AMD Moving 28nm GPU Process To GlobalFoundries

H

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
If I am reading this correctly (translated via Google), AMD has plans to move its 28nm GPU process away from TSMC in favor of GlobalFoundries.

AMD has decided to take advantage of the 28nm SHP ( Super High Performance ) of GlobalFoundries GPU to realize that we will see in the first half of 2015, so as to avoid the high taxes which are currently forced customers to chip in 28nm at TSMC. The 28nm SHP GloFo also should allow higher frequencies, with the same operating voltage, and a less extensive surface of the chip (higher density) compared to 28nm TSMC, thus ensuring both a faster product is less expensive to produce . All this, remember, with excellent yields.
Click to expand...
 
5ldHuh1.jpg
 
It makes sense that AMD actually produce something at GLF than to pay penalties for not living up to contractual business volumes for CPUs. It's a special olympics win at best.
 
Or a way of putting pressure on TSMC. AMD is leaving and taking their ball with them.
 
Corvette said:
Or a way of putting pressure on TSMC. AMD is leaving and taking their ball with them.
Click to expand...
Or what actually did happen: AMD paid GLF a $320 million fine to cut CPU orders for 3 quarters due to lack of demand. That period just ended. AMD moves GPU wafers to GLF and won't have to pay a large fine again, in a time when AMD has had its lowest cash reserves since 2001 (dipped below $1 billion months ago and AMD continues to bleed cash).

Or maybe because everything is hunky-dory and AMD is using its might to put pressure on TSMC, which isn't exactly hurting for low cost 28nm foundry business. :p
 
Or they went with the better deal that was cheaper and gives the results they want.
 
Story says faster clocks at same voltage, But gotta wonder given current heat output for amd gpu's and power draw. Could that cause some issues on those new gpu's?
 
cageymaru said:
Or they went with the better deal that was cheaper and gives the results they want.
Click to expand...
Well, yeah. Spending $320 million for nothing in return was likely a huge motivator to prop up volume manufactured at GLF. Because that's the math AMD is dealing with, not just a relatively small better deal it could have gotten for processed wafers.
 
Yup, I think pxc has it right: ever since AMD got it in their fool head that "real men have fabs," that fab has been bleeding them dry every singe quarter trying to fill idle capacity (due to spotty demand for AMD chips). They got rid of the fab, but not without a supply agreement, so they're STILL paying for idle capacity on a fab they non-longer own :D

This amendment they signed back in March spells it out: If AMD can't meet their chip supply agreement stably through x86 chips alone, they'll bring over a more diversified product line to GF:

http://www.amd.com/en-us/press-releases/Pages/amd-amends-wafer-2014apr1.aspx
 
Depending on how good the info was, a few weeks back it was stated that Glofo 28nm was far better than TSMC. So I guess now it isn't AMD vs Nvidia it is Glofo vs TSMC.
 
If this leads to a better 28nm GPU then I don't really care what the reasoning behind it is. It doesn't sound like either AMD nor NVIDIA will be launching a 20nm or 16/14nm GPU this year.
 
durquavian said:
Depending on how good the info was, a few weeks back it was stated that Glofo 28nm was far better than TSMC.
Click to expand...
I think several things got mixed up in that claim, likely interpreted by someone who doesn't know what they're talking about. :p

Just because AMD is lowering voltage on GLF's 28nm SHP process vs TSMC's HP process, that doesn't mean the process is superior. It's a matter of tuning parameters and restrictions, and making processes available which appeal to the widest range of customers. That, plus yields and gate performance characteristics must be included when judging what's "better" using standard metrics.

If GLF's SHP process works out better for AMD's GPUs, that's a benefit. However, the much larger collection of customers TSMC has than GLF suggests GLF isn't attracting all the boys to the yard with a process which has much more limited appeal. See also: IBM's literally worthless fab business which operated on the same model.
 
TSMC is notorious for horrid yields. Quite certain GloFo is doing much better, which lowers the avg. cost. 20nm at TSMC was *never* going to work for GPUs. Low power ARM CPUs, sure. But never GPUs. Additionally, GloFo licensed Samsung's 14nm process, which AMD could just slide into since the majority of its products aren't with TSMC now. Apple is also leaving TSMC and going back to Samsung for 14nm.
 
Moving its 28nm process to GLOFO doesn't mean they can't produce 20nm GPU's at TSMC... I still hope that 390x will be 20nm
 
i heard a discussion several weeks ago about amd considering the 20nm at tsmc only for APUs, and under no circumstances GPUs
 
Briguyatx said:
Moving its 28nm process to GLOFO doesn't mean they can't produce 20nm GPU's at TSMC... I still hope that 390x will be 20nm
Click to expand...
Highly unlikely that you will see 20nm from TSMC for GPUs, going off what is known about their processes. TSMC's 20nm process is great for low power tablet/mobile chips, not for big power GPUs.
 
You must log in or register to reply here.
Back
Top