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

[FordGT90Concept]

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.

 

[StrayKAT]

TSMC does stand for Taiwanese Semiconductor Manufacturing Company. It's based there for a reason.

I already knew that. The original article already said that. I myself already said that by mentioning Taiwan in my first question.

I only wanted to know where Global Foundry held their facilities. Not TSMC. This was what cadaveca mistook and we already cleared it up. I hope I've cleared things up here further.

 

[HD64G]

As others already wrote, if Vega 11 is the successor of Polaris 10, to make it have HBM2 VRAM is ultimate stupidity of the highest level imho. GDDR5 11Gb would more than enough for it, economically feasible and bad for mining also as GTX1080 is now because of this type of VRAM. And thus, the perfect combo for budget gamers.

 

[StrayKAT]

As others already wrote, if Vega 11 is the successor of Polaris 10, to make it have HBM2 VRAM is ultimate stupidity of the highest level imho. GDDR5 11Gb would more than enough for it, economically feasible and bad for mining also as GTX1080 is now because of this type of VRAM. And thus, the perfect combo for budget gamers.

That's assuming AMD doesn't want the miners. Looking objectively, it seems like a good thing for them... unfortunately.

 

[HD64G]

That's assuming AMD doesn't want the miners. Looking objectively, it seems like a good thing for them... unfortunately.

For high priced-low volume GPUs like Vega 10 I agree. But not for the mainstream-middle budget and profit level Vega 11. They MUST go to gamers to gain them back from nVidia in massesif possible. Polaris 10 already decreased the gap in market share before the mining boom occured. They need to keep pushing on this direction. It will help them with game devs having more gamers on their side as well.

 

[uuuaaaaaa]

So AMD finally moving to TSMC, great news! (Lets see how they will handle all the global foundries contract situation tho...)

 

[the54thvoid]

As for Taiwan, I mean this new operation they seem to be moving to (not GoFlo).

I already knew that. The original article already said that. I myself already said that by mentioning Taiwan in my first question.

I only wanted to know where Global Foundry held their facilities. Not TSMC. This was what cadaveca mistook and we already cleared it up. I hope I've cleared things up here further.

Apologies, the first quote above said 'not glo-fo', so I thought you meant TSMC. I wasn't trying to be cheeky.

 

[FordGT90Concept]

There's a reason why 8 GiB RX 580s vanished first. Memory demands grow with time when mining. 4 GiB is less attractive to miners but still fantastic for gaming.

 

[StrayKAT]

For high priced-low volume GPUs like Vega 10 I agree. But not for the mainstream-middle budget and profit level Vega 11. They MUST go to gamers to gain them back from nVidia in massesif possible. Polaris 10 already decreased the gap in market share before the mining boom occured. They need to keep pushing on this direction. It will help them with game devs having more gamers on their side as well.

I didn't know they were already closing the gap before mining. Hmm..

But yeah, there's a plethora of things the gaming market brings that it seems stupid to ignore it. Gamers have brand loyalties and future purchases.. not necessarily miners. And working with devs is the other thing. As well as tie in products that miners would never be interested in, like Freesync. Or peripherals.

But fast money has a way of blinding people. And they already got beat down by Intel/Nvidia once.

 

[ppn]

TSMC 7nm is for Arm processors only, keep dreaming. 7nm GPU in 2019.

 

[jabbadap]

TSMC 7nm is for Arm processors only, keep dreaming. 7nm GPU in 2019.

Afaik there's now two nodes on risk production, mobile 7nm FF and high performance compute 7nm FF.

TSMC's 7nm Fin Field-Effect Transistor (FinFET) process technology provides the industry's most competitive logic density and sets the industry pace for 7nm process technology development by delivering 256Mb SRAM with double-digit yields in June 2016. Risk production started in April 2017.

We expect double digit customer product tape-out in 2017.

Compared to its 10nm FinFET process, TSMC's 7nm FinFET features 1.6X logic density, ~20% speed improvement, and ~40% power reduction. TSMC set another industry record by launching two separate 7nm FinFET tracks: one optimized for mobile applications, the other for high performance computing applications.

 

[T1beriu]

1. Vega is 486 mm2.

2. Ryzen die (Zeppelin) is 213 mm2.

3. Vega 11 with HBM is fake news. The original source said it's going to be packaged at the same company as the one that does Vega 10 packages. That's it! Putting a GPU chip on a PCB substrate it's also called "packaging".

 

[silentbogo]

1. Vega is 486 mm2.

2. Ryzen die (Zeppelin) is 213 mm2.

3. Vega 11 with HBM is fake news. The original source said it's going to be packaged at the same company as the one that does Vega 10 packages. That's it! Putting a GPU chip on a PCB substrate it's also called "packaging".

I was busy drinking on my cousin's wedding, so T1beriu beat me to it.
All I want to add is that this whole thing is a speculation based on speculation with a speculation on its tail.
1) Manufacturing chips and packaging is not the same thing
2) GN did a little research based on Vega samples and they came to conclusion that there are at least three packaging partners based on major differences in spacing/filling between GPU and HBM
3) Regardless of how much you want to blame miners, they are still on the same boat with everyone else - there is no vega on the market.

Another thing I can't wrap my head around is the contents of the "source" articles.
First one only mentions that SPIL will keep the contract on packaging (theoretically they are one of those three that package Vega10 right now), which may or may not include future 2.5D/3D stacking (putting HBM on top of GPU or vice versa).
Secon one is about ASE revenues and not a word or a hint about which way their partnership with AMD goes.
Third one is about shortages and possible reasons behind it (e.g. HMB supply, packaging yields etc, but not a word about mining).

Lastly, when will you forget that bullshit from twitter about a claim of 43MH/s at 130W. It was already debunked by... I'm gonna say it with a bit of pride and respect this time.... WCFTech and recreated by a dozen people on youtube (averaging at 175W-185W @ 43.5MH/s).

 

[Vya Domus]

TSMC 7nm is for Arm processors only, keep dreaming. 7nm GPU in 2019.

There is no such thing as a node meant for a particular architecture. A node is simply a manufacturing process , if it's optimized in such a way that some chips benefit more than a others it's a completely different matter. Use a better choice of wording , TSMC's 7nm node is not "only" for ARM processors , ARM processors are simply meant to be the first ones to use it.

 

[Rehmanpa]

I hope I don't sound completely stupid, but what's wrong with using hmb and hbm 2.0 ram? Isn't it significantly faster and is more power efficient than normal gddr5 ram, plus its more space efficient due to stacking abilities?

 

[efikkan]

I hope I don't sound completely stupid, but what's wrong with using hmb and hbm 2.0 ram? Isn't it significantly faster and is more power efficient than normal gddr5 ram, plus its more space efficient due to stacking abilities?

-- Price
-- Supplies - It delayed the RX Vega launch by three months. How can they get enough for even more products?
-- Assembly issues

 

[Octopuss]

Could you stop using Vega 10 and Vega 11 names please? It's confusing.

 

[efikkan]

Could you stop using Vega 10 and Vega 11 names please? It's confusing.

Why? It's the code names of the chips decided by AMD.

 

[Octopuss]

Why? It's the code names of the chips decided by AMD.

https://en.wikipedia.org/wiki/AMD_RX_Vega_series
Do you see any Vega 10 in there?

 

[Vya Domus]

https://en.wikipedia.org/wiki/AMD_RX_Vega_series
Do you see any Vega 10 in there?

Could you stop using Vega 10 and Vega 11 names please? It's confusing.

Currently there is just Vega 10 , Vega 11 is the supposed future RX 580 replacement. Don't know why it is so confusing.

 

[Octopuss]

Because there is Vega 64 and 56, not 10 and 11.

 

[efikkan]

Because there is Vega 64 and 56, not 10 and 11.

The chips are called Vega 10, Vega 11 and Vega 20, similarly to Nvidia's GP100, GP102, GP104, etc.
The products are called RX Vega 64, RX Vega 56, Vega Frontier Edition, etc.
You can blame AMD for the confusion here.
Edit: See https://community.amd.com/thread/218613

 

[Vya Domus]

Because there is Vega 64 and 56, not 10 and 11.

Uhm , who even said that Vega 64 is Vega 10 and Vega 56 is Vega 11 ?

Both Vega 64 and 56 are Vega 10 , Vega 11 does not exist yet. I think this is a case of people just simply not paying enough attention , things have been clear as daylight.

Both AMD and Nvidia are often manufacturing a single GPU and releasing different variants based on it. This has been happening for ages but for some reason now it sparks confusion.

 

[Octopuss]

Uhm , who even said that Vega 64 is Vega 10 and Vega 56 is Vega 11 ?

Both Vega 64 and 56 are Vega 10 , Vega 11 does not exist yet.

WTF?
I give up.
I guess you have to be this geek or nerd or whatever thing to understand this shit. I have better things to do in my life than memorizing cryptic and confusing hardware names.

 

[Vya Domus]

I give up too , no idea how this ended up being so hard to figure out. I suppose this goes all the way back to 1 year ago when there were leaks about Vega 10 and 11 and people just sort of assumed that Vega 64 and 56 correspond to that.