AMD has never been afraid to try new things, from hitting 1GHz first, to creating a true multicore processor, most recently adopting HBM and HBM2 into their graphics cards. That move contributed to some of their recent difficulties with the current generation of GPUs; HBM is more expensive to produce and more of a challenge to implement. While they were the first to implement HBM, it is NVIDIA and Intel which are benefiting from AMD's experimental nature. Their new generation of HPC solutions, the Tesla P100, Quadro GP 100 and Lake Crest all use HBM2 and benefit from the experience Hynix, Samsung and TSMC gained fabbing the first generation. Vega products offer slightly less memory bandwidth as well as lagging behind in overall performance, a drawback to being first.
On a positive note, AMD have now had more experience designing chips which make use of HBM and this could offer a new hope for the next generation of cards, both gaming and HPC flavours. DigiTimes briefly covers the two processes manufacturers use in the production of HBM here.
"However, Intel's release of its deep-learning chip, Lake Crest, which came following its acquisition of Nervana, has come with HMB2. This indicates that HBM-based architecture will be the main development direction of memory solutions for HPC solutions by GPU vendors."
Here is some more Tech News from around the web:
- FCC Approves First Wireless 'Power-At-A-Distance' Charging System @ Slashdot
- Guidemaster: Everything Amazon’s Alexa can do, plus the best skills to enable @ Ars Technica
- INQ's best and worst of tech in 2017
- Linksys LAPAC2600 AC2600 Dual Band MU-MIMMO Access Point Review @ NikKTech
- Acoustic Attacks on HDDs Can Sabotage PCs, CCTV Systems, ATMs, More @ Slashdot
- 10 Tech Products That Are Next to Impossible to Repair @ Techspot
- noblechairs ICON PU Faux Leather Chair @ TechPowerUp
The problems and cost of HMB2
The problems and cost of HMB2 may be the reason that AMD has shut down Vega 56 production. Yes every once in a while a batch of 56’s show up and promptly sell out even at prices 60% over MSRP.
http://www.nowinstock.net/computers/videocards/amd/rxvega56
At the same time custom Vega’s are MIA and word on the street is that a lot of vendors are reconsidering and may not ever sell custom Vega’s even though they have initially stated that they would.
I believe Vega is produced at
I believe Vega is produced at GlobalFoundry and its a very small foundry with very limited production.
But AMD is forced to use them, and GF is not increasing capacity to meet AMD product demand.
Its unclear if GF will invest in capacity in 2018 or if AMD will take advantage of its revised wafer agreement (that AMD paid over half a billion to redact) before it expire in 2016.
AMD is being hurt really badly by GF once again.
When AMD fail to order enought chips, GF make AMD pay 100% penalty and resold the unused capacity to AMD “competitors”
Now that AMD need higher capacity, GF is simply telling AMD “Sorry, just be patient” all the while AMD cant supply demand.
I thought they had a deal
I thought they had a deal with Samsung for supply if GloFlo reached capacity. Maybe RX Vega is being sold at a loss or no profit so the lack of cards are by design. It would be interesting to see sales/availability of Radeon Pro & MI25.
GF is not an HBM/HBM2
GF is not an HBM/HBM2 producer as that HBM2 production is not related to GF or AMD’s foundry agreement with GF for CPU/GPU dies. HBM2 dies are sourced through SK Hynix and Samsung mostly and Samsung was quicker out the door with its JEDEC complient HBM2 supplies while SK Hynix had issues with its HBM2 production. Samsung is much larger than SK Hynix so Samsung had the resources to get its HBM2 to market faster and take advantage of the great demand for HBM2 from AMD, Nvidia, Intel, others.
P.S. AMD is smart not to increase its Vega production and get burned again if the coin mining market cools off again and the Vega 10 base die design top out at 64 ROPs so AMD Lacks that same amount of availale ROPs compare to Nvidia’s GP102 based GTX 1080TI(88 ROPs). GP102 has at least 96 ROPs available if Nvidia needs to spin up a more powerful variant than the GTX 1080Ti so Nvidia will not lose in any FPS metrics form Vega 10 based gaming variants.
GF must be meeting AMD’s Zeppelin die production goals else we would be hearing that AMD has optioned some alternative Production at Samsung’s/Other’s foundries.
The one with the most GPU ROPs is the one who will win the gaming FPS metrics race as sure as the sun will come up every day because it’s those ROPs that matter!
Vega is manufactured at quite
Vega is manufactured at quite a few different foundry locations, such as from GlobalFoundries, Advanced Semiconductor Engineering (ASE), Siliconware Precision Industries (a.k.a SPIL, who are in charge of the bulk of Vega 11 orders) are all supposedly building Vega packages. Next year the Vega 20 (a.k.a 600 series maybe?) will see Taiwan Semiconductor Manufacturing Company (TSMC) enter into a contract with AMD for Vega chips on 7nm (or maybe just 12nm). GlobalFoundries will remain their go to Zen partner and maybe even build some Vega 20’s themselves, as the foundry landscape can obviously change. But it is NOT one company manufacturing the Vega GPU’s. If anyone thinks that they would be sadly mistaken. AMD can’t supply Vega demand because of the complexity with HBM2 package and the size of the Vega die combined, not to mention cryptocurrency which had consumed everything vega based. This is why rumors point to them in talks with TSMC because it is well known that TSMC can handle the extra load, just look at the insane size of Volta. GlobalFoundries needs to expand, so I am not sure why they haven’t yet. Or maybe they are and we just do not realize it yet? OR, maybe they realize the small dies of Navi on 7nm will be a piece of cake for them and don’t want to expand when Navi is as close as it is. I am just speculating though, as it makes plenty of sense.
it wouldn’t matter how many
it wouldn’t matter how many video cards AMD could make, they can never produce enough at this point. the mining craze is what is really hurting the supply, doesn’t have much to do with gf.
They stopped producing
They stopped producing reference cards as you not compete with the AIBs since they won’t offer any performance increases. They didn’t just stop making Vega 56 altogether.
AIB’s Vega’s are MIA.
AMD
AIB’s Vega’s are MIA.
AMD shot themselves in the foot with that move. By not selling reference cards until AIB cards are available they basically gave the market to Nvidia this holiday season.
Vega is sold out everywhere
Vega is sold out everywhere and those miners’ money is as good as any gamer’s money and Nvidia needs sales also! So its happy XMAS/New-Year for the Billionares they are all a trillion dollars richer for 2017 and a big tax break for 2018 and old JHH over at Nvidia and our Lord Gaben will have even more dosh in 2018!
Now if those Billionares would just invest more with that Musk Billionare’s auto operation so he can get an electric Pickup Truck to market then Bubba can be Happy for that!
Just Imagine Bubba at the camp site Playing his Nvidia GPU based rig powered by his Tesla Pickup in that camper top with plenty of solar cells and battery capacity. Hell Bubba could haul a trailer behind with some extra battery capacity and go for hundreds of miles before needing access to any plug.
Oh Musk is going to create an electric pickup truck with a torque curve(Damn near a line) sent down from the Gods just for Bubba’s use and let’s hope that there will be some Vans on the horizon for all those that want to do some camper conversions. Yee Haw! Bubba’s done gone all electric!
AMD does not need no stinking gamers only to make that mad dosh so enjoy them Nvidia extra ROPs, Vern, because ROPs are the GPU’s NADS for flinging out them FPS metrics but the miners do not give a rat’s red A$$ ’bout no ROPs! So AMD just needs to keep up the good work and not overproduce Vega 10 dies and keep them prices high, the AIB partners will be all over AMD wanting to pay more for them Vega 10 dies and it’s a GPU maker’s market to up them wholesale prices a good bit and get them margins higher.
Bubba’s got him some Nvidia ROPs and an all electric Pickup truck can power it all!
The very reason that AMD/SK
The very reason that AMD/SK Hynix worked with JEDEC to get HBM established as a open standard is intended to have Nvidia, Intel, and others adopt the now JEDEC HBM/HBM2 standards. AMD alone could not provide enough HBM usage to allow for HBM’s/HBM2’s overall economy of scale. All memory technologies have to be adopted by a larger industry or the economy of scale will never have a chance of working out. A single CPU/GPU maker does not represent a large enough economy of scale for a complicated product such as HBM/HBM2 to succeed in any manner as such is the great initial costs involved with the packaging and die stacking technology and physical plants that need to be built up around HBM’s specialized production.
Any radically new memory technology like HBM/HBM2 requires a massive capital investment in physical plant resources and that includes some very expensive and specialized HBM packaging equipment. The only way to properly amortize the massive HBM memory development costs requires an industry level of investment that requires a larger production volume and one company’s needs can not properly fill and allow HBM’s massive development and R&D cost to be properly amortized over a reasonable amount of time. HBM/HBM2 is not only an AMD/SK Hynix project it’s a standard with Samsung and others creating their own hardware solutions that adhere to the JEDEC standard. So AMD and SK Hynix have shared none of their in house HBM/HBM IP with any other HBM/HBM2 maker and each individual HBM/HBM2 maker creates their own IP that adheres to that JEDEC HBM/HBM2 standards.
Any new technology like the HBM2/HBM2 Standard needs wide industry adoption or the cost will never come down. HBM2’s costs currently is the result of more of a supply and demand issue compounded by the need to properly amortize those initial capitol investments by the HBM makers. And Samsung has the lead in having the resources to get its JEDEC standard complaint HBM2 to market sooner than AMD/SK Hynix. Eventually as the market shifts over to using HBM2 as the overall standard way for GPU’s and other Accelerators and eventually even CPUs then that volume and economy of scale will kick in and HBM2 will become the lower cost solution.
This will not happen overnight as every new technology takes years for any proper economy of scale to take hold, just look at some of the early PCs and some of those cost $6000+ dollar range for single core processors that ran in the 100-200MHz clock range and came with 5MB of Hard Drive storage that cost $1500 in 1980( ST-506 first 5 1⁄4 inch drive released with capacity of 5 megabytes, cost $1500 USD ).
You can thank AMD/SK Hynix for spearheading the adoption of that AMD/SK Hynix work that resulted in JEDEC standard but AMD’s goal was hardly altruistic as AMD just needs as many others to begin using HBM2 so the economy of scale will quickly take hold and AMD will be able to benefit along with others at a more reasonable cost from HBM/HBM2 intrinsic advantages over the more power hungry and space inefficient GDDR# memory standard. Everybody and their dog wants HBM2 usage to become affordable.
Really folks the entire history of the PC market is replete with costly new technology examples that are gone through that economy of scale process exactly like HBM is going through! So what’s so special about HBM in this regard. This is just the standard growing pains of any new technology and the economics are exactly the same.
AMD is not making and more new Reference GPU cards because no one will want the reference cards once the AIB products are fully there with better performance and AMD needs to get its margins higher. So it’s better to just sell the dies at a higher markup and let the AIB folks take on the cost of development of the PCIe cards/SKUs. If you do not like that price of AMD’s(Now The AIB partners’ SKUs) and if Nvidia’s cards cost less and have better FPS gaming metrics then by all means buy Nvidia’s AIB/Partner products because AMD’s GPUs will sell out regardless because of the current BlockChain related market. And the professional compute/AI markets are getting more of those Vega 10 base dies(WX 9100/MI25 SKUs) than any coin mining folks, and the pro market includes Apple’s iMac Pro/Mac Pro.
AMD is not going to produce an excess of Vega 10 base dies just to placate gamers when that HBM2 on the Vega 10 base die designs cost so much that AMD is barely breaking even with its RX Vega 64/56 gaming SKUs. Lisa Su has stated many times that AMD is no longer the low cost solution and that’s the way it will remain going forward. Nvidia has always been charging the proper amount on its GPU SKUs to profit and AMD has to charge like Nvidia or AMD will never be able to compete and that requires billions in R&D, hardware, and software/firmware/driver development costs.
The professional market wants HBM2 for the power savings alone with HBM2’s space saving more icing on the cake and the professional market pays the bills while the gaming market(For AMD Mostly) is a loss leader for AMD. Consumer gaming leaches heavily from the professional market revenue streams as the consumer market does not ever want to pay its fair share to cover those billions in investments that it takes to develop CPUs/GPUs in the first place.
Quote: AMD is not making and
Quote: AMD is not making and more new Reference GPU cards because no one will want the reference cards
Funny I guess I need to change my name to “no one” because I only want reference cards. I do not need AIB over priced cards with barely any performance improvement dumping their heat inside the case.
Nvidia marketed their cards properly with the FE cards $50 over MSRP thus allowing AIB vendors a profit. Currently Nvidia FE cards can still be bought. AMD Vega’s not available. Big screw up.
Oh stop your whining and pay
Oh stop your whining and pay the proper markup, AMD is not bound by that Refrence MSRP on these new AIB/Partner cards and AMD only has to worry about the GPU Die/GPU die costs which AMD should be charging more for in the first place so they can make a little revenue off the transaction. You consumer/gamers have an inflated sense of entitlement with regards to AMD having to drop everything and kiss your little behinds! AMD made those Vega 10 dies for the professional compute/AI markets and only added just enough ROPs to compete with the GTX 1080 and Vega 10’s base die design was frozen long before Nvidia dusted up its GP102 base die based GTX 1080Ti variant to take the FPS crown.
You are not paying enough money for Vega 64/56 to justify any more of AMD’s attention this time around. Those MSRPs and Margins need to go hiher for AMD so AMD can fund a better level of GPU development next time around and AMD’s gaming GPU market share is not large enough for AMD to pay gamers too much mind. You get what you pay for and AMD needs to move on and focus on next years designs and let it’s AIB partners have some revenues or AMD will lose those AIB partners loyality. It’s not big screwup if some butthurt gamers get all hot under their dirty Ts about not getting any more refrence designs and AMD needs to get out of those loss leader Vega refrence sales that are costing too much and not producing the needed revenues to justify that business.
The AIB partners have the Vega 10 dies now to produce some tweaked designs at their own expense with AMD getting a better price for those Vega dies/HBM stacks without worrying about MSRP and all the MSRP whining from the little gaming scrubs that won’t pay a fair market price for Vega. Let the miners have it all if the gamers will not pay and Nvidia needs to up its MSRP to compensate for AMD’s higher average pricing as Nvidia needs to get those higher pricing benifits for its billionare investors. The PC market is shrinking and prices need to go up to allow for a proper pricinge level to support any lower sales because R&D and fab/process nod expenses are going up not down and those costs need to be amortized over less units sold.
PC are about to become more expensive as PC sales drop to cover all those related expenses that have to be paid before and profits can be calcualted. AMD needs to maximum its existing sales margins above 40% for its share holder value to go up so higher MSRPs are in order to pay for its all. Nvidia needs to be properly compensated for giving gamers all those extra ROPs with their mad frame flinging FPS metrics that gamers so love so that justifies Nvidia increasing its MSRP even higher and making even more revenues also.
Prices need to go up not down for AMD until AMD has massive amounts of resources to compete with Nvidia in the consumer gaming market and Nvidia needs to always be be higher priced because that’s what Nvidia is known for and the gamers will pay or thay will not play. If the AMD game customers will not buy Vega the coin miners will so that means put of the cash or hot the street as AMD is all about the same thing that Nvidia is about and that’s maximum revenues and a fair markup for its GPU SKUs.
Won’t someone think of the of the billionares, they need that extra trillion dollars in wealth that they amassed this year. And they will get more next year those 500 billionares and JHH(4.6 billion) of Nvidia is one of those billionares that need higher MSRPs!
Edit: put of the cash or hot
Edit: put of the cash or hot the
To: put up the cash or hit the
Quote: Oh stop your whining
Quote: Oh stop your whining and pay the proper markup
Like this ASUS ROG Radeon RX Vega 56 STRIX-RXVEGA56-O8G-GAMING for $649.99.
https://www.newegg.com/Product/Product.aspx?Item=N82E16814126233&ignorebbr=1
So you accept a 62.5% markup over the AMD reference card price of $399.99 as proper.
That’s what the demand curve
That’s what the demand curve pricing function is for and that SKU shoude have had a 500+ MSRP at introduction because AMD’s actuaries where not doing their job and insisting that AMD’s marketing department was making a mistake in setting that SKUs MSRP so low in the first place.
That’s on the coin mining demand for GPUs and the gaming demamnd because AMD’s GPUs are shader heavy and great for computer and good enough for competing with the GTX 1080.
Remember that AMD only has one base Vega die design(Vega 10) that has to be a jack of all trades so ROPs where cut back to make room for shaders. Nvidia has its many GPU base die designs to make use of GP100, GP102, GP104, GP106, GP108 and plent more GPU variants based on those many base die designs vurses AMD single jack of all trades Vega 10 base die, with Polaris still used for the mainstream market currently by AMD.
ROPs are the GPUs NADS and the GPU with the most Nads wins the FPS bubba Gameing metrics race hands down. No ROPs no Frames at all no matter the TMUs, Tessellators, or Shader cores!
ROPs! ROPs! ROPs! is what the gamers should be demanding from AMD!
Never have so many electrons
Never have so many electrons been used to say so little.
It did it job and pissed you
It did it job and pissed you stinking sportswear wearing Chav A$$ off you product of and unholy union between siblings.
Did that supply micro/macro economics lesson hurt your poor little single brain cell floating in a sea of lipids. That poor little cell so enraged is just what that post was good for in your case. Oh you so hate that economy of scale that is so neccessary for the technology market to function efficiently! AMD, Nvidia, Intel, and their dogs are all about HBM2 and getting that unit cost down so the entire market can benifit. Sure JHH over at Nvidia will be using GDDR6 along with others but Nvidia will not pass and cost savings from GDDR6 on the consumer, oh no JHH is all about share holder value going up and JHH has his billions and all you have is a dirty stinking tracksuit and a scabies infested baseball cap! You Chavs never heard of antiperspirant or deodorant or reading and writing!
Edit: It did it job
To: It
Edit: It did it job
To: It did its job
Edit: of and unholy
to: of an
Edit: of and unholy
to: of an unholy
Needs more proof reading! But what the Hell it’s still fun to piss WhyMe off!
How much coke did you snort
How much coke did you snort or Adderall did you take before writing this. Good lord.
Haha, I was thinking the same
Haha, I was thinking the same thing!
Poor Jeremy just writing an
Poor Jeremy just writing an article and folks accusing him of abusing substances!
Looks like you are just some Chav with anger issues there bubba!
Something tells me they have
Something tells me they have no experience with what they speak of, but at least they read it. I am disappointed that neither took the opportunity to accuse me of being a shill though.
Something tells me they have
Something tells me they have no experience with what they speak of, but at least they read it.
And HBM 3 is supposed to be
And HBM 3 is supposed to be around the corner
If AMD goes with bleeding
If AMD goes with bleeding edge MBM 3 then they will screw themselves once again.
GDDR6 is what they should go with.
Leave HBM 2/3 for the ultra high Enterprise/HPC market where low yields and high prices reign.
It’s been stated that they
It’s been stated that they will use gddr6, but it’s more likely that once again all the eggs will be placed in the HBM basket.
No HBM2 is good enough for
No HBM2 is good enough for flagship GPUs and AMD can go GDDR6 for it’s other GPU SKUs but AMD needs to charge more for all its GPUs if the miners continue to be willing t pay more. AMD needs to set an artifically high MSRP and make more revenues off of consumer deadbeats. The Miners are sure willing to pay and AMD can still sell GPUs as compute/AI accelerators and make even more off of its Vega 10 dies and their HBM2. The Vega Refresh SKUs all need really highe MSRPs and then the retailers can lower the pricing if they are not selling AMD’s GPUs at that higher MSRP. There is not restrictions on retailers selling below Retail as retailers get the GPUs and wholesale pricing and AMD needs to raise its wholesale pricing and MSRP higher to make a proper markup on its GPUs. GPU prices need to go up for AMD as their consumer GPU market share on discrete GPUs is still too small.
AMD is in a much better position with its integrated APU graphics this time around and Zen is very close to Intel in IPC. So that’s good news for AMD, but HBM2 is the future and it’s about time that gamers are forced to foot the bill until AMD’s professional Epyc/Radeon Pro WX/Radeon Instinct products get AMD back to its Opteron Levels of server market share that put AMD’s share price around $93 a share! Make them Gamers pay!
What…. XD. LMBO.
What…. XD. LMBO.
It all boils down to a price
It all boils down to a price increase for GDDSR6 and HBM2, and consumers need to start paying their fare share for all that R&D and software/driver expense or consumers better hope that, for AMD at least, the Red Team’s enterprise sales take off over the next few years. And that enterprise includes not only AMD’s Epyc CPU SKUs but also those Vega based Radeon Pro WX and Radeon Instinct Compute/AI GPU SKUs also. AMD needs to be working on a interim Dual Vega 10 die based Compute/AI SKU that make use of the Infinity Fabric to wire up the dual Vega 10 dies and make them work more efficiently for heavy compute/AI workloads. Vega’s support for the Infinity Fabric means that any Vega 10 die based SKUs that use dual Vega 10 dies on a single PCIe card will perform well against the Nvidia GV100 based Tesla compute variants with those Tensor Cores.
In Gaming, AMD has a serious ROP deficiency similar to a serious nougat deficiency and Bubba Gamer don’t want no candy bars with nothing but thin air inside instead of chewy nougat! ROPs are what fling the frames and win that frame flinging contest and Bubba will spend serious dosh for the bragging rights alone. Just look at Bubba spending to get all the flashing LEDs on his gaming rig and GPUs are a vanity purchase for plenty of Bubbas out there, frame quality be damned! As that FPS race is just like a NASCAR race to Bubba. So it’s good for both AMD and Nvidia when the demand for their GPU SKUs remains high for whatever the hell reason.
It’s no time to pass any savings on to any damn whining consumers that for so many years have not paid their fair share for R&D and other expenses and HBM2 is needed for the energy savings alone! So HBM2 needs to be used on all GPU SKUs no matter the Initial costs! It’s flat out time to tell the consumer to pay up or get to stepping because the Professional market has had the consumer market leaching off of the professional market’s higher mark-ups that really pay for those R&D bills. Let the consumer market help get the HBM2 volumnes up so that HBM2 market’s economy of scale can kick in sooner rather than later.
Make fat Bubba Pay Pay Pay just to play or have his vanity based bragging rights and 500LBS of LED flashing trash lights! Look at Bubba’s rig you can spot is at high noon from the ISA with all them LEDs! Set them MSRP’s very high and increase those wholesale prices Nvidia and AMD if the GPU SKU damand is there then the price must go UP!
Start praying for the red
Start praying for the red team. I left them a long time ago.
Well Folks need to tell AMD
Well Folks need to tell AMD to get the ROP counts up there on par with the GP102 die design and maybe AMD will be in for some better gaming sales on discrete GPU for gaming! But with the miners snapping up AMD’s GPUs for compute AMD has no problems selling whatever consumer stocks of Vega SKUs are available.
AMD’s EPYC sales will carry AMD into their second stock price golden age if AMD’s Server market share gets back up to 20-25% territory. Vega 10 is actually a better compute design than most of Nvidia’s offerings except the GV100 but some dual Vega 20 die based SKUs will be monsters if AMD needs to get even more compute fitting on a single PCIe card. Vega 20’s DP FP ratio to its SP FP ratio will be at a 1/2 DP to SP rate and maybe there needs to be some DP FP units that can do some 16 bit packed math to bump up that 16 bit math rate for AI/Inferencing workload ability.
AMD needs to develop some Tensor Core IP of its own and get that on the market but that Project 47 system loads 80 Vega 10 die based MI25 compute/AI accelerators into an single cabinet along with 20 single socket Epyc 32 core CPUs/MBs so that’s plenty of Vega 10 die sales right there.
AMD does not need gamers that much when the professional GPU markups are much higher and the revenues that come from the professinal market will get AMD back to that 90+ dollar per share price range that AMD enjoyed from its Opteron business in 2005, and that was on server CPUs only because AMD had yet to acquire ATI.
AMD stands a better chance of gaing more integrated GPU market share with its Raven Ridge APUs than any of its discrete gaming GPU sales as AMD’s Raven Ridge/Vega graphics can not be matched by Intel currently and more than likely not within the next few years. And the majority of the GPU market is Integrated Graphics if the overall numbers are looked at. So that bodes well for the Red team’s outlook.
Gamers need ROPs and at 30-120+ FPS not much else will be noticed with any per frame quality as long as there are no jagged edges and the frame rate variance is within the limits that the mind can not take notice of. The human brain is designed to dectect edges and evolutionarily speaking there has never been a problem with frame rates/frame variance in real life. So keeping that artifical frame rate variance in check is necessary because GPUs can’t really match the frame rate of trillions upon trillions of photons “bouncing off” of trillions of trillions of atoms per second. So Nvidia wins with More ROPs over AMD with less but the Vega 10 base die design was frozen longe before the GTX 1080Ti became available so AMD really designed Vega 10 to compete with the GTX 1080!
One need only look at Nvidia’s financial ability to fund the design of the GP100, GP102, GP104, GP106, GP108 base die designs compared to AMD’s only one Vega 10 compute heavy design to see that Nvidia has more designs to work with. And the GTX 1080Ti is derived from the GP102 base die and not the same GP104 base die that the GTX 1080 comes from. Nvidia is more than 10 times larger than AMD in market cap value so that’s enough for Nvidia to maintain a gaming lead always over AMD on discrete GPUs at the tip top end of gaming and not necessarily at compute where AMD holds its own quite well against Nvidia’s competition.
https://www.pcper.com/news/Ge
https://www.pcper.com/news/General-Tech/AMD-Working-GDDR6-Memory-Controller-Future-Graphics-Cards 🙂