Digitimes Says AMD RX Vega Card Shortages Will Persist

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,532
We all know that there has been greater demand for AMD RX Vega 64 than there has been supply for. There is not arguing that. Speculating on why can lead you in a few different directions. However, Digitimes is reporting that the supply issue is due to low packaging yield rates, according to sources from the upstream supply chain. This has been very much different from the information we have gotten here at HardOCP in the last 7 days. Our information about initial shortages have pointed to AIB lag in the channel as we have been expected to see more Vega 64 inventories back in the channel at MSPR in the next week. We will see what happens, and we have reached out to AMD for direct comment on what Digitimes has reported.

Some sources pointed out that Vega's design of integrating high bandwidth memory (HBM) into the GPU has significantly raised the difficulty of the related packaging process and hence resulted in low yield rates. However, some other sources claimed that the issue may be down to problems with Advanced Semiconductor Engineering's (ASE) packaging technology.
 
This is why we can't have nice things at affordable prices. So much speculation and demand of these things among the mining community. The price was awesome to start but its not at the level of 1080, which sucks because the bang for the buck was there and now its kinda not so much anymore....
 
We've seem the samples of the dies and it appears the Taiwan packager didn't do epoxy and had an uneven GPU/ HBM2 die profile. This is what has been suggested as the issue. Samsung memory chip dies were underfilled with epoxy and even die/memory profile.
 
The person who signed off on using HBM should be fired. The performance there doesn't warrant the use of HBM. Vega will probably cost AMD money in the end with the pile of bad chips they got.
 
HBM is AMD technology. Yields aren't there yet on HBM2. NVidia Volta modules use HBM2 also
 
This is why we can't have nice things at affordable prices. So much speculation and demand of these things among the mining community. The price was awesome to start but its not at the level of 1080, which sucks because the bang for the buck was there and now its kinda not so much anymore....
Rumor is that most of the cards are being used for the data centres/rendering farms due to the higher compute rates.
 
It boggles the mind how badly AMD screwed up the Vega launch. I really don't know why anyone would want to buy one of these cards. They aren't even that great at mining. Polaris is better for the cost.

It's sad really, because AMD absolutely nailed it with Ryzen and Threadripper.
 
HBM is AMD technology. Yields aren't there yet on HBM2. NVidia Volta modules use HBM2 also

Nvidia uses and will continue to use HBM (which isn't 'AMD technology' as much as 'AMD jumped on for marketing') where it makes sense for the product. AMD is using it as a catchword for marketing copy.

Fun part: while using traditional GDDR would lower production costs and (apparently!) increase final product yields, it would also increase power draw and heat output.
 
Rumor is that most of the cards are being used for the data centres/rendering farms due to the higher compute rates.
Its not really rumor anymore its pretty much fact at this point. Data centers are buying them up by the truck loads....
 
It boggles the mind how badly AMD screwed up the Vega launch. I really don't know why anyone would want to buy one of these cards. They aren't even that great at mining. Polaris is better for the cost.

It's sad really, because AMD absolutely nailed it with Ryzen and Threadripper.
Shortage is a good problem to have, so ill say they are okay.
 
Nvidia uses and will continue to use HBM (which isn't 'AMD technology' as much as 'AMD jumped on for marketing') where it makes sense for the product. AMD is using it as a catchword for marketing copy.

Fun part: while using traditional GDDR would lower production costs and (apparently!) increase final product yields, it would also increase power draw and heat output.

Yeah but the heat would have been distributed and not huddled next to the die... making Vega even harder to cool.
 
http://www.amd.com/Documents/High-Bandwidth-Memory-HBM.pdf <-- they developed it. amd pushed for it being made a jedec standard for everyone to use not just them. if nvidia developed it 100% it would of been proprietary...
And there would be nothing wrong with that either. AMD did not push it out of the goodness of their heart. There are strategic business reasons. When you are the company so far behind getting people to adopt your technology, even if you don't get direct monetary compensation for it, can benefit the sales of your hardware. In AMD eyes the more people optimizing to take advantage of HBM the better positioned they are.
 
Nvidia uses and will continue to use HBM (which isn't 'AMD technology' as much as 'AMD jumped on for marketing') where it makes sense for the product. AMD is using it as a catchword for marketing copy.

Fun part: while using traditional GDDR would lower production costs and (apparently!) increase final product yields, it would also increase power draw and heat output.

According to Kitguru, AMD owns some patents regarding HBM, but its been open standard. If this is true, I wouldn't call it just "marketing copy". Seems pretty likely to me that the statement AMD and SK Hynix and others co-developed it is true.
 
AMD is happy just to recoup costs and get out from under the Arctic Islands architecture.

Now if Navi bombs, there are problems because that is 100% on Raja.

But Polaris and Vega are RTG making lemonade from lemons
 
Im not sure that vega should be called a failure, since amd is selling each one that they can build.

Sucks for gamers, but as a company, they dont have the problem of worrying about unsold stock.
 
Im not sure that vega should be called a failure since amd is selling each one that they can build.

Sucks for gamers, but as a company, they don't have the problem of worrying about unsold stock.
It is easy to sell out if it is a small volume of cards. We don't really have a good picture of the actual volume of sales, so it is sold out literally tells us nothing other than it is sold out.
 
Im not sure that vega should be called a failure, since amd is selling each one that they can build.

Sucks for gamers, but as a company, they dont have the problem of worrying about unsold stock.
They have a bigger problem - turning away potential sales and sending those buyers next door to Nvidia. And once once they run out of fanboys and launch euphoria, they're in trouble.

Nvidia doesn't have that problem with new cards sold out at launch for a few weeks, because people will wait for them to come back into stock, since they know there's nothing better on offer or coming soon from AMD.
 
They have a bigger problem - turning away potential sales and sending those buyers next door to Nvidia. And once once they run out of fanboys and launch euphoria, they're in trouble.

lol funny it's to see most of the die-hard AMD fans looking options to buy 1080Ti what's worse even looking for 1080 and even worse yet, the fact that some of those are willing to spend top dollar buying TitanXp due the delay of Nvidia with Volta.
 
This is why we can't have nice things at affordable prices. So much speculation and demand of these things among the mining community. The price was awesome to start but its not at the level of 1080, which sucks because the bang for the buck was there and now its kinda not so much anymore....
RX VEGA 56 actually beats the 1080 Ti at Blender so it's not just mining the demand and RX Nano likely will as well once it launches so that's a pretty big deal if that's a core usage or interest for people beyond gaming. It's funny that people seem to care so much about the issue yet on the same hand also act standoffish like they don't want RX Vega anyway I mean that's a bit irritating.
 
According to Kitguru, AMD owns some patents regarding HBM, but its been open standard. If this is true, I wouldn't call it just "marketing copy". Seems pretty likely to me that the statement AMD and SK Hynix and others co-developed it is true.

Oh, I won't deny that AMD did some development work to push the tech forward, and obviously secured some IP in the process, but my point is that they did all of that and then went through the trouble of designing HBM into and using it in cards whose performance simply doesn't justify the cost.
 
*Shrug* Eventually, they will come back down in price, this holiday season could be good one for technology.
 
Oh, I won't deny that AMD did some development work to push the tech forward, and obviously secured some IP in the process, but my point is that they did all of that and then went through the trouble of designing HBM into and using it in cards whose performance simply doesn't justify the cost.

You worded that pretty badly.
 
Fun part: while using traditional GDDR would lower production costs and (apparently!) increase final product yields, it would also increase power draw and heat output.

HBM2 using less power than GDDR5X/GDDR6 is a myth. Even Hynix says so :)
 
Its not really rumor anymore its pretty much fact at this point. Data centers are buying them up by the truck loads....

Are they? I doubt. Its not a mining card, its not a gaming card, its not an AI card. And compute is lacking big time too.
 
http://www.amd.com/Documents/High-Bandwidth-Memory-HBM.pdf <-- they developed it. amd pushed for it being made a jedec standard for everyone to use not just them. if nvidia developed it 100% it would of been proprietary...

Its about equal to claiming Nvidia developed GDDR5X and GDDR6. The only part AMD did was to accept being a customer and then play the PR game. It was Hynix that developed it. Development began in 2011. AMD first "joined" in late 2013 for a mid 2015 Fiji release.

But hey, AMD developed DX12 too or something, right? ;)
 
Last edited:
It was known long ago that Vega would be in short supply. And as pricing also shows, its a cost issue more than anything else. HBM2 devices are plentiful in P100 and FPGAs.

http://www.tweaktown.com/news/57418/amd-radeon-rx-vega-less-20-000-available-launch/index.html

On the other hand there is hardly any buyers willing to buy it. So 20000 cards will go a real long way. Vega is still in plentiful supply at most of the retailers that actually got the cards.

https://www.mindfactory.de/Hardware/Grafikkarten+(VGA)/Radeon+RX+VEGA.html
 
AMD is happy just to recoup costs and get out from under the Arctic Islands architecture.

Now if Navi bombs, there are problems because that is 100% on Raja.

But Polaris and Vega are RTG making lemonade from lemons

People need to stop the excuses that Navi will be a Raja design and act like he had nothing to do with everything before and is a single handed god. Also dont expect Navi to be anything than a shrinked and glorified Vega as the HPC roadmap suggest as well. 4096SP and 7nm perhaps with GDDR6. Vega 20 will be the big GPU of that timeframe.
 
Last edited:
I mean I get the frustration that nothing is list price.. Get that 100%, but AMD showings are good at their intended price target, apparently they do have are reasonable margin of profit as they can do rebates somehow, that is good too.
Hoping things will stabilize.
If nvidia had this card with performance level X that beats Vega, for a million years ago,
I dont think that is too important.. the great volta is far away, we know shit of how its going to do, and we don't really know if current Nvidia's architecture is tapped out in a meaningful way or not.
 
the great volta is far away, we know shit of how its going to do, and we don't really know if current Nvidia's architecture is tapped out in a meaningful way or not.
Not that far, and the performance is already somewhat known. Going to be a massive jump in performance.
 
It is easy to sell out if it is a small volume of cards. We don't really have a good picture of the actual volume of sales, so it is sold out literally tells us nothing other than it is sold out.

Doesnt matter, as i said, their main concern as a company is selling all the produced goods, instead of having them sitting in a warehouse.

They have a bigger problem - turning away potential sales and sending those buyers next door to Nvidia. And once once they run out of fanboys and launch euphoria, they're in trouble.
.

The only downside is as you said, losing the chance of selling because they don't have enough available. But as i said above and before, as a company, they should be happy that their products are all sold, instead of sitting in shelves and warehouses.

Yes, it does sucks not having more at the moment, but nobody seems to know how many they are producing on a daily basis.

Could be 10, could be a million, but in the end, they are selling faster than they can provide, so it is kind of good problem to have.
 
Doesnt matter, as i said, their main concern as a company is selling all the produced goods, instead of having them sitting in a warehouse.
No that is not how this works. Their main concern is getting a return on the investment and if they can't actually produce enough cards to break even then selling out means absolutely nothing to them other than not sitting on stock. Selling out is not necessarily an indication of success, There are lots of other factors at play.
 
GamersNexus had a great article explaining that while HBM2 costs $175, AMD had no choice but to use it because GDDR5 would require an additional 80-100W of power consumption for similar memory bandwidth. The Vega core wasn't power efficient enough to support that additional power draw. Imagine rather than 300W consumption at stock clocks the card consumed >400W.

http://www.gamersnexus.net/guides/3032-vega-56-cost-of-hbm2-and-necessity-to-use-it

On the other hand, given that GDDR5 is 3x cheaper than HBM2 ($50-60 vs $175), AMD is in a bit of a pickle. About half the price of the RX 56 for example is going to the memory makers. Given the cost of the other components, and the yield issues that were brought up in the Digitimes article, people are speculating that AMD is currently making little to no money on these cards.
 
GamersNexus had a great article explaining that while HBM2 costs $175, AMD had no choice but to use it because GDDR5 would require an additional 80-100W of power consumption for similar memory bandwidth. The Vega core wasn't power efficient enough to support that additional power draw. Imagine rather than 300W consumption at stock clocks the card consumed >400W.

http://www.gamersnexus.net/guides/3032-vega-56-cost-of-hbm2-and-necessity-to-use-it

On the other hand, given that GDDR5 is 3x cheaper than HBM2 ($50-60 vs $175), AMD is in a bit of a pickle. About half the price of the RX 56 for example is going to the memory makers. Given the cost of the other components, and the yield issues that were brought up in the Digitimes article, people are speculating that AMD is currently making little to no money on these cards.

GDDR5X/GDDR6 is as good as HBM2 in power consumption. GP102 also got more bandwidth than any Vega card. (547.7GB/sec).

The only reason why anyone uses HBM2 is ECC and perhaps size. (Vega isn't using it for size). And no, HBCC also works on GDDR cards like Polaris ;)

Using GDDR5 as the compare metric when GDDR5X have been out for 15 month+ is just silly.
 
Back
Top