"AMD has tapped both Globalfoundries and TSMC to manufacture the new GPUs on their 28nm process. But since both the processes are vastly different, AMD would need independent GPU designs for each foundry."
I hope that means each foundry gets a different class of GPU to produce. Like say, TMSC gets the 79xx-series and Global gets the 77xx-series. I really don't want to be playing lottery with GPUs if BOTH foundries produce GPUs AMD markets under the same brand name.
Person A: "I just got myself a hot new 7980!" Person B: "Which version? One from TMSC or one from Global?" Person A: "Huh? What's the difference? Does it matter?" Person B: "Different designs for each manufacturer due to process differences..." Person A: "Uh, laymans please?" Person B: "Your performance mileage will vary depending on which design you got."
It's kinda of like the SSD thing with I think OCZ? I believe there was a noticebale performance discrepancy in a single SSD model line from them depending on if it used I think Toshiba or IMFT NAND chips. I do'nt recall the exact details, but I know AnandTech covered it.
I don't think it's quite the same with GPU's or CPU's. The problem with the NAND chips that were produced by different foundries is that the specs (i.e. speeds) turned out different as a result. If the same GPU was being fab'd at two different foundries, they would either perform exactly the same or the clocks would be different. If the clocks were different, it would be very obvious - and changing the clocks have always meant a different class of product, so it would be unthinkable for AMD to market different clocked chips as the same chip.
What would be a more likely (but subtle) problem is that the tolerance/overclocking-headroom would likely be different, depending on which foundry it came from.
But I don't think it makes sense that they'd produce the same chip at two different foundries, because as it was pointed out, you'd have to have slightly different designs for each foundry. What's more likely is what you pointed out - different classes of GPUs being fabbed at different foundries.
So, I guess in summary: thanks for pointing that the (likely?) possibility that they'll be fabbing different classes of GPUs at the different foundries :P
Of course, this would also kill some of the mods that would change a lower end product into a higher end product with a simple firmware update, which I am sure AMD would not mind.
As you said the quality of the chip could be different and this could lead to more heat on one over the other if they were clocked the same. This is what people are worried about. It is unlikely if you think practically about it but companies have been doing it with all sorts of tech products for a long time so maybe it is not really that unlikely.
I think you'll likely see certain price ranges coming from certain foundries. So 7200-7500 would come from one and 7600-7900 would come from the other. It would be insane to design the whole line up to made by both sides. However, designing products for different product segments for one or the other wouldn't be nearly as intense.
TSMC and GlobalFoundries both use ASML machines for manufacturing - so at their core, processed will be the same. No need to differentiate the designs.
At 130nm, there were still many major players in the equipment industry - it's not the same as now, where ASML owns 80% of the market...
According to Xbitlabs AMD will use bulk process which is already different than TSMC HP 28nm Nvidia will most likely use. If the rumors are true, AMD will be ~6 months ahead of Nvidia because the 28nm HP process has problems like 40nm had.
I don't think AMD will make same GPUs in TSMC and GF. In any case isn't GF still behind TSMC and their 28 nm bulk isn't ready yet?
In the same way that AMD caught up to Intel with the original Athlon vs. Pentium 3, it is possible that GF managed to catch up in terms of fab process improvements. We have seen that TSMC has had problems moving to a new process node, so with the help of IBM, GF may have caught up.
GF is second to only Intel and possibly IBM (if low-volume fabs are considered)
If you mean the Charered part of GloFo playing catch-up, you better hold your breath. Chartered was actually able to compete in the last cycle maily due to AMD/GF's help and the 28nm will come from Dresden aka "AMD" part of GF, at least the initial stuff, that is.
Tax "optimization" coupled to high investment-to-turnover ratio. Has pretty much nothing to do with unit's performance as the article is trying to suggest.
I wonder why they are still using TSMC for manufacturing their chips? Is it that Global Foundries still doesn't have the capacity to handle the entire GPU lineup? I would assume it would be much more cost effective to use only Global Foundries to manufacture the chips. Isn't GF's building a new FAB in New York? Wonder what that status is on that...
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
17 Comments
Back to Article
Stahn Aileron - Thursday, July 28, 2011 - link
"AMD has tapped both Globalfoundries and TSMC to manufacture the new GPUs on their 28nm process. But since both the processes are vastly different, AMD would need independent GPU designs for each foundry."I hope that means each foundry gets a different class of GPU to produce. Like say, TMSC gets the 79xx-series and Global gets the 77xx-series. I really don't want to be playing lottery with GPUs if BOTH foundries produce GPUs AMD markets under the same brand name.
Person A: "I just got myself a hot new 7980!"
Person B: "Which version? One from TMSC or one from Global?"
Person A: "Huh? What's the difference? Does it matter?"
Person B: "Different designs for each manufacturer due to process differences..."
Person A: "Uh, laymans please?"
Person B: "Your performance mileage will vary depending on which design you got."
It's kinda of like the SSD thing with I think OCZ? I believe there was a noticebale performance discrepancy in a single SSD model line from them depending on if it used I think Toshiba or IMFT NAND chips. I do'nt recall the exact details, but I know AnandTech covered it.
Paulman - Thursday, July 28, 2011 - link
I don't think it's quite the same with GPU's or CPU's. The problem with the NAND chips that were produced by different foundries is that the specs (i.e. speeds) turned out different as a result. If the same GPU was being fab'd at two different foundries, they would either perform exactly the same or the clocks would be different. If the clocks were different, it would be very obvious - and changing the clocks have always meant a different class of product, so it would be unthinkable for AMD to market different clocked chips as the same chip.What would be a more likely (but subtle) problem is that the tolerance/overclocking-headroom would likely be different, depending on which foundry it came from.
But I don't think it makes sense that they'd produce the same chip at two different foundries, because as it was pointed out, you'd have to have slightly different designs for each foundry. What's more likely is what you pointed out - different classes of GPUs being fabbed at different foundries.
So, I guess in summary: thanks for pointing that the (likely?) possibility that they'll be fabbing different classes of GPUs at the different foundries :P
Targon - Friday, July 29, 2011 - link
Of course, this would also kill some of the mods that would change a lower end product into a higher end product with a simple firmware update, which I am sure AMD would not mind.PubFiction - Sunday, July 31, 2011 - link
As you said the quality of the chip could be different and this could lead to more heat on one over the other if they were clocked the same. This is what people are worried about. It is unlikely if you think practically about it but companies have been doing it with all sorts of tech products for a long time so maybe it is not really that unlikely.jordanclock - Thursday, July 28, 2011 - link
I think you'll likely see certain price ranges coming from certain foundries. So 7200-7500 would come from one and 7600-7900 would come from the other. It would be insane to design the whole line up to made by both sides. However, designing products for different product segments for one or the other wouldn't be nearly as intense.Nighteye2 - Sunday, July 31, 2011 - link
TSMC and GlobalFoundries both use ASML machines for manufacturing - so at their core, processed will be the same. No need to differentiate the designs.At 130nm, there were still many major players in the equipment industry - it's not the same as now, where ASML owns 80% of the market...
Pantsu - Thursday, July 28, 2011 - link
According to Xbitlabs AMD will use bulk process which is already different than TSMC HP 28nm Nvidia will most likely use. If the rumors are true, AMD will be ~6 months ahead of Nvidia because the 28nm HP process has problems like 40nm had.I don't think AMD will make same GPUs in TSMC and GF. In any case isn't GF still behind TSMC and their 28 nm bulk isn't ready yet?
Targon - Friday, July 29, 2011 - link
In the same way that AMD caught up to Intel with the original Athlon vs. Pentium 3, it is possible that GF managed to catch up in terms of fab process improvements. We have seen that TSMC has had problems moving to a new process node, so with the help of IBM, GF may have caught up.mino - Friday, July 29, 2011 - link
GF "catch up" with TSMC?Let laugh out loud!
GF is second to only Intel and possibly IBM (if low-volume fabs are considered)
If you mean the Charered part of GloFo playing catch-up, you better hold your breath. Chartered was actually able to compete in the last cycle maily due to AMD/GF's help and the 28nm will come from Dresden aka "AMD" part of GF, at least the initial stuff, that is.
jfelano - Friday, July 29, 2011 - link
How on Earth did AMD lose 7 million in the gpu business this quarter?? How is that possible when they lead in price/ performance and sales??Mr Perfect - Friday, July 29, 2011 - link
Probably a combination of things: expenses where to high and that great price/performance ratio(for us) means a low margin for them.mino - Friday, July 29, 2011 - link
Tax "optimization" coupled to high investment-to-turnover ratio. Has pretty much nothing to do with unit's performance as the article is trying to suggest.BSMonitor - Monday, August 1, 2011 - link
If they lead in sales, and are losing money.. Means they are not charging enough to cover their expenses.Price/Performance is great until you have to make a profit.
mino - Friday, July 29, 2011 - link
It was Charlie who broke this couple weeks ago. But yeah, AMD prolly decided to confirm given the word was on the street ...minijedimaster - Saturday, July 30, 2011 - link
I wonder why they are still using TSMC for manufacturing their chips? Is it that Global Foundries still doesn't have the capacity to handle the entire GPU lineup? I would assume it would be much more cost effective to use only Global Foundries to manufacture the chips. Isn't GF's building a new FAB in New York? Wonder what that status is on that...minijedimaster - Saturday, July 30, 2011 - link
Hmmm... so sounds like about a year out for production equipment to be installed and working for phase 1 of the new fab in new york.http://fab8update.com/2011/07/15/globalfoundries-f...
austonia - Friday, August 5, 2011 - link
using an HD 5850 for almost 2 years now.. waiting on new 28nm chips to upgrade.looks like they won't make it before Skyrim hits :(
hurry up guys and i'll buy your stuffs :)