AMD Reassembles the Radeon Technologies Group
anandtech.com
anandtech.com
AMD will likely never compete with Nvidia again in high performance unless they outsmart them somehow. Their next architecture, Navi, could be a very smart idea but I wouldn't bet much on it. It will also not launch before 2019. Their current Vega architecture does not seem to have what it takes to compete against Pascal, even with scheduled process improvements.
Let's not forget that AMD is not dependant on having the top end gaming performance [1]. They can make money with their GPU tech in lots of different ways and they are in fact doing this quite well. So while we don't have to worry about AMD's viability as a GPU company, we should expect the market to become stale from a performance competition perspective. Which is what has driven the market up to around 2016.
[1]:
In fact their 2013 Radeon R9 290X, their last GPU that was the highest gaming performance in the market by most measures, performed very poorly because of a cryptocurrency mining bubble (gamers couldn't buy one at first, and cheap second hand 290Xes a few months later destroyed AMD's sales) at the time and simply because the card did not have Nvidia's brand recognition. You have to wonder whether they even want to compete in this segment if they were even able to.
Nvidia has huge advantages in their architecture, resources and most importantly in their game driver team (both explicitly and implicitly because game devs use Nvidia cards). Those aren't going to disappear any time soon.
AMD will compete the way they always have: with a lower gross margin. Nvidia's 60% gross margin vs AMD's 35% means that you get more transistors per dollar buying AMD. Nvidia uses their transistors more efficiently, so there is always trade-offs and room for AMD.
And they've taken some really smart moves to exploit that. Shifting the industry away from DX11 towards DX12 and Vulkan has significantly eroded the advantage that Nvidia gets from their driver group.
I also see ROCm and open source Linux drivers as a good way of eroding Nvidia's CUDA dominance.
I'm watching, with interest, what happens with tether over the coming weeks and months. The tether company (basically bitfinex) have issued almost a billion dollars in tether tokens this year so far, around 2 billion in total. It seems that they get issued and used by bitfinex to buy bitcoin on bitfinex whenever the price dives. The latest printing was just yesterday.
I'm not willing to say whether this will explode, or soon, or what effect it might have when it does, but it certainly looks as if someone is printing dollar tokens out of thin air to prop up BTC. If it causes a BTC crash, it'll trigger an all-round crash of the cryptocurrency space, and this may finally reduce pressure on graphics cards.
If true, it would represent around 10% of the fiat that has ever been put into cryptocurrency, going in by that one route, in three weeks, during a bear market. Does that seem likely?
Isn't that a good thing from AMD's perspective? "People can't get our GPUs because they're sold out everywhere" seems like a nice problem to have.
But it risks them having much of a position in the gaming market, long term. And right now whilst the mining market is lucrative, I would be trying to cater to the gaming market too, for longer term stability.
AMD's cards are all sold out all the time - it seems they cannot produce enough to meet the market demand. There are rumors of them using a 12nm process better tailored for high power components like GPUs (their current 14nm process is apparently tailored for smartphone chips), but they don't need to. They will sell everything they can make anyway.
Nvidia basically also sells everything they have - Nvidia cards in general mine worse but are still profitable if you can't get AMD. Moreover, AMD is barely competitive with their current offerings (especially in performance per watt), so even without mining they have no compelling reason to start pushing new GPUs.
There's no incentive at the moment to release new GPUs. Hell, Nvidia and especially AMD could be wise to take some time and restrategize to specifically start catering to the new cryptocurrency market. Find ways to produce more rather than faster cards.
It's a stale market, and will likely continue to be so in the near future.
Not only do consumers get hurt but also PC Makers would need to pay the higher inflated prices.
This is 100% a supply issue. The miners want video cards but the manufacturers are afraid the whole world of miners will burst and then they have way to much supply.
Consumers are already hurting from lack of supply. If they could reinvest these profits in speeding along increases in production capacity, seems like it would generally improve the market for everyone but the PC makers who are getting underpriced hardware.
I think production volume will be more important than hardware performance in terms of how successful AMD is in 2018. If they can supply enough mid-tier cards to meet unsatisfied demand, they could see huge growth.
I can't imagine this really having a palpable effect. Time will tell though.
> Does anything think this goes beyond simple corporate politics?
The new RTG leaders changing the organization to better fit their wants, needs and strategy? I mean, I guess its basically corporate politics because that's all executive level positions really do.
I doubt that AMD will very strongly change their future strategy. Its clear that AMD needs strong investments into GPGPU compute and needs to get back into the supercomputer-level accelerator game to survive in the future.
The 480 and 580 series were just not very competitive with nVidia for graphics performance (mining is a different matter, of course) and Vega turned out to be too low volume and expensive to make much of an impact AFAICT.
Ryzen and EPYC look like a real leap forward for AMD. I hope we see something awesome and (more importantly) deliverable come out of Radeon soon.
???
By nearly every account, the 480 was on par with or slightly behind the 1060 6GB at launch, and now the 480/580 is on par with or slightly faster than the 1060.
Shortly afterwards the prices started to rocket.