AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

[Deleted member 13524]

The "Vega V" logo has been used in Radeon Instinct technical marketing at least, so new logo doesn't necessarily mean anything regarding any consumer products

Well the Mi50 and Mi60 were already formally announced, so I doubt this new logo was registered for those.
Plus, AMD showed the Vega logo for the 7nm Vega 20 when they talked about it in June at Computex, but I can't find the Vega logo in any PR material of the formal announcement they made last month. The Instinct line seems to have been separated from the Vega moniker.

Are you suggesting there will be other non-consumer products carrying the Vega 20 chip besides the Instinct cards? Radeon Pro and Frontier Edition?
It could be possible, but I don't remember a single time when AMD ever kept a chip from the consumer market.
Unlike nvidia, they can't really afford to make super-specialized chips. And even nvidia only ever did that with GP100, as far as I remember. All others found their way into Geforces and/or Titans.
The GP102 was faster for gaming workloads anyway, as it provides ~20% higher FP32 throughput. 1:2 FP64 and 2:1 FP16 proved to be relatively useless for Pascal's lifetime in gaming.
It's easy to see why the GP100 never went into a consumer or prosumer (Titan) product.


With AMD being absent from the ultra-high / top-end range for so long, the only reasons I could see for them to not have Vega 20 in consumer products are if:
1 - They can't make enough chips (doubtful with TSMC announcing they won't be using the 7nmn fabs at full capacity throughout 2019), or
2 - Vega 20 isn't competitive with the TU104 / GP102 in gaming workloads so they'd have to price such cards below 700€.
3 - AMD doesn't want to spend marketing money to push another Vega into the market with Navi being relatively close.

 

[entity279]

with TSMC announcing they won't be using the 7nmn fabs at full capacity throughout 2019

sorry, where's that from?
The reason for it would be interesting. Out of context, I'd read this as "we're not ready to max out or 7nm node" rather . Doesn't mean necessarly that there's spare capacity at non-absurd prices and high enoungh yields

 

[Tofu]

sorry, where's that from?
The reason for it would be interesting. Out of context, I'd read this as "we're not ready to max out or 7nm node" rather . Doesn't mean necessarly that there's spare capacity at non-absurd prices and high enoungh yields

https://www.tomshw.de/2018/12/07/tsmcs-7nm-fertigung-nicht-ausgelastet-amd-wird-es-freuen-freuen/

They refer to Digitimes (paywall) stating that for 2019 TSMC will only use 80-90% of their 7nm capacity. This is because Apple, HiSilicon and Qualcomm reduced the number of ordered chips.

 

[Deleted member 13524]

sorry, where's that from?
The reason for it would be interesting. Out of context, I'd read this as "we're not ready to max out or 7nm node" rather . Doesn't mean necessarly that there's spare capacity at non-absurd prices and high enoungh yields

It comes from a digitimes article behind a paywall:

https://www.digitimes.com/news/a20181205VL201.html]

There's an extremetech article about it:

https://www.extremetech.com/mobile/281816-tsmcs-7nm-node-supposedly-running-below-capacity

The reason isn't "TSMC isn't ready for more", but rather that their clients cut back on the orders:

Supposedly Apple, HiSilicon, and Qualcomm have all cut their orders for the first half of the year. It’s not entirely clear why this would be the case.

Given the reports about 7nm not hitting the performance estimates, that could be part of the reason.

 

[Deleted member 13524]

So this news is a couple of days old but it didn't find its way here:

https://www.phoronix.com/scan.php?page=news_item&px=AMD-New-Vega-10-20-PCI-IDs

Last friday's Mesa drivers added support for a new Vega 20 PCI ID, adding to a pre-existing 5 (total of 6 at the moment).
It also added 6 new Vega 10 PCI IDs.

In the meanwhile, a "Radeon Pro Vega 48" was spotted in the PCMark 10 database.
I'm guessing at least some of these Vega 10 variants are coming to a new Mac Pro to be announced soon. After all, the garbage-bin model hasn't been updated in quite a while (it uses Ivy Bridge + DDR3 and Tahiti graphics cards!).
Maybe there will be Vega 20 cards in there, too.

 

[del42sa]

In the meanwhile, a "Radeon Pro Vega 48" was spotted in the PCMark 10 database.
I'm guessing at least some of these Vega 10 variants are coming to a new Mac Pro to be announced soon. After all, the garbage-bin model hasn't been updated in quite a while (it uses Ivy Bridge + DDR3 and Tahiti graphics cards!).
Maybe there will be Vega 20 cards in there, too.

it says GFX901, which is Vega 10

7nm Vega has GFX 906 and GFX 907

 

[Kaotik]

The new "Vega 10" aka "Vega 1x" models are most likely for the Vega 12(?) in MacBook Pro's at the moment, since they've already been spotted in macOS Mojave update

 

[del42sa]

The new "Vega 10" aka "Vega 1x" models are most likely for the Vega 12(?) in MacBook Pro's at the moment, since they've already been spotted in macOS Mojave update

AFAIK Vega 12 (Vega Mobile) is GFX904 not GFX901

 

[Kaotik]

AFAIK Vega 12 (Vega Mobile) is GFX904 not GFX901

Could be, manflu is taking it's toll on the brain. There was an odd Radeon Pro Vega 48 spotted somewhere, maybe upcoming Mac Pro GPUs then?

 

[Deleted member 13524]

it says GFX901, which is Vega 10

Yes.
I didn't imply otherwise.

Could be, manflu is taking it's toll on the brain. There was an odd Radeon Pro Vega 48 spotted somewhere, maybe upcoming Mac Pro GPUs then?

Flu indeed.

In the meanwhile, a "Radeon Pro Vega 48" was spotted in the PCMark 10 database.
I'm guessing at least some of these Vega 10 variants are coming to a new Mac Pro to be announced soon.

 

[Geeforcer]

I wonder how crypto collapse and what I can only imagine is a huge decline in BitMain delivers is impacting TSMC 7nm orderbook.

 

[del42sa]

Could be, manflu is taking it's toll on the brain. There was an odd Radeon Pro Vega 48 spotted somewhere, maybe upcoming Mac Pro GPUs then?

Yes it could, or it could be 12nm Vega10 refresh from GlobalFoundries, otherwise what´s the point of release 6 "new" PCI IDs of old VEGA 10 in 2019 ?

https://twitter.com/x/status/1072048805997105152

 

[Kaotik]

Yes it could, or it could be 12nm Vega10 refresh from GlobalFoundries, otherwise what´s the point of release 6 "new" PCI IDs of old VEGA 10 in 2019 ?

Unlikely, why do a 12nm shrink with very limited gains when you've already done 7nm "shrink"?
A, B and F could just be Radeon RX versions of the Radeon Pros or Radeon Pros for Win/Linux if they have separate ID's compared to Apple ones

Flu indeed.

Manflu. Completely different things, if you compare the severity, manflu is to flu like lung cancer is to cough (I really wish you'd understand finnish, I'd have perfect video for this, but can't find even a subtitled version of it)

 

[Deleted member 13524]

Manflu. Completely different things, if you compare the severity, manflu is to flu like lung cancer is to cough (I really wish you'd understand finnish, I'd have perfect video for this, but can't find even a subtitled version of it)

Dude I'm a man and I've had the flu before...
In fact I was infected with Influenza A when I was in Mexico (first time I ever seriously thought I was going to die from the flu).

 

[del42sa]

Unlikely, why do a 12nm shrink with very limited gains when you've already done 7nm "shrink"?

yes, so was Radeon RX590.... Why ? because of WSA

 

[no-X]

Polaris 30 is not a good example for several reasons. 7nm shrink wouldn't be possible, because the resulting die would be too small to fit 256bit bus (128bit bus would make it even slower than Polaris 10/20). Also, WSA is not an issue. Why would AMD manufacture Polaris 30 at Samsung, if it needed to fulfill contracts with GlobalFoundries?

 

[del42sa]

Polaris 30 is not a good example for several reasons. 7nm shrink wouldn't be possible, because the resulting die would be too small to fit 256bit bus (128bit bus would make it even slower than Polaris 10/20). Also, WSA is not an issue. Why would AMD manufacture Polaris 30 at Samsung, if it needed to fulfill contracts with GlobalFoundries?

I don´t know exactly the current state of WSA between AMD and GloFo, but I firmly believe it still stands valid. About Samsung being manufacturing Polaris 30, do we know any volume numbers ? From my point of view AMD divide production to another (backup) source. Whats the point to manufacture something on older 14nm node if they offer actually better 12nm variant ?

 

[no-X]

The question is: Do they even manufacture Vega 10? Because it seems, that they have excessive stock of them.

 

[Kaotik]

The question is: Do they even manufacture Vega 10? Because it seems, that they have excessive stock of them.

What exactly makes it look like that?

 

[Babel-17]

Unlikely, why do a 12nm shrink with very limited gains when you've already done 7nm "shrink"?
A, B and F could just be Radeon RX versions of the Radeon Pros or Radeon Pros for Win/Linux if they have separate ID's compared to Apple ones


Manflu. Completely different things, if you compare the severity, manflu is to flu like lung cancer is to cough (I really wish you'd understand finnish, I'd have perfect video for this, but can't find even a subtitled version of it)

You inspired me to Ecosia that. http://manflu.info/index.htm Possibly because because I've been going through the nose wipes this morning. At least now I know the proper treatment! http://manflu.info/self_help/self_help.htm