Nice find...Try for yourself.
http://isine.com/DieYieldCalculator.html
Samsung was claiming around 0.2 defects/cm^2. Accurate? Play with higher.
You only get the Full working die yield not partial ones.
Rx480 specs strongly suggests a RX480X 40 CU model
In any case, not that many wafers are needed and we can safely assume that GloFlo is not swamped by orders.
edit:
I get 152 full die out of 238 possible using a 0.2/cm^2 defect density. Harvested will be a lot of the remaining 86.
Ballpark numbers, yes, but still somewhere to start.
What can AMD do? Have a mandatory survey before someone buys a GPU to make sure they aren't a miner? Force companies like Newegg to only sell one per customer? I just don't see any practical way AMD can "prepare" for the miners. They will make every card they can, and sell every card they make.
Well.....that isn't completely true. If AMD really really wants to keep miners away they need to sabotage mining in their driver and make it so Polaris only gets like 11MH. But then that GPU better damn well beat the GTX 1060 by a good margin day one or I don't know who is gonna buy them.
I believe those numbers will be buried in the financials, and still got a few months before they report anything.Do we know if the chips for new PS4 are coming from GF, or if Samsung are doing those directly?
Definitely makes a big difference to how confident you can be in the volume of 480's they'll have right away - Sony will want massive stock built up before a launch and would probably get priority.
On the other hand if there delays due to that it'll help them bringing production on the desktop/mobile stuff up to full speed later on, so whatever
Whatever, I definitely don't think AMD will want to wait long to build up excess 480 stock. The 1060/70 are coming soon enough, and they'll want as long as they can possibly get uncontested.
Maybe put software into the driver package that disables the card if it senses hash activity.
Maybe put software into the driver package that disables the card if it senses hash activity.
I'm not criticizing mining with it, I mean it's free money if this card ends up being a 150W 390x. I just meant it's a real kick in the nuts to AMD. Gamers who want these cards at MSRP are going to be pissed, and then when they buy these cards used with high fail rates after 24/7 mining use they'll blame AMD for making crap.
Limited addition mining card with optimized drivers would be a better approach I'd imagine. Probably easier optimizing drivers for mining than 1000's of games anyways. Could probably even up the MSRP some to top it off.
Ya, if AMD designed a card for ethereum that has 2-3X the output of an R9 390, many miners would pay 2-3X what the RX 480 costs, but by the time AMD does this, ethereum mining will switch to Proof of Stake making a $200-300 million investment into a new dedicated mining 14nm FinFET ASIC a gigantic loss of $. The reason AMD couldn't capitalize on mining for almost a decade is because it's not something that's certain and you cannot guarantee that by the time a mining specific GPU is designed (2-3 years) that mining will still be in demand. AMD is strapped for cash and have limited engineering and financial resources to pull something like this off.
I thought mining had moved on to ASICs/FPGAs? What happened?
with how much amd messed up 200 series launch and supplies/stock issues, they are that incompetent at timesMy point was if you remove miners entirely, the target market which AMD is going after is ~50X the size of Ethereum miner's market. Using the current Ethereum network rate of ~ 3,300,000 MH and assuming a worst case scenario that on average a GPU ~ 20MH/sec and all of them are AMD cards (they are not), that means there are probably at most 165,000 AMD graphics cards mining ether. Ok, now look at the context for RX 480:
AMD's potential market with $199 RX480/$229 RX 480 8GB is up to 8-10 Million PC gamers a quarter. The 165,000 mining cards is a lifetime number to date!
I am sorry but if AMD or anyone else tries to use mining as an excuse why AMD cannot fulfill demand for RX 480 (i.e., by pointing a finger at miners), it is because AMD failed to build-up sufficient inventory and failed themselves to be true to their word of targeting 8-10M potential PC gamers a quarter. Ether mining might have an initial effect on shortages of RX 480 but if AMD actually legitimately plans to target the mainstream + performance graphics card market for gamers, the demand from miners should be a tiny spec on their supply side. Think about it, let's say we use Q3 2015 #s, even if there is a demand of 100,000 units of RX 480 cards from miners in Q3 2016, that's only 1.25% of the Mainstream/Performance market in the upcoming Q3 2016 quarter! (100K/ 8M cards).
AMD's job is to manufacture as many as they can to meet demand. The nature of the mining demand is irrelevant, since mining is just another general purpose compute activity a graphics card can perform (just like gaming, rendering, streaming video, OpenCL computing, distributed computing, financial analysis, Photoshop, etc.).
Ya, if AMD designed a card for ethereum that has 2-3X the output of an R9 390, many miners would pay 2-3X what the RX 480 costs, but by the time AMD does this, ethereum mining will switch to Proof of Stake making a $200-300 million investment into a new dedicated mining 14nm FinFET ASIC a gigantic loss of $. The reason AMD couldn't capitalize on mining for almost a decade is because it's not something that's certain and you cannot guarantee that by the time a mining specific GPU is designed (2-3 years) that mining will still be in demand. AMD is strapped for cash and have limited engineering and financial resources to pull something like this off.
That's the good thing about GloFlo/Samsung 14nm. Once AMD has used at least the WSA quota on Gloflo they can start using Samsung. I think 1 fab with Gloflo and 3 with Samsung are rated for 14nm production.Do we know if the chips for new PS4 are coming from GF, or if Samsung are doing those directly?
Definitely makes a big difference to how confident you can be in the volume of 480's they'll have right away - Sony will want massive stock built up before a launch and would probably get priority.
On the other hand if there delays due to that it'll help them bringing production on the desktop/mobile stuff up to full speed later on, so whatever
Whatever, I definitely don't think AMD will want to wait long to build up excess 480 stock. The 1060/70 are coming soon enough, and they'll want as long as they can possibly get uncontested.
with how much amd messed up 200 series launch and supplies/stock issues, they are that incompetent at times
Plus Newegg will probably do lame shit like making it only available for purchase in overpriced bundles.
I thought mining had moved on to ASICs/FPGAs? What happened?