ATI DX11 Cards Pushed Back to November?

[Gikaseixas]

poor Keys you're being misquoted left and right hehehe.

As long as the majority of us see the real meaning in your msg's it's ok

sometimes you do get out of the line a bit but not this time, you've been accused while being 100% correct.

back to the topic, i wouldn't read too much into this, let's wait a bit more. As we get closer to launch date we will know more and more details

 

[Keysplayr]

Originally posted by: Gikaseixas
poor Keys you're being misquoted left and right hehehe.

As long as the majority of us see the real meaning in your msg's it's ok

sometimes you do get out of the line a bit but not this time, you've been accused while being 100% correct.

back to the topic, i wouldn't read too much into this, let's wait a bit more. As we get closer to launch date we will know more and more details

Hehe. Woe is me.. 🙂

Anyway, I think due to the delay in 40nm fabbing from TSMC, most likely AMD and Nvidia will be launching their products around the same time. Hopefully before the holidays.

 

[Idontcare]

Originally posted by: Keysplayr
Anyway, I think due to the delay in 40nm fabbing from TSMC, most likely AMD and Nvidia will be launching their products around the same time. Hopefully before the holidays.

If the delay is expected/projected to impact Q4 plans I would not be surprised if we see AMD qualify GF's to produce 4x nm GPU parts in the meantime as well as NV qualifying UMC for 4x nm parts too.

It is not the preferred route as it raises GPU development costs all the more but it is the prudent thing for management to do from a risk-management standpoint and I'm sure the critical decisions have already been made by now.

 

[Scali]

Did anyone else think about how this is VERY convenient for Intel and its Larrabee, if the problems with 40 nm productions for GPUs keep up?
Intel will have no problems churning out 45 nm chips in high volumes, and by the end of the year, they'll be starting mass-production of 32 nm aswell.

I don't think AMD can move GPU production to GF on short notice. Their 45 nm process is completely different from TSMC's 40 nm process, and requires a redesign of the chip.

 

[Keysplayr]

Originally posted by: Idontcare

Originally posted by: Keysplayr
Anyway, I think due to the delay in 40nm fabbing from TSMC, most likely AMD and Nvidia will be launching their products around the same time. Hopefully before the holidays.

If the delay is expected/projected to impact Q4 plans I would not be surprised if we see AMD qualify GF's to produce 4x nm GPU parts in the meantime as well as NV qualifying UMC for 4x nm parts too.

It is not the preferred route as it raises GPU development costs all the more but it is the prudent thing for management to do from a risk-management standpoint and I'm sure the critical decisions have already been made by now.

I agree. It would have been prudent of both companies to have both fabs in mind when designing the cores. I remember when Nvidia actually used IBM's fabs (for a limited amount of cores) for NV40? I think? Might be way off here, but I remember something.... ah google.

http://www.eetimes.com/news/se...tml?articleID=10801052

"SAN JOSE ? Moving into the damage-control mode, Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) today (March 27, 2003) restated that the foundry alliance between IBM Corp. and Nvidia Corp. would not have a material impact on the Taiwanese company.

TSMC today sent a report from an analyst via e-mail to SBN, which states that the Taiwanese foundry giant will continue to be Nvidia's primary foundry for the foreseeable future. The report, issued by Pacific Crest Securities Inc., also states that TSMC would process 95 percent of Nvidia's wafers in 2003.

However, the report also states that IBM would make Nvidia's next-generation NV40 or Quadro workstation graphic chip and that TSMC stands to lose $7.5-to-$11 million per quarter to IBM in 2003. That figure could go up to $35-to-$45 million per quarter in 2004 as a result of the IBM-Nvidia foundry deal, according to the report.

TSMC initially moved into the damage control phase on Tuesday, when IBM's Microelectronics Division announced a multi-year foundry alliance with TSMC's largest customer ? Nvidia. IBM will make Nvidia's next-generation GeForce graphics processor line, based on its 0.13-micron technology." (click link for full article)

So Scali, If Nvidia was able to utilize both IBM and TSMC in the same time frame, and you have to assume that IBM's fab process is nothing like TSMC's, then I believe it's not a stretch or as cumbersome of an undertaking for Nvidia to use UMC or AMD to use GF for limited production runs. At least until TSMC gets it's buttocks in gear.

 

[Scali]

Originally posted by: Keysplayr
So Scali, If Nvidia was able to utilize both IBM and TSMC in the same time frame, and you have to assume that IBM's fab process is nothing like TSMC's, then I believe it's not a stretch or as cumbersome of an undertaking for Nvidia to use UMC or AMD to use GF for limited production runs. At least until TSMC gets it's buttocks in gear.

That's the problem, you assume.
Now, this was long ago, and I don't know exactly what specs IBM and TSMC had at the time.
If IBM had a smaller/more advanced process, then yes, nVidia could probably have manufactured chips designed for TSMC with little or no changes... We also don't know how long it took to prepare for this transition to IBM's process.

We DO know that GF has a very different process from TSMC. TSMC's minimum feature size is smaller, obviously (40 nm vs 45 nm), which could already be a dealbreaker, the chip would have to be 'magnified' to get it on a 45 nm process. Then there may be differences in the materials used and all that (high-k, strained silicon, hafnium, and all that sort of details which affect how you design your chip).
It will take months to develop new masks, and verify the modified chip design AGAIN for the new manufacturing process etc. They can't just say "Oh, we'll just produce our chips at GF then".

UMC and TSMC are far more similar (if not identical, they could both be buying the same machines from the same supplier).
Either way, I don't think anyone can touch Intel in production.

 

[Keysplayr]

Originally posted by: Scali

Originally posted by: Keysplayr
So Scali, If Nvidia was able to utilize both IBM and TSMC in the same time frame, and you have to assume that IBM's fab process is nothing like TSMC's, then I believe it's not a stretch or as cumbersome of an undertaking for Nvidia to use UMC or AMD to use GF for limited production runs. At least until TSMC gets it's buttocks in gear.

That's the problem, you assume.
Now, this was long ago, and I don't know exactly what specs IBM and TSMC had at the time.
If IBM had a smaller/more advanced process, then yes, nVidia could probably have manufactured chips designed for TSMC with little or no changes... We also don't know how long it took to prepare for this transition to IBM's process.

We DO know that GF has a very different process from TSMC. TSMC's minimum feature size is smaller, obviously (40 nm vs 45 nm), which could already be a dealbreaker, the chip would have to be 'magnified' to get it on a 45 nm process. Then there may be differences in the materials used and all that (high-k, strained silicon, hafnium, and all that sort of details which affect how you design your chip).
It will take months to develop new masks, and verify the modified chip design AGAIN for the new manufacturing process etc. They can't just say "Oh, we'll just produce our chips at GF then".

UMC and TSMC are far more similar (if not identical, they could both be buying the same machines from the same supplier).
Either way, I don't think anyone can touch Intel in production.

Ok, a few questions then.

When was the last time TSMC and IBM had anything to do with each other? I ask because, how could their process technologies be similar to each other if they never had any dealings with each other. Kind of like comparing Intel to TSMC (differences only).

How does having a smaller, more advanced process make it any easier or little to no changes in design? I honestly don't know. Could you explain? Also, about the transistion to IBM's process. These are huge multibillion dollar companies. You don't get there without having a "Plan B" even if it's not the best Plan B. I would wager to guess that these companies have the sense to realize from past transgressions, that things can and will go wrong. "What can we do to minimize the impact if example A happens. What about example B..." and so on.

And about TSMC's "40nm" label. Idontcare had a nice writeup about that. TSMC fell quite a bit behind in their 45nm development, and to keep current, they gave their 45nm process label a new name. "40nm". They could call it whatever they wish. For all intents and purposes, TSMC's "40nm" process IS their 45nm process.

And about AMD saying, "Well, we'll just produce our chips at GF then. What would you say if I said, they planned for this all along? Why wouldn't they? AMD and GF are practically sisters.

"UMC and TSMC are far more similar (if not identical, they could both be buying the same machines from the same supplier)."

Just the same, maybe they didn't.

Look, there are far too many maybe's here to develop any sort of conclusive argument or outcome thereof. I do believe in those Plan B's however. Every Multi-Billion dollar company should have one, or seven.

 

[Scali]

Originally posted by: Keysplayr
When was the last time TSMC and IBM had anything to do with each other? I ask because, how could their process technologies be similar to each other if they never had any dealings with each other. Kind of like comparing Intel to TSMC (differences only).

Just because they don't directly have anything to do with eachother doesn't mean that they can't use similar technologies.
In fact, a lot of the chip manufacturing equipment isn't developed by these companies themselves, but is bought from companies such as ASML.
Aside from that, things have diversified in recent years. In the old days, there were far more similarities between different manufacturers in general, because there just weren't as many different technologies around (many technologies were mainly developed because of issues that started arising with 90 nm and smaller processes, trying to reduce leakage and such).

Originally posted by: Keysplayr
How does having a smaller, more advanced process make it any easier or little to no changes in design? I honestly don't know. Could you explain?

Well, a '45 nm' process simply says that the *minimum* feature size is 45 nm, nothing more. It's just the upper limit of the process. You can just as easily use the process to manufacture larger chips, like 65 nm or 90 nm etc. But you can't do it the other way around, obviously...

Originally posted by: Keysplayr
And about TSMC's "40nm" label. Idontcare had a nice writeup about that. TSMC fell quite a bit behind in their 45nm development, and to keep current, they gave their 45nm process label a new name. "40nm". They could call it whatever they wish. For all intents and purposes, TSMC's "40nm" process IS their 45nm process.

Yes, but their 45 nm process is not the same as AMD's, IBM's, or Intel's. It's not going to be entirely trivial to transition to a different process, even if the '40 nm' name is just TSMC marketing.

Originally posted by: Keysplayr
And about AMD saying, "Well, we'll just produce our chips at GF then. What would you say if I said, they planned for this all along? Why wouldn't they? AMD and GF are practically sisters.

I doubt that AMD had planned this. They missed the perfect time to transition to GF: the 4770. AMD has been struggling with TSMC and 40 nm GPUs for ages now. If they had planned it all along, wouldn't we have had GF-built 4770s now?

Originally posted by: Keysplayr
Look, there are far too many maybe's here to develop any sort of conclusive argument or outcome thereof. I do believe in those Plan B's however. Every Multi-Billion dollar company should have one, or seven.

Well, AMD and Intel don't really have plan B's. We can name plenty of examples of both companies struggling with a new manufacturing process, and never outsourcing their production to compensate (okay, AMD has outsourced SOME of its production, but because of x86 licensing issues, they NEED to produce their own chips).
I think if nVidia and AMD have any plan B for GPUs at all, it's UMC, nothing else.

 

[BFG10K]

Originally posted by: tuteja1986

LOL :!

Your intelligence really surprises me , So if you use Nvidia card you need 2GB of system ram while if you use ATI card you need 6GB :!
Your logic is flawed and stupid :!

Feel free to attack his argument, but leave his intelligence out of it.

Video Mod BFG10K.

 

[VisceralM]

Was there ever a consensus on this? I'm thinking of picking up a 4890 at the end of August but if the dx cards are out earlier than November...

 

[ronnn]

Lots of places seem to still be saying september with low availability - but who knows. AMD has been good at keeping their secrets lately. I am in the same boat - I have a new system picked out - but than don't push the buy button.