Lordhawkwind wrote:AMD and Hynix co-developed HBM and probably in lieu of taking a licensing fee AMD would have negotiated preferential terms for the production of most, if not all, of Hynix's HBM production. This obviously left Nvidia in a bit of a hole and that's why they worked closely with Micron on GDDRX5/6 and as it turned out this was a very wise move. Whilst Hynix have struggled to get mass production in gear Samsung has stepped into the breach no doubt paying Hynix a licensing fee for the privilege but it's an arrangement that suits both parties.
As we've seen, GDDR5(X)/6 have all been (will be in the case of v.6) more than needed to feed even the highest performance cards for gaming. They're technologies that the big memory makers are already very good at producing in volume, and Nvidia (and AMD) are very good at implementing. Very safe bet here.
HBM, while a proven technology in terms of deliverable performance, is rather new. Nvidia's choice to limit HBM use to their big-chip compute product line makes a whole lot of sense here, and AMD's gamble- which could certainly have paid off- appears foolish.
Lordhawkwind wrote:Both companies want volume business and at the moment that's to AMD's advantage as they use HBM for their consumer and HPC cards. As Nvidia only use it for their HPC cards they can't provide the volume required and this means they will be in line behind AMD and most likely paying a high cost in excess of what AMD are charged per module. Nvidia probably don't mind because their profit margins in this market are huge.
Sure, volume is good. The challenge here is that AMD simply cannot make very many of these cards; not only is production of large GPUs hard, but AMD is relying on a technology that is apparently unproven, and they're doing it with their consumer GPUs.
Would you argue that HBM wouldn't have been available for anyone to use if AMD hadn't committed to use it for their consumer products?
Lordhawkwind wrote:I also don't buy all the miners are buying Vega cards because the overall economics don't add up. Who I can see buying Vega even at $699 are content creators and small studios who can pick up each RX Vega at $300 less than a Vega FE and probably saving more on a Titan purchase. If a small company needs 20 cards that's a saving of at least $6,000 thank you very much, not exactly small change. Maybe this is why Nvidia suddenly released this 'miraculous' driver for Titan in the last month or so that has increased it's compute performance. Whilst Vega is not going to hurt Nvidia too badly in the 1080ti gaming market I bet it's beginning to erode Nvidia's share of the 'prosumer' market which overall is far more profitable for both companies.
Miners don't add up. They're definitely buying those cards that they can and they're tweaking them for profitability, despite the initial numbers not looking so great.
I see others buying them to tinker with, a few actually got cards at MSRP, those that are invested in FreeSync, and a few zealots that will buy AMD no matter how bad the numbers stack up. But mostly, I don't see regular gamers buying Vega.
As for the pro markets... yeah. You'd have to provide sourced numbers to make any kind of compelling argument, showing that both the volume is there and that Vega is actually a good buy for a commanding percentage of those markets, and hell, that they can actually buy them in the first place.
And no, $6000 isn't even pocket change for a small business. You and me, sure. But even a small business would spend that $6000 if it increased productivity.