Wednesday, January 16th 2019

AMD's Initial Production Run of Radeon VII Just 5,000 Pieces, Company Denies it

More news coming in on AMD's upcoming high-end graphics card, the Radeon VII, with Chinese media reporting that AMD's initial production run for the card is set to ship just 5,000 pieces worldwide. This comes hot on the heels of another report that the Radeon VII won't come in custom-designs by AMD's add-in board (AIB) partners, and that only the reference design will be repackaged and sold by them. What's worse, the source which leaked this production size also revealed that AMD is selling the card below cost-price, i.e., with each card sold, AMD is losing money. This probably explains Wall Street's cold response to the Radeon VII launch, but with a batch size of just 5,000 (roughly $3.5 million in sales at $699 a piece), this card has a negligible impact on AMD's bottom-line.

AMD posted a swift denial to both pieces of news, the size of its production run and the product's profitability. In a statement to MyDrivers, AMD said (translated): "We will not release production figures, but when released on February 7, AMD.com official website and AIB vendor partners will have products on sale, and we expect the supply of Radeon VII to meet the needs of gamers." In short, Radeon VII is shaping up to be the card you'd want to buy if you've sworn a blood-oath never to buy an NVIDIA product, and you need something to play games in 2019 at 4K with.
Sources: MyDrivers, MyDrivers (2)
Add your own comment

124 Comments on AMD's Initial Production Run of Radeon VII Just 5,000 Pieces, Company Denies it

#3
HD64G
No way any GPU sold at over $500 is causing any money loss, apart for an ultra big die, which surely isn't the case here. If Vega 56 was sold at $400-450 back at the time of its launch and before the crypto craziness bulit up, this one which is also a cut version of the full Vega 20, will be surely profitable. And I think that AMD has put the price up there in order to be able to lower it IF nVidia lowers tha 2080's price. I am positive that they can sell it even at $600 at profit. 16GB HBM2 can cost up to $200 max (probaly closer to $150) and the die is relatively small at 330mm2. A small batch at first (not only 5000 units though) might exist just to not be found with a big stock in their hands in 1-2 months if the gaming community isn't fond of it after the reviews. They need to be precautious after all as the small player in this market, especially for an expensive product.
Posted on Reply
#4
notb
Fairly common practice - product released for shareholders, not clients. Yes, we can deliver 7nm GPUs. Sit tight and wait till we make it profitable (well... until TSMC does that).

Also, AMD will be able to use a "first consumer 7nm GPU" headline.
Looking at all the R&D and manufacturing struggle, 7nm could be the last silicon node for mainstream GPUs ever.

HD64G, post: 3976912, member: 95052"
No way any GPU sold at over $500 is causing any money loss
And this statement is based on what...? :-)

AMD marketing strategy lately is all about being first to 7nm in CPUs and GPUs (after they took the lead in core count). Having such a limited, flagship product makes a lot of sense for AMD, so TSMC could have asked any price they wanted.

Even if this whole move costs them $3.5mln (as much as the revenue), it's still less than 30s commercial during superbowl.
Posted on Reply
#5
Gasaraki
HD64G, post: 3976912, member: 95052"
No way any GPU sold at over $500 is causing any money loss, apart for an ultra big die, which surely isn't the case here. If Vega 56 was sold at $400-450 back at the time of its launch and before the crypto craziness bulit up, this one which is also a cut version of the full Vega 20, will be surely profitable. And I think that AMD has put the price up there in order to be able to lower it IF nVidia lowers tha 2080's price. I am positive that they can sell it even at $600 at profit. 16GB HBM2 can cost up to $200 max (probaly closer to $150) and the die is relatively small at 330mm2. A small batch at first (not only 5000 units though) might exist just to not be found with a big stock in their hands in 1-2 months if the gaming community isn't fond of it after the reviews. They need to be precautious after all as the small player in this market, especially for an expensive product.
Dude. Don't say stuff you are not 100% sure about. It costs around $750 to make the Radeon7/Vega 20/Instinct card.
Posted on Reply
#6
FordGT90Concept
"I go fast!1!11!1!"
There is absolutely no way Radeon VII is priced below cost. They wouldn't bring it to market if that were the case (would have kept selling them as MI50).

I get the distinct impression from the OP that there's a smear campaign in the works. All of these things said target lowering AMD's stock valuation.
Posted on Reply
#7
EarthDog
I mean, why research and get the facts out with one article when one posts click bait and then a follow up... :(
Posted on Reply
#9
moproblems99
Gasaraki, post: 3976923, member: 168493"
Dude. Don't say stuff you are not 100% sure about. It costs around $750 to make the Radeon7/Vega 20/Instinct card.
How do you know it costs $750 to make with 100% certainty?
Posted on Reply
#10
Casecutter
Chinese media = Nvidia dis-information.

How many Radeon MI60 Instinct Professional units would we think AMD produce/sells? How many chips on a 7nm wafer? How many on a wafer are just bad can't be salvage for anything? Then what's the mix of full MI60 (4096SP), MI50/Vega7 (3840SP), and I might say there possibly could be a 56CU/3584SP part being binned. That could point to how many parts AMD would have to play with.

As to price the existing Vega 64 has been down to $400... what's different? The 7nm wafer TSMC is charging more, although you get more per wafer, also remember the V64 was a full chip. This is a gelding so per chip could be slightly more, but I could say similar? Cost for 8Gb "more" HBM2, I might say $80. The interposer and assembly is same if not getting better so that's similar. There is plenty of money left in the BOM to provide profit, that's bunk!

Lastly, AIB's... We don't think existing Vega 64 design don't easily "port-over"? Everything is on the interposer, so the power section might need nothing as isn't the MI50 300 TDP while I thought Vega 7 isn't suppose to be that high. AIB's have existing PCB and coolers that should not have an issue flipping from their existing Vega 64 customs to this interposer that in all probability a "pin-for-pin" drop-in interposer and nothing more.

Logically this Chinese media report is not based on factual consideration.
Posted on Reply
#11
Kaotik
In short, Radeon VII is shaping up to be the card you'd want to buy if you've sworn a blood-oath never to buy an NVIDIA product, and you need something to play games in 2019 at 4K with.
Seriously @btarunr ? How biased can you be? You're actually blatantly saying on TPU frontpage that only reason anyone would buy Radeon VII is some oath never to buy NVIDIA?
Posted on Reply
#12
Renald
Kaotik, post: 3976951, member: 101367"
Seriously @btarunr ? How biased can you be? You're actually blatantly saying on TPU frontpage that only reason anyone would buy Radeon VII is some oath never to buy NVIDIA?
Even if his quote is brutal an non-professional, I don't think he's wrong. I defended AMD all these years for what they were for : good performance/$ with enough performance to play.

But let's face it, the Radeon VII is nothing more that an attempt to fill the high-end user case.
And for that, there's the 2060, and nobody can deny it's performance on actual games. Three weeks ago I thought this card (2060) would be a very slow version of RTX cards, but it's not the case and I revised my point of view.

Slight problem is Win 10 with Fall update... Quite hard to in.
Posted on Reply
#13
20mmrain
Kaotik, post: 3976951, member: 101367"
Seriously @btarunr ? How biased can you be? You're actually blatantly saying on TPU frontpage that only reason anyone would buy Radeon VII is some oath never to buy NVIDIA?
I really don't see how this is a blast on AMD or NVidia. Just saying. Someone's a little easily offended. It's just a video card.
Posted on Reply
#14
unikin
The truth is AMD hasn't developed high end GPU for gamers, period. They have some Instinct MI50 GPUs with a defective compute unit or two and decided to push them to AMD gaming fan boys for free advertising purposes. R7 will sadly be collectors edition GPU and nothing more.
Posted on Reply
#15
notb
Casecutter, post: 3976948, member: 94772"
How many Radeon MI60 Instinct Professional units would we think AMD produce/sells?
Zero?
Posted on Reply
#16
xkm1948
I wonder which TPU member will be the first to unbox and bench one of these, well minus W1zzard of course.
Posted on Reply
#17
Bones
Kaotik, post: 3976951, member: 101367"
Seriously @btarunr ? How biased can you be? You're actually blatantly saying on TPU frontpage that only reason anyone would buy Radeon VII is some oath never to buy NVIDIA?
I'm not so sure it's actual bias, more like a certain unfortunate truth stated and yes you do have folks like that out there - Same with AMD vs Intel, we've seen more than enough biased comments here and elsewhere to prove the truth of that statement.

No real bias here, I have both brands onhand for my use and just grabbed a GTX 780 for my DD, not the newest by any means but enough it serves the needs I have. Also have a R9 380X I run sometimes, it's a good card too so in truth either one can and will do for me. Only difference is what would run what I'm currently wanting to do best and that's it.

Same with AMD vs Intel, currently running a 7350K in this setup and before that it was a FX-8300, what I have runs along the fence and honestly I haven't seen enough difference to make me want to prefer one over the other, be it a card or CPU.

If it works and does the job to make you happy, go for it.
I do.
Posted on Reply
#18
Ferrum Master
The main question is. Can you even order pancackes at TSMC at such amount. Considering how complex it is to set it up?

I call this bollox.
Posted on Reply
#19
Casecutter
Knowing Apple's lackluster iPhone/iPad sales numbers, I wonder if TSMC could have some 7nm Wafer capacity that underutilized?

While "The data center accelerator market is expected to grow from 2.84 billion by 2018 to USD 21.19 billion by 2023"
https://www.prnewswire.com/news-releases/global-data-center-accelerator-market-2018-2023---focus-toward-parallel-computing-in-ai-data-centers-300705003.html


Some would want to think AMD could do perhaps 500,000 GPUs to Data Center and Deep Learning HPC over the next 2-3 years I personally think that's low-ball.

https://seekingalpha.com/article/4103435-china-catalyst-amds-radeon-instinct-ai-processors
Posted on Reply
#20
theoneandonlymrk
Casecutter, post: 3976948, member: 94772"
Chinese media = Nvidia dis-information.

How many Radeon MI60 Instinct Professional units would we think AMD produce/sells? How many chips on a 7nm wafer? How many on a wafer are just bad can't be salvage for anything? Then what's the mix of full MI60 (4096SP), MI50/Vega7 (3840SP), and I might say there possibly could be a 56CU/3584SP part being binned. That could point to how many parts AMD would have to play with.

As to price the existing Vega 64 has been down to $400... what's different? The 7nm wafer TSMC is charging more, although you get more per wafer, also remember the V64 was a full chip. This is a gelding so per chip could be slightly more, but I could say similar? Cost for 8Gb "more" HBM2, I might say $80. The interposer and assembly is same if not getting better so that's similar. There is plenty of money left in the BOM to provide profit, that's bunk!

Lastly, AIB's... We don't think existing Vega 64 design don't easily "port-over"? Everything is on the interposer, so the power section might need nothing as isn't the MI50 300 TDP while I thought Vega 7 isn't suppose to be that high. AIB's have existing PCB and coolers that should not have an issue flipping from their existing Vega 64 customs to this interposer that in all probability a "pin-for-pin" drop-in interposer and nothing more.

Logically this Chinese media report is not based on factual consideration.
Agreed , Hype and drama for what.
Kaotik, post: 3976951, member: 101367"
Seriously @btarunr ? How biased can you be? You're actually blatantly saying on TPU frontpage that only reason anyone would buy Radeon VII is some oath never to buy NVIDIA?
A bit like the one before with its salty BS, hype drama and certainly ,I remember TPU's covergage of Nvidias Founders edition announcment with no such saltiness for what was the same thing, hype drama, no need their will be enough drama and hype on TPU Radeon 7 review day, Also if this were about Nvidia the site might find itself without review samples like some neg V bloggers did recently with its predetermined attitude.


ALSO WORTH NOTING A LOT ON HERE WERE VERY HAPPY WITH 10% improvement for years even lauding its greatness ,vega 7 is 25%+ better then vega64 i have one , it games at 1680Mhz for no more then 240 watts but typically around 180 watts @4k,, Vega 7 runs Upto 300 watts , i very much doubt it runs that high in any game and its priced about the typical 2080 price point here in the uk with 2080's going for between 650 and 800 dependant on variant, of which there are two , one a bit shitter then the other in silicon terms, that alone gets AMD the sale for me , they are not being deceptive, but im not buying tbh waterblocked cards need using for longer to recoup costs, shame.
Posted on Reply
#21
Mistral
btarunr"
AMD's Initial Production Run of Radeon VII Just 5,000 Pieces, Company Denies it
Do we no longer care how language works?
Posted on Reply
#22
Vya Domus
Ferrum Master, post: 3977035, member: 90058"
The main question is. Can you even order pancackes at TSMC at such amount. Considering how complex it is to set it up?

I call this bollox.
AMD gets wafers with Vega 20s on them in X quantities, TSMC doesn't care nor do they have any control over what happens with them after that. These 5000 units or whatever wont change anything with regards to how much TSMC charges AMD.
Posted on Reply
#23
Casecutter
I suppose the 5000 unit allocation has some merit. As that would mean AMD figures initially gifting 1% of production (if looking at the 500,000 above) as Vega 7 at $700, instead of the $5000 they can probably get as a Instinct MI50. I suppose if FP64 was some "flaw" that they use to "bin" a Vega 7 but doubt that and why only 1% initially. For Vega 7 it's probably more about meeting power envelope, and less on shader count. So does that mean their 7nm production has little to no terrible chips once actual yield of unusable/bad parts are weeded out?


theoneandonlymrk, post: 3977059, member: 82332"
Agreed , Hype and drama for what
Jensen Huang cronies still looking to bad mouth.
Posted on Reply
#24
erocker
*
Gasaraki, post: 3976923, member: 168493"
Dude. Don't say stuff you are not 100% sure about. It costs around $750 to make the Radeon7/Vega 20/Instinct card.
That $750 number is made up as well. So.....
Posted on Reply
#25
Kaotik
20mmrain, post: 3976985, member: 76476"
I really don't see how this is a blast on AMD or NVidia. Just saying. Someone's a little easily offended. It's just a video card.
Yes, as part of the media I'm easily offended when I find media outlets that publish so blatantly biased texts. Media is the one that's supposed to stay neutral.

Bones, post: 3977027, member: 144474"
I'm not so sure it's actual bias, more like a certain unfortunate truth stated and yes you do have folks like that out there - Same with AMD vs Intel, we've seen more than enough biased comments here and elsewhere to prove the truth of that statement.
Of course there are always folks like that, but that doesn't change the fact that there's plenty of other reasons to buy Radeon VII even as it stands now - announced but not released and half the specs unclear when one AMD rep is saying one thing and the next something else.
Biased comments is one thing, I don't care about those, biased news is something else which I do care about (as should everyone else, media is the one that's supposed to stay neutral (and yes, I know I repeat that in this message, but you can never repeat it enough))
Posted on Reply
Add your own comment