TR Forums

While the whole OMG MINING aspect on the demand side for Rx Vega has been discussed to death, it's also pretty obvious that even long after Vega's initial release AMD is still having problems manufacturing the GPUs in sufficient quantity to actually let the board makers produce custom Vega boards:

https://videocardz.com/72712/amd-partne ... ir-designs

So it looks like there are both demand and supply issues affecting Vega even right now.

You do understand the relationship between supply and demand, right? Vega's HBM2 supply issue has been rehashed repeatedly since (and even before) Vega's launch.

The vast majority of the issues with Vega (delays, pricing fiasco, even lackluster performance due to lower-than-expected memory speed/bandwidth, and now OEM delays for AIB cards) have all been due to Hynix coming up short on their HBM2 timeline. The cryptocurrency demand, while significant, is a secondary factor. AMD partnered with Hynix to produce HBM2, and so far, production ramp up has been delayed, then delayed again, and then delayed some more. The result is that AMD had to scramble to get HBM2 from Samsung. (at least partially explaining multiple package designs, etc).

Though news that AMD's HBM2 supply issues are persisting is unfortunate.

cynan wrote:

You do understand the relationship between supply and demand, right? Vega's HBM2 supply issue has been rehashed repeatedly since (and even before) Vega's launch.

Though news that AMD's HBM2 supply issues are persisting is unfortunate.

The HBM issues were certainly a known cause of why Vega didn't launch sooner, but there was also some thought that the supply issues had been ironed out now that Vega is on the market. Although it turns out that may not be the case.

Is there any NEW information here? No where in the article does it say where the problem lies... Is it in the VEGA chip manufacturing? Is it in the supply (or lack thereof) of HBM2? Is it issues with the Chip Packaging process? Or Cost related? Probably a combination of several...

As the Wendy's commercials with the little old ladies from the 1980s said, "Where's the Beef?" to this article?
https://www.youtube.com/watch?v=riH5EsGcmTw

On another note, NEWS FLASH: Intel Delays 10nm Cannon Lake CPUs To End Of 2018
http://wccftech.com/intel-delays-10nm-c ... -end-2018/

vega's a big chip. big chips have lower yields.

this is nothing we didn't already know 6 months before the damn thing even launched. The only people who act surprised by all this are the crazy lunatics who somehow believe AMD can have complete control over store pricing and are blaming AMD for not flooding the market with so many GPUs that everything stays at MSRP no matter what...

Is there any NEW information here?

Yes there absolutely is: AMD said third-party board makers would have Vega parts in-hand by over a week ago and it never happened. Meaning that this is a real story about a real event, while you prefer to hide the severe issues at AMD that have likely cost Raj his job and want to chase after the usual suspects whining about Intel to generate clicks.

As for rumor articles whining about Intel, bear in mind that Intel was showing off 10nm wafers less than 24 hours before Wccftech needed to post more clickbait.

https://hexus.net/tech/news/industry/11 ... jing-show/

And remember that Coffee Lake, which is launching in just over 2 weeks and well ahead of quad-core APUs based on Ryzen wasn't supposed to launch until next year too.

And remember that Skylake X wasn't coming out until 2018 either https://www.reddit.com/r/intel/comments ... d_to_2018/

chuckula wrote:

Is there any NEW information here?

Yes there absolutely is: AMD said third-party board makers would have Vega parts in-hand by over a week ago and it never happened. Meaning that this is a real story about a real event, while you prefer to hide the severe issues at AMD that have likely cost Raj his job and want to chase after the usual suspects whining about Intel to generate clicks.

As for rumor articles whining about Intel, bear in mind that Intel was showing off 10nm wafers less than 24 hours before Wccftech needed to post more clickbait.

https://hexus.net/tech/news/industry/11 ... jing-show/

And remember that Coffee Lake, which is launching in just over 2 weeks and well ahead of quad-core APUs based on Ryzen wasn't supposed to launch until next year too.

And remember that Skylake X wasn't coming out until 2018 either https://www.reddit.com/r/intel/comments ... d_to_2018/

Well, 1st off, you stated AMD had manufacturing issues, where I don't see any proof of that in the article or anything you said either. The delay in OEMs getting hold of Vega modules (GPU & HBM2 combined on an interposer) can have numerous issues as I stated before none of which have been PROVEN to be a GPU MANUFACTURING ISSUE as you STATED. Maybe they do maybe they don't maybe it is with the building of the Interposer package or maybe Harvey or IRMA washed away all their packaging. Needless to say you can still buy an RX Vega 56 or 64 if you think the price is right, just not an OEM designed one. And I for one bought a RX Vega 56 with the game bundle for $499 because I thought it was a good deal, considering most GTX 1070s were going for $450+ when I bought it.

What severe issues at AMD??? Fudzilla was claiming AMD is losing money on early Vega cards (which maybe, I have no idea, but I wouldn't be surprised if it is nearly a break-even scenario on the RX Vega 56) but if that is true wouldn't it be "smart" on their part to "slow roll" the release of a money losing card? You are acting like this is a "Barcelona" delay or manufacturing problem, when I highly doubt it is... more likely just high cost of HBM2 and perhaps with 3rd party packaging of the GPU & HBM2 on the interposer. Overpriced is also subjective to the buyer is it not? I thought the 1080 & 1070 were over priced when 1st released, but the 1070 turned out to be a "steal" at $400 and under prior to the Crypto-craze driving up prices. Prices on the 1070s are finally coming down some (may go right back up due to GDDR5 memory price increases however) to where you can find them back under $450 and as low as $415 I've seen this week.

And these "severe issues"with AMD manufacturing (which technically is done by GF or TSMC) don't seem to have an affect on AMD producing Ryzen, ThreadRipper, EYPC or AMD making inroads on GPU compute sales to Amazon and now an AI SOC for TESLA which was just announced tonight... so maybe, just MAYBE AMD is devoting resources to products that generate better revenue. I find it just as plausible they got a contract for Vega Compute boards they can sell for 1000s of dollars and "stole" the Vega allotments destined for OEMs. Is that speculation? SURE IS, but it has as much proof as stating AMD has manufacturing problems...

freebird wrote:

What severe issues at AMD??? Fudzilla was claiming AMD is losing money on early Vega cards (which maybe, I have no idea, but I wouldn't be surprised if it is nearly a break-even scenario on the RX Vega 56) but if that is true wouldn't it be "smart" on their part to "slow roll" the release of a money losing card? You are acting like this is a "Barcelona" delay or manufacturing problem, when I highly doubt it is... more likely just high cost of HBM2 and perhaps with 3rd party packaging of the GPU & HBM2 on the interposer. Overpriced is also subjective to the buyer is it not? I thought the 1080 & 1070 were over priced when 1st released, but the 1070 turned out to be a "steal" at $400 and under prior to the Crypto-craze driving up prices. Prices on the 1070s are finally coming down some (may go right back up due to GDDR5 memory price increases however) to where you can find them back under $450 and as low as $415 I've seen this week.

I think you just about nailed it there.
My opinion-reading between the lines in various stories-
AMD offering $100 rebate to sellers to keep prices down puts them in a position where they could actually be loosing money on each sale.
This has come about by being forced to use the more expensive Samsung option until cheaper Hynix option becomes available.

I don't think AMD has manufacturing problems with its Vega chips,but being the beginning of a new chip when yields are at their lowest don't help much.

I look at it pretty simply: between gamers, high-end compute, and cryptocurrency miners, AMD can't make enough; rather, if they could make more, they would. There's a production problem somewhere, else AMD would step up to the demand.

I'd also hesitate to draw any conclusions between the yields and availability of Ryzen and Vega. Even if produced at the same foundry (and I don't believe they are), they're two different animals. Size and HBM2 are both real manufacturing limitations that Ryzen doesn't have to deal with.

Airmantharp wrote:

I look at it pretty simply: between gamers, high-end compute, and cryptocurrency miners, AMD can't make enough; rather, if they could make more, they would. There's a production problem somewhere, else AMD would step up to the demand.

I'd also hesitate to draw any conclusions between the yields and availability of Ryzen and Vega. Even if produced at the same foundry (and I don't believe they are), they're two different animals. Size and HBM2 are both real manufacturing limitations that Ryzen doesn't have to deal with.

It also takes time to ramp up production and built up inventory even if there weren't any "manufacturing" issues. These so-call speculators need to wait until next year to see if the shortages continue to persist.

Airmantharp wrote:

I look at it pretty simply: between gamers, high-end compute, and cryptocurrency miners, AMD can't make enough

Plus demand from Apple: Vega is in the iMac Pro, which starts shipping in December, and is a prominent component per Apple's product page

https://www.apple.com/uk/imac-pro/

Vega graphics. The beast behind the beauty.

When we considered how much we wanted this iMac to be capable of, it was clear that only one graphics chip would do — but that chip didn’t exist yet. So iMac Pro is debuting a new one. The Radeon Pro Vega is over three times faster than any previous iMac GPU, packing the power of a double-wide graphics card into a single chip. All of which translates to higher frame rates for VR, real-time 3D rendering, more lifelike special effects and gameplay at max settings. It’s one huge reason iMac Pro is power incarnate.



Radeon Pro Vega
11 teraflops single precision
22 teraflops half precision
8GB or 16GB High Bandwidth Memory
400GB/s memory bandwidth

(The "over three times faster" is weasel wording, it seems to come from a comparison to the performance of the top-end GPU in the iMac (Late 2015), the Radeon R9 M395X, not to that in the iMac (2017), the Radeon Pro 580.)

HERETIC wrote:

freebird wrote:

What severe issues at AMD??? Fudzilla was claiming AMD is losing money on early Vega cards (which maybe, I have no idea, but I wouldn't be surprised if it is nearly a break-even scenario on the RX Vega 56) but if that is true wouldn't it be "smart" on their part to "slow roll" the release of a money losing card? You are acting like this is a "Barcelona" delay or manufacturing problem, when I highly doubt it is... more likely just high cost of HBM2 and perhaps with 3rd party packaging of the GPU & HBM2 on the interposer. Overpriced is also subjective to the buyer is it not? I thought the 1080 & 1070 were over priced when 1st released, but the 1070 turned out to be a "steal" at $400 and under prior to the Crypto-craze driving up prices. Prices on the 1070s are finally coming down some (may go right back up due to GDDR5 memory price increases however) to where you can find them back under $450 and as low as $415 I've seen this week.

I think you just about nailed it there.
My opinion-reading between the lines in various stories-
AMD offering $100 rebate to sellers to keep prices down puts them in a position where they could actually be loosing money on each sale.
This has come about by being forced to use the more expensive Samsung option until cheaper Hynix option becomes available.

I don't think AMD has manufacturing problems with its Vega chips,but being the beginning of a new chip when yields are at their lowest don't help much.

I don't think Vega SKUs are being sold at a "loss" or razor-thin margins. Performance GPUs are never sold at a loss. They always have some kind of profit margin in their MSRP. It is just Nvidia is making killer profit margins on their GP104 (mainly 1070s) and GP106 due to the lack of fierce competition like Evergreen versus Fermi days. The main problem is that high-end GPUs are high-margin/low volume products and they ultimately make small money. The real money in discrete GPUs is mid-tier stuff through sheer volume and Nvidia is holding onto the market with an iron-fist (1060s-1070s).

Probably another one of those multiple contributing factor situations, not just an HBM2 shortage. Especially when there are two different versions of the Vega 64 product as mentioned before.

By AMD's own slide deck the Z-height is different between versions. The HBM2 stacks are not flush with the die on one of those versions. The width and length also differ between versions. As such I imagine OEMs are having to sort out Vega packages into two different manufacturing lines to match them to an appropriately fitted cooler.

Kougar wrote:

Probably another one of those multiple contributing factor situations, not just an HBM2 shortage. Especially when there are two different versions of the Vega 64 product as mentioned before.

By AMD's own slide deck the Z-height is different between versions. The HBM2 stacks are not flush with the die on one of those versions. The width and length also differ between versions. As such I imagine OEMs are having to sort out Vega packages into two different manufacturing lines to match them to an appropriately fitted cooler.

How is this most likely not all due to the HBM2 shortage issue? Hynix came up short with HBM2. AMD is scrambling to cope with 1) not enough HBM2 2) increased cost of HBM2 3) variety in HBM2 chips resulting in variation in GPU-HBM2 packaging .

Krogoth wrote:

I don't think Vega SKUs are being sold at a "loss" or razor-thin margins. Performance GPUs are never sold at a loss. They always have some kind of profit margin in their MSRP. It is just Nvidia is making killer profit margins on their GP104 (mainly 1070s) and GP106 due to the lack of fierce competition like Evergreen versus Fermi days. The main problem is that high-end GPUs are high-margin/low volume products and they ultimately make small money. The real money in discrete GPUs is mid-tier stuff through sheer volume and Nvidia is holding onto the market with an iron-fist (1060s-1070s).

This seems to be at least somewhat of an obtuse bit of reasoning. "High end GPU products have characteristically had high margins, therefore Vega has high margins".

Again, when you factor in the HBM2 shortage and resulting increased cost to AMD, both in chip price and those related to having to design multiple chip packages and deal with OEM-related issues, etc, how can you be so certain AMD is able to maintain high margins? And this says nothing of lost sales due to delays and lack of supply.

If Vega had such thing margins then why bother with making gaming SKUs at all? It would just make more fiscal sense to sell them as Froniter/FireGL SKUs. Vegas are making ~$50-$100 of profit per unit sold, while Nvidia is enjoying 2x to 3x times that much on their gaming SKUs. Nvidia's recent fiscal reports makes this painfully evident. Nvidia shareholders would love to maintain such margins.

Airmantharp wrote:

I look at it pretty simply: between gamers, high-end compute, and cryptocurrency miners, AMD can't make enough; rather, if they could make more, they would. There's a production problem somewhere, else AMD would step up to the demand.

I'd also hesitate to draw any conclusions between the yields and availability of Ryzen and Vega. Even if produced at the same foundry (and I don't believe they are), they're two different animals. Size and HBM2 are both real manufacturing limitations that Ryzen doesn't have to deal with.

I'm not drawing conclusions between Ryzen and Vega yields since they are completely different chips, but used that to state that I doubt "manufacturing problems" are the issue. My issue was with the title of the post... I don't believe manufacturing "problems" are holding Vega back. Unless it is a manufacturing "problem" when overall demand for it is too HIGH that prices move UP at the consumer level. YES, Chuckula stated AMD can't supply enough Vegas to the OEMs to make custom boards; I don't have an issue with that either. Actually, I don't have a problem with anything Chuckula wrote in the whole post, except that he states that AMD has "manufacturing problems" from the title of the post. If the title was "VEGA high prices and lack of 3rd party custom boards due to high demand and limited supply", I probably wouldn't have even replied. I may have thought somebody knocked him on the head though...because that wouldn't sound like "our" Chuckula But if AMD can only get XXXX amount of working Vega chips from GF on XXXX amount of contracted wafers and supply is limited you can be guaranteed that the supply will go to who is highest in the pecking order... Apple, Professional boards, cloud computing which all either lead to additional business or increased revenue or both.

The reality is VEGA was designed as a COMPUTE GPU 1st; which also covers AMD for higher end gaming also. We might as well face the fact that with the advent of GPU computing and now GPU AI, GPUs are no longer just for "gamers". I don't see an end to it anytime SOON. Crypto-currencies WILL NOT go away... even with a crash in the market is my belief and the GPU compute & AI business is just getting started. So until AMD has the money/resources to build an EXCLUSIVE compute/AI GPU and a separate "dumbed-down" gaming GPU, I see the same thing playing out, probably until 2020. AMD does not have endless resources, they are in the CPU(server,workstation,desktop & laptop), GPU (compute, AI & gaming) and semi-custom business(gaming consoles,SOCs leveraging CPU, GPU, compute or AI), all of which are looking good in the near-term but all require working capital, which for AMD is limited.

I know this is a long response, but another reason why i DO NOT believe that VEGA has manufacturing or yield issues are the fact that my RX Vega 56 has no problems loading the Vega 64 BIOS and running at higher clocks both memory and GPU, what it DOES have is a cooling problem... and I'm looking for an aftermarket cooler as soon as someone makes one at a decent price. I'm not paying $200 plus for one from Europe.
Also, an interesting issue pops up with the AMD compute driver however, Claymore's miner reports 64 compute engines with the compute driver loaded, it doesn't mine any faster from my experience unless I let it run the fan full bore to keep it cool, but with the non-compute Vega driver Claymore only reports 56 compute units.

Again, VEGA isn't overpriced if AMD is selling every one they make. I'm actually surprised I could get an RX Vega 56 for $499 with 2 games, which is a better deal than I got on my 2 1070s (with no games) at $450 & $500 for a water-cooled EVGA, but when I was buying them they were they CHEAPEST prices I could find. Does that make them OVER PRICED? Not if I'm willing to pay that much for them...

A separate gripe would be how Newegg sells the bundle... $129 for two NON-RETURNABLE games...and charged me $376 for the RX Vega 56, so if I wanted to return it; they would only reimburse the $376. That being said, I have NO plans on returning it...

Krogoth wrote:

If Vega had such thing margins then why bother with making gaming SKUs at all? It would just make more fiscal sense to sell them as Froniter/FireGL SKUs. Vegas are making ~$50-$100 of profit per unit sold, while Nvidia is enjoying 2x to 3x times that much on their gaming SKUs. Nvidia's recent fiscal reports makes this painfully evident. Nvidia shareholders would love to maintain such margins.

Because AMD is desperate to stay relevant as a gaming "market leader" for the sake of things like game developer relations, the viability of Freesync, VR - which all relates heavily to their competitiveness of their Vega APU products and as chip suppliers for future consoles, etc.

cynan wrote:

Krogoth wrote:

If Vega had such thing margins then why bother with making gaming SKUs at all? It would just make more fiscal sense to sell them as Froniter/FireGL SKUs. Vegas are making ~$50-$100 of profit per unit sold, while Nvidia is enjoying 2x to 3x times that much on their gaming SKUs. Nvidia's recent fiscal reports makes this painfully evident. Nvidia shareholders would love to maintain such margins.

Because AMD is desperate to stay relevant as a gaming "market leader" for the sake of things like game developer relations, the viability of Freesync, VR - which all relates heavily to their competitiveness of their Vega APU products and as chip suppliers for future consoles, etc.

AMD already concided being a leader in the gaming market since Maxwell. Their shareholders would never allow product to sold at a loss to retain a trivial mindshare. It would be fiscal suicide. It appears you quickly forgotten how high-end gaming GPUs used to have MSRP around $399-$499 when there was heated competition. Nvidia took over with Kepler and Maxwell. They decided to push premiums up and knew that high-end market could bear it. There was little or no price pressure due to lack of competition. The new paradigm is $599+ for high-end GPUs. The cost of making them hasn't gone up. It is actually going down for smaller SKUs.

Krogoth wrote:

AMD already concided being a leader in the gaming market since Maxwell. Their shareholders would never allow product to sold at a loss to retain a trivial mindshare. It would be fiscal suicide. It appears you quickly forgotten how high-end gaming GPUs used to have MSRP around $399-$499 when there was heated competition. Nvidia took over with Kepler and Maxwell. They decided to push premiums up and knew that high-end market could bear it. There was little or no price pressure due to lack of competition. The new paradigm is $599+ for high-end GPUs. The cost of making them hasn't gone up. It is actually going down for smaller SKUs.

When AMD was mapping out Vega, they probably had at least 20% of the non-pro discrete GPU market. That's not exactly trivial. Further, while AMD obviously wouldn't have knowingly put themselves in a position of "fiscal suicide", hindsight is 20/20. By all indications, HBM2 costs many times that of GDDR5X. And prices are negotiated per customer and obviously based on volume. AMD having to scramble to fill a short gap from Samsung is probably paying through their nose (maybe approaching $200 just for 8GB HBM2?). AMD probably banked on better yields (and somewhat better performance) and less lead time for the Hynix HBM2 agreement to mature. And if it does ramp up in the near future, AMD could begin to make a decent profit on Vegas going forward. But for now, I think it highly likely that the profit margins on RX Vega 56 could be thin to non-existent.

cynan wrote:

When AMD was mapping out Vega, they probably had at least 20% of the non-pro discrete GPU market.

34-41% from 2010 to 2014, fluctuating, of the dGPU market according to Jon Peddie Research, and given NVIDIA's greater dominance in the workstation GPU market I'd say you're out by a factor of 2.

I am an AMD fan, but Vega is hot and expensive. On a side note, I suspect that it will take a good amount of time for some of the new features baked into the chip to be used fully in gaming and pro software, much like older GCN cards that kept getting faster as they aged. So, assuming Vega's production issues are not HMB2 related, and AMD is churning out dies with high defect rates, where is the heavily cut down GPU? If the dies are brain damaged, why no Vega 48 or 36 with 2 stacks of slow HBM2, one stack, or low density stacks of HBM? The die is huge, so unless the defect renders them useless, I would think there would be rumblings of OEM only or Asian market cards with low numbers of operating ROPs and stream processors. That or talk of an incoming a respin of the die. It's a hot mess, but no one seems to be offering any solution to this "huge" problem, publicly at least.

Topinio wrote:

cynan wrote:

When AMD was mapping out Vega, they probably had at least 20% of the non-pro discrete GPU market.

34-41% from 2010 to 2014, fluctuating, of the dGPU market according to Jon Peddie Research, and given NVIDIA's greater dominance in the workstation GPU market I'd say you're out by a factor of 2.

Guess you've not familiar with understatement for emphasis.

cynan wrote:

Guess you've not familiar with understatement for emphasis.

Guess understatement for emphasis fails to translate every time on discussion boards

Stands to reason. Also, on the mining front, "but we think it's just a short term fad" isn't such a great justification for not increasing supply any more given that GPU mining has been a thing for, what, about five years now? It ebbs and flows, sure, but so does gamer demand based on what AAA titles and/or console hardware improvements are happening in a given year.

NovusBogus wrote:

Stands to reason. Also, on the mining front, "but we think it's just a short term fad" isn't such a great justification for not increasing supply any more given that GPU mining has been a thing for, what, about five years now? It ebbs and flows, sure, but so does gamer demand based on what AAA titles and/or console hardware improvements are happening in a given year.

It's also an exceedingly poor justification given that the demand for high-performance compute is skyrocketing in the enterprise space. When AMD talks about a lot of the features that uninformed AMD gamers get excited about (that will never be used for gaming), that's where they're actually going with it.

And that reinforces my point above, in support of the OP: something's broke. There's no rational reason for AMD to not be producing as literally as many Vega parts as they possibly can.

They're making as many as they can, but HBM prices have quadrupled in the past year...

EDIT: DRAM has jumped nearly 100% as well. Everyone is switching to mass producing flash...

Airmantharp wrote:

NovusBogus wrote:

Stands to reason. Also, on the mining front, "but we think it's just a short term fad" isn't such a great justification for not increasing supply any more given that GPU mining has been a thing for, what, about five years now? It ebbs and flows, sure, but so does gamer demand based on what AAA titles and/or console hardware improvements are happening in a given year.

It's also an exceedingly poor justification given that the demand for high-performance compute is skyrocketing in the enterprise space. When AMD talks about a lot of the features that uninformed AMD gamers get excited about (that will never be used for gaming), that's where they're actually going with it.

And that reinforces my point above, in support of the OP: something's broke. There's no rational reason for AMD to not be producing as literally as many Vega parts as they possibly can.

It is because making big pieces of silicon in bulk isn't easy or cheap to do. Nvidia isn't exactly making a ton of GP100 and GP102 chips either but they really don't need to from a fiscal standpoint. Their bread and butter in the gaming market has always been mid-tier SKUs. GP100 is still kinda hard to get if you really want one. GP102 stuff isn't exactly that common either. There's just enough stock to satisfy current demand for high-end gaming market and professionals. The key difference is that Nvidia typically waits for their reserves to build-up before "officially" launching their product to the general market. AMD RTG couldn't afford to do the same thing. If AMD RTG waited too long then Volta would have practically destroy Vega line-up. AMD RTG had to launch Vega before Volta or it would have never sold in any significant numbers = total loss which is far worse than a lackluster launch.

Krogoth wrote:

It is because making big pieces of silicon in bulk isn't easy or cheap to do. Nvidia isn't exactly making a ton of GP100 and GP102 chips either but they really don't need to from a fiscal standpoint. Their bread and butter in the gaming market has always been mid-tier SKUs. GP100 is still kinda hard to get if you really want one. GP102 stuff isn't exactly that common either. There's just enough stock to satisfy current demand for high-end gaming market and professionals. The key difference is that Nvidia typically waits for their reserves to build-up before "officially" launching their product to the general market. AMD RTG couldn't afford to do the same thing. If AMD RTG waited too long then Volta would have practically destroy Vega line-up. AMD RTG had to launch Vega before Volta or it would have never sold in any significant numbers = total loss which is far worse than a lackluster launch.

Sure, and it means positioning an expensive, hard to produce product against Nvidia's consumer-focused half-GPU, which since Kepler, has been competitive against AMD's best in the gaming market.

And that's just poor design and planning from AMD for gaming, both in unoptimized architectures and the exceedingly poor decision to go with HBM/HBM2.

Airmantharp wrote:

Krogoth wrote:

It is because making big pieces of silicon in bulk isn't easy or cheap to do. Nvidia isn't exactly making a ton of GP100 and GP102 chips either but they really don't need to from a fiscal standpoint. Their bread and butter in the gaming market has always been mid-tier SKUs. GP100 is still kinda hard to get if you really want one. GP102 stuff isn't exactly that common either. There's just enough stock to satisfy current demand for high-end gaming market and professionals. The key difference is that Nvidia typically waits for their reserves to build-up before "officially" launching their product to the general market. AMD RTG couldn't afford to do the same thing. If AMD RTG waited too long then Volta would have practically destroy Vega line-up. AMD RTG had to launch Vega before Volta or it would have never sold in any significant numbers = total loss which is far worse than a lackluster launch.

Sure, and it means positioning an expensive, hard to produce product against Nvidia's consumer-focused half-GPU, which since Kepler, has been competitive against AMD's best in the gaming market.

And that's just poor design and planning from AMD for gaming, both in unoptimized architectures and the exceedingly poor decision to go with HBM/HBM2.

You mean since Pascal. Tahiti surpass GK104 and managed to catch up to the binned GK110 (vanilla 780)in a year or so, while Hawaii kept up until full GM204. Maxwell owes most of its performance gain in gaming over Kepler from tiled rendering which alleviates bandwidth and geometry resources. Maxwell had no little or no gain in general compute and shading performance over Kepler. Pascal is basically a tweaked Maxwell with massive clockspeed roids due to the 16nm process. Titled rendering is what allowed Pascal to get away with GDDR5 and GDDR5X in gaming SKUs without having resort to HBM in order to keep chips happy.

Airmantharp wrote:

And that's just poor design and planning from AMD for gaming, both in unoptimized architectures and the exceedingly poor decision to go with HBM/HBM2.

Yet AMD dominates in HPC / compute intensive workloads. When you only have the budget for one fabbed design, you make it count.