Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : AMD, ARMH, INTC, NVDA -- Ignore unavailable to you. Want to Upgrade?

To: neolib who wrote (16094)	6/2/2016 5:04:08 AM
From: pgerassi	Respond to of 72327

What you think is most applicable to gamers. It doesn't matter whether a card has one or multiple GPUs as long as the performance is there.. And with DX12/Vulkan, it is a given that shaders and memory are shared. Trouble is that SLI takes a back seat to Crossfire through the PCIe bus. Nvidia admits that there isn't enough bandwidth in their new high speed SLI bridge to connect more than two 1070/80s together effectively. Crossfire OTOH can effectively connect 4 Radeon Fury Xs together and likely could do 4 Pro Duos as well. 1080 CF isn't going to be anywhere near that. So yes, Radeon Pro Duo is the fastest GPU card period. That is why reviewers don't include the Pro Duo in their reviews of the 1080. Nvidia knows they would lose to it. 2) We know that GCN does have an architectural advantage over Maxwell/Pascal especially in optimized DX12/Vulkan games. Which given current consoles, will be almost all of the newer games. And that is before any additional advantages from GCN4. The primitive discard accelerator is intriguing. It might make Gameworks shenanigans ineffective except to drag down Nvidiea's own GPUs. So that performance /per Watt metric could be heavily favoring the Rx480 and related Polaris based members. 3) With the GP104 being twice the size of Polaris 10, if not more, Given typical new process metrics, there should be about 3 to 4 times the number of good dies per wafer. That means the dies will be 1/3 to 1/4th the cost with a much larger percentage being being fully capable wrt the GP104. AMD may even get more margin and profit per wafer with Rx480 than Nvidia gets with 1070/1080. And AMD is likely to launch with 1 to 2 orders of magnitude more cards than the 1070/1080 did. And for risk runs like the 1070/1080 paper launch, R&D costs far outweigh marginal costs. Say Pascal R&D costs were $1 billion, at a production of 100K cards (far more than they have sold so far by all accounts) that means each card cost $10,000 for R&D. Even if Polaris cost AMD as much (which likely includes Vega and Zen APUs), 10 million cards puts the R&D portion at $100 each. Given the current prices, even if AMD gets half of the target market of 50 million (and unless Nvidia greatly reduces prices and their profits AMD will get a higher percentage) R&D costs will be under $20 per GPU. 4) Which is why AMD is giving its AIB customers 4 weeks to reduce R9 380 and up GPU card inventories. There are lots of sales for 380, 380X and 390s and even some for 390X. Due to the larger DP GPGPU capabilities, the 290/290X/390/390X may still get a higher price for 0.6-0.8TF performance. The Rx480 will be more in the 0.3-0.4TF DP range. We could get surprised though. Of course if the Polaris 11 is launched as well, the hard launch of the whole line would make your point moot. AMD would likely sell more Polaris to make GPU revenue to go up rather than down. And that makes AMD to not care, if the older generation tanks. Pete