Friday, September 8th 2017

AMD To Change Suppliers for Vega 20 GPUs on 7nm, HBM2 Packaging for Vega 11

AMD's RX Vega supply has seen exceedingly limited quantities available since launch. This has been due to a number of reasons, though the two foremost that have been reported are: increased demand from cryptocurrency miners, who are looking towards maximizing their single node hashrate density through Vega's promising mining capabilities; and yield issues with AMD's Vega 10 HBM2 packaging partner, Advanced Semiconductor Engineering (ASE). It's expected that chip yield for Vega 10 is also lower per se, due to it having a 484 mm² die, which is more prone to defects than a smaller one, thus reducing the amount of fully-enabled GPUs.

AMD's production partner, GlobalFoundries, has historically been at the center of considerations on AMD's yield problems. That GlobalFoundries is seemingly doing a good job with Ryzen may not be much to say: those chips have incredibly small die sizes (192 mm²) for their number of cores. It seems that Global Foundries only hits problems with increased die sizes and complexity (which is, unfortunately for AMD, where it matters most).
Due to these factors, it seems that AMD is looking to change manufacturers for both their chip yield issues, and packaging yield problems. ASE, which has seen a 10% revenue increase for the month of August (not coincidentally, the month that has seen AMD's RX Vega release) is reportedly being put in charge of a much smaller number of packaging orders, with Siliconware Precision Industries (SPIL), who has already taken on some Vega 10 packaging orders of its own, being the one to receive the bulk of Vega 11 orders. Vega 11 is expected to be the mainstream version of the Vega architecture, replacing Polaris' RX 500 series. Reports peg Vega 11 as also including HBM2 memory in their design instead of GDDR5 memory. Considering AMD's HBM memory history with both the original Fury and and now RX Vega, as well as the much increased cost of HBM2's implementation versus a more conventional GDDR memory subsystem, this editor reserves itself the right to be extremely skeptical that this is true. If it's indeed true, and Vega 11 indeed does introduce HBM2 memory to the mainstream GPU market, then... We'll talk when (if) we get there.

As to its die yield issues, AMD is reported to be changing their main supplier for their 7 nm AI-geared Vega 20 from GlobalFoundries to Taiwan Semiconductor Manufacturing Company (TSMC), who has already secured orders for AI chips from NVIDIA and Google. TSMC's 7nm and CoWoS (chip-on-wafer-on-substrate) capabilities have apparently proven themselves enough for AMD to change manufacturers. How this will affect AMD and GlobalFoundries' Wafer Agreement remains to be seen, but we expect AMD will be letting go of some additional payments GlobalFoundries' way.Sources: DigiTimes on Vega 11, DigiTimes ASE, DigiTimes on vega Shortages
Add your own comment

62 Comments on AMD To Change Suppliers for Vega 20 GPUs on 7nm, HBM2 Packaging for Vega 11

#1
efikkan
Vega11 on HBM would be a huge mistake, AMD needs a volume GPU in this segment. There is also a large gap between GTX 1060 and GTX 1070 for anyone to fill. And Volta is coming early next year…
Posted on Reply
#2
Fouquin
efikkan said:
Vega11 on HBM would be a huge mistake, AMD needs a volume GPU in this segment. There is also a large gap between GTX 1060 and GTX 1070 for anyone to fill. And Volta is coming early next year…
I think RTG designed themselves into a corner with Vega in general and are at the point of no return to change over to GDDR. While it's true we don't know what they taped Vega 11 to be, the popular chorus of 'switch to GDDR5(x)' requires a completely reworked ASIC, PCB and reference HSF assembly (assuming they don't reuse an old design). All of that costs time and money that AMD can no longer afford for Vega. At this point I would not be surprised if the other execs told Raja to focus his teams on Navi and just get all of Vega out the damn door.

So yeah, HBM on Vega 11 is probably a massive mistake, but if that's the way they taped it out then it's a mistake they're stuck with.
Posted on Reply
#3
efikkan
Sure, it takes a while to redesign a memory controller.
If AMD have chosen HBM2 for Vega11 as well, they've dug their own grave. These chips would have to last until Navi.
Posted on Reply
#4
Thimblewad
I guess:
  1. Keep GlobalFoundries as supplier for Ryzen?
  2. New GPU supplier?
profit for everyone?
Posted on Reply
#5
HisDivineOrder
If this is true, AMD really needs to fire the people they're using to do future forecasting for technology design. Because HBM on their mainstream product is just a horrible decision. It was already a bad enough decision on their high end part...
Posted on Reply
#6
theoneandonlymrk
HisDivineOrder said:
If this is true, AMD really needs to fire the people they're using to do future forecasting for technology design. Because HBM on their mainstream product is just a horrible decision. It was already a bad enough decision on their high end part...
While I get you i think something to remember is HBM have yeilds have bins too plus I personally see the Vega chip type's as Amds stepping stone to Navi , if they indeed do tie 2-4+ gpus on an interposer with each having Hbm they clearly need to put some time in the lab with something similar and these could be seen as affirmative steps in that direction, ie what is the best config 2x4096 ,4x4096 4x2356 who knows , It is not going to help with costs but maybe they see mining sticking around, and the market remaining at higher norms.
Posted on Reply
#7
RejZoR
Advantage of split GPU's would also be heat distribution. Even if it's higher as a whole, it won't be concentrated in a "tiny" spot like current massive monolithic GPU's. It's all speculation for Navi, but if they can really pull it off, it's gonna pay off big time. The problem is taking away 300W from a single spot. Taking away 300W of heat from a larger surface is a lot easier. We'll see in about 2 years...
Posted on Reply
#8
cadaveca
My name is Dave
Raevenlord said:
As to its die yield issues, AMD is reported to be changing their main supplier for their 7 nm AI-geared Vega 20 from GlobalFoundries to Taiwan Semiconductor Manufacturing Company (TSMC), who has already secured orders for AI chips from NVIDIA and Google. TSMC's 7nm and CoWoS (chip-on-wafer-on-substrate) capabilities have apparently proven themselves enough for AMD to change manufacturers. How this will affect AMD and GlobalFoundries' Wafer Agreement remains to be seen, but we expect AMD will be letting go of some additional payments GlobalFoundries' way.
I had been begging AMD to switch to GloFo for GPUs because of info leaks on unreleased GPUs that came out of TSMC. Turns out, this time, with VEGA, nobody had any real idea as to what the real story was with VEGA, so perhaps moving to GloFo did improve the security of the data on AMD's unreleased products. It will be very interesting to me to see them go back to TSMC, and to see how the next launch works.
Posted on Reply
#9
Xzibit
cadaveca said:
I had been begging AMD to switch to GloFo for GPUs because of info leaks on unreleased GPUs that came out of TSMC. Turns out, this time, with VEGA, nobody had any real idea as to what the real story was with VEGA, so perhaps moving to GloFo did improve the security of the data on AMD's unreleased products. It will be very interesting to me to see them go back to TSMC, and to see how the next launch works.
This could be foundry diversifying due to demand both in CPU and GPU. If in fact they are gaining market share.

If VEGA is being received well they might be hedging their bets. We still havent seen & heard much of VEGA x2 since ROCM leaks and since releases started moving months back we might not see it until next year Q1 or even Q2. Imagine the ruccus thats going to cause if things havent calmed down
Posted on Reply
#10
StrayKAT
Curious where the original Global Foundry fab was located?

Why must everything be in freaking Taiwan.

Well, except Intel thankfully.

edit: Even more sad since AMD used to be in my home state... along with others.
Posted on Reply
#11
xkm1948
I am wondering the spec of possible Vega11. Gotta to be a further cut down version of Vega56. Or a Vega56 with GDDR5X?
Posted on Reply
#12
cadaveca
My name is Dave
StrayKAT said:
Curious where the original Global Foundry fab was located?

Why must everything be in freaking Taiwan.

Well, except Intel thankfully.

edit: Even more sad since AMD used to be in my home state... along with others.
iirc, GloFo has manufacturing facilities in Singapore, Germany, and the US. Not in Taiwan at all. They might have an office in Taiwan though. GloFo used to be AMD-owned, but they split off the foundry business many years ago when they went to an "Asset-Light" business model, and since then, the only new things that I can think of for foundry stuff was in Fishkill? I cannot recall directly. I'm sure that they must have this sort of info on their site though. Anyway, so iirc, they had their main foundry in Germany then, Dresden I think.
Posted on Reply
#13
StrayKAT
cadaveca said:
iirc, GloFo has manufacturing facilities in Singapore, Germany, and the US. Not in Taiwan at all. They might have an office in Taiwan though. GloFo used to be AMD-owned, but they split off the foundry business many years ago when they went to an "Asset-Light" business model, and since then, the only new things that I can think of for foundry stuff was in Fishkill? I cannot recall directly. I'm sure that they must have this sort of info on their site though. Anyway, so iirc, they had their main foundry in Germany then, Dresden I think.
I never looked at their site. It's usually difficult to find for most companies so I gave up. Thanks.

As for Taiwan, I mean this new operation they seem to be moving to (not GoFlo).
Posted on Reply
#14
cadaveca
My name is Dave
Oh, that's TSMC, and yes, most definitely is in Taiwan. ATI used TSMC prior to being bought by AMD, and AMD continued to use TSMC for a while, too, but then recently switched to GLoFo for their GPUs.
Posted on Reply
#15
semitope
all making assumptions about the cost of HBM to AMD.
Posted on Reply
#16
EarthDog
If it was cheaper and more available, wouldn't it make sense for there to be more of them? Why would they trickle it into the market when miners and gamers alike would (and are) swoop these up?
Posted on Reply
#17
Perihelion
efikkan said:
Vega11 on HBM would be a huge mistake, AMD needs a volume GPU in this segment. There is also a large gap between GTX 1060 and GTX 1070 for anyone to fill. And Volta is coming early next year…
Volta is NOT coming soon. 1080 and TI are selling well enough. And there are also Miners.. Damn Miners... neither of them will rush
Posted on Reply
#18
eidairaman1
The Exiled Airman
cadaveca said:
iirc, GloFo has manufacturing facilities in Singapore, Germany, and the US. Not in Taiwan at all. They might have an office in Taiwan though. GloFo used to be AMD-owned, but they split off the foundry business many years ago when they went to an "Asset-Light" business model, and since then, the only new things that I can think of for foundry stuff was in Fishkill? I cannot recall directly. I'm sure that they must have this sort of info on their site though. Anyway, so iirc, they had their main foundry in Germany then, Dresden I think.
I believe that was the famous FAB51.

I doubt glofo can adapt properly or hold up a dual front, perhaps another fab needs to handle gpus since glofo is not up to the task, same can be said when amd changes cpu designs too...
Posted on Reply
#19
FordGT90Concept
"I go fast!1!11!1!"
Taiwan, Malaysia, etc. often have final assembly which appears on the integrated heat spreader. Most electronic components are individually made in the west (e.g. Intel has a lot of fabs in the USA and GloFo has a few).

I'm more concerned about the memory. Presumably Vega 11 is only going to get 4 GiB but how are they going to package it? 1x 4 GiB or 2 x 2 GiB? 1 x 4 GiB would be cheaper to manufacturer but would make for an odd looking chip.

I heard that most of the supply problems did stem from the interposer integrator. Not surprised at all that they're changing.
Posted on Reply
#20
StrayKAT
FordGT90Concept said:
Taiwan, Malaysia, etc. often have final assembly which appears on the integrated heat spreader. Most electronic components are individually made in the west (e.g. Intel has a lot of fabs in the USA and GloFo has a few).
Huh. Most components in general are made in the West or just these companies?

Where can I read up on this? I don't mean to derail, but it's a subject of interest to me.
Posted on Reply
#22
StrayKAT
FordGT90Concept said:
https://en.wikipedia.org/wiki/List_of_Intel_manufacturing_sites
https://en.wikipedia.org/wiki/GlobalFoundries
https://en.wikipedia.org/wiki/Semiconductor_device_fabrication

"Packaging" is rarely done in the USA (or west in general).
I knew about Intel, but not GF. Actually been out of the loop on PCs for years as well... Last time I paid attention to AMD, they still operated more in Austin near here. So yeah... it's been awhile.
Posted on Reply
#23
the54thvoid
StrayKAT said:
I never looked at their site. It's usually difficult to find for most companies so I gave up. Thanks.

As for Taiwan, I mean this new operation they seem to be moving to (not GoFlo).
TSMC does stand for Taiwanese Semiconductor Manufacturing Company. It's based there for a reason.
Posted on Reply
#24
Vya Domus
So much for those "huge gains on 7nm" , to me this would be another confirmation that GloFo sucks and it's dragging AMD down.

That being said , it's very good news that they are trying to switch to TSMC.
Posted on Reply
#25
FordGT90Concept
"I go fast!1!11!1!"
Threadripper and EPYC...Navi...Infinity Fabric...HBCC

Certainly I can't be the only one that sees where this is going. AMD wants to produce smaller GPU chips and stick a bunch of them on a package exposing a single GPU to the system.

Vega may be the last monolithic, high performance GPU from AMD.
Posted on Reply
Add your own comment