your own when you have capacity.
 
Doesn't AMD need the bulk process so they can go the HKMG route to better compete with Intel? I wouldn't think GPU production would play much of a role in this decision...
 
But the 45nm process is what CPU fabs use. They don't seem to do the half nodes, so not sure if it means much that TSMC has 40 and GF/AMD has 45. Even Intel never bothered with half nodes.

I understand they are targetting 32 fast and that is what they are trying to sell to potential customers. I think that when the fab in New York is up and running, we will see all GPU's off TSMC as soon as reasonable, technically. The key to making any money on Fabs is to have them running as close to full capacity as possible.

GF is targetting to get as many customers as possible so therefore BULK is being added. ATI will be needed to help fill that new fab and that will also aid by example in the selling of additional capacity to others. By putting ATI on GF that potentially makes nVidia look to GF "if" its process does better. TSMC's current problems is bad for ATI now but is probably a godsend for GF in their sales pitch.
 
But, but, Charlie says 40nm production is now good to go, despite all the bad rumors... :LOL:

No, I'm pretty sure rjc mentioned that last month as well. they did a lot of work on 40nm and should be getting increased yields from the waffers in August. Probably the same source for the rumor.
 
But the 45nm process is what CPU fabs use.
ATI makes GPUs.
We're talking about migrating their GPU manufacturing.
GPUs are big so half nodes are very important to GPUs & GF has no track record there.
 
Hoom, I understand that but comparing GF to TSMC is not fair at this point. Until GF was created, it was AMD and there was not enough capacity to make CPU's let alone additional GPU's for ATI. So we had ATI using tsmc. WIth the creation of GF and the cash to make a new FAB, they announced they would also do BULK that is needed for processes like GPU's.

Therefore GF (newly created from AMD that did only CPUs) naturally would only have 45 and not 40. However with the creation of GF and the ability to finally have another fab they now can target stuff that ATI does, however I understand that they are taking a miss on 45/40 process for bulk and aiming for 32 . Maybe they will start the work on BULK early and get stuff ready for the new bulk process and who better to do the first 32 but an in-house requirement of ATI. My original point is they probably will still use TSMC for stuff already being manufactured and test 32 on a selective part.
 
The reality is that nobody really wants SOI, except for high performance CPU vendors like IBM and AMD. And not all CPU vendors want it - Intel, Sun, Fujitsu all passed, as did most low power companies like PA-Semi, Cavium, RMI, etc.

SOI is expensive and isn't very suitable for products with lower margins (e.g. 30-40%), or high volume. It is very useful for space, as SOI has better resilience to SERs than bulk silicon.

So to get additional volume business, GF really needs bulk.

AMD hadn't done bulk in a long time, and their circuit libraries for SOI aren't going to be very helpful for a bulk process as the two behave differently. So they need to spend quite a bit of time getting up to speed there. Right away that means that doing bulk at 40 or 45 is useless, since by the time its production ready, people would be using 32.

Also, note that "32nm" vs. "40nm" is really just a marketing distinction. The only thing that matters are physical characteristics like contacted gate pitch, SRAM density, Ion vs. Ioff, etc. Once upon a time, 180nm referred to a physical characteristic, but 32nm doesn't really. it's more short hand for a certain class of process technologies, but within that class there's a great deal of variation.

For example, IBM's 32nm bulk process has Ion vs. Ioff curves that look a lot more like a Fujitsu 45nm process rather than a 32nm process from TSMC (let alone Intel).

GF has to make some huge investments to compete with TSMC for cutting edge business, and it will probably take 3-5 years to see a pay off. Fortunately, the folks at Abu Dhabi have a long term mindset (I hope) and will see it through.

My guess is that GF will get combined (somehow) with IBM's semiconductor operations to put the R&D under one roof.

one other thing to contemplate is that foundries probably make little money off cutting edge tech, but use advanced customers like Qualcomm, NV and ATI to finance a process which is later sold to other people. i.e. I bet TSMC makes a lot of profit on 130nm and above.

Another issue is whether customers will migrate to newer processes. As design costs go up, fewer and fewer folks will be able to justify moving to a new process. Plenty of folks are still happy with 180nm and may never move (except under extreme duress).

Lots of stuff to think about. Bottom line though is that ATI has a lot of physical design people who will be very helpful for GF in moving to bulk.

DK
 
Good explanation!

I agree with the idea that ATI will be extremely helpful in getting the bulk process ready. They probably be the first ones using the Bulk. Once one line is successful, then they will migrate off of TSMC as they switch each line.
 
GF has to make some huge investments to compete with TSMC for cutting edge business, and it will probably take 3-5 years to see a pay off. Fortunately, the folks at Abu Dhabi have a long term mindset (I hope) and will see it through.
Thanks to large amounts fo cash, I think ATIC are going are going a more direct route - WSJ Part 1 and Part 2. Haven't heard anything for a couple of weeks now though.

The idea is that GF is merely a R&D centre for an ecosystem of Fabs. If the ATIC bid for Chartered doesnt work am sure we will be reading about an ATIC attempt on UMC in a few months or so.

Around Computex in what i at first thought looked like an interesting "pincer movement" AMD initiated contracts for Chartered to produce chipsets for them and also (maybe) contracts at UMC for their low end GPUs. The idea being to provide ATIC with enough inside knowledge of these companies to make successful takeovers.

Above is more than slightly conspirational, there is also an equal possibility AMD is searching for ways to get their production costs down as much as possible, particularly on their volume lines. What AMD produces at TSMC is an interesting question, the way they have been behaving lately looks like TSMC is also going to get an increasing share of their business. ie Something like GF - High end CPUs, TSMC - High end GPUs, Fusion, UMC/Chartered - Low end GPUs, chipsets.
 
Back
Top