AMD’s Radeon Expertise Group (RTG) held a Reddit AMA session yesterday and answered quite a few questions on their upcoming Vega GPU. These of you hoping for a fast launch are going to be disillusioned. Whereas the corporate demoed and mentioned its Vega Frontier Version this week at its financial analyst day, the corporate won’t be launching a client Vega GPU at Computex or in June. The Frontier Version, for professionals and scientists, will completely run video games. However Raja recommends fairly strongly that buyers not purchase one in the event that they’re searching for any form of worth on the acquisition (because it occurs, we agree, since skilled playing cards are vastly dearer than their client counterparts).
Right here’s that individual part of the AMA:
We’ll be exhibiting Radeon RX Vega off at Computex, however it received’t be on retailer cabinets that week. We all know how keen you’re to get your palms on Radeon RX Vega, and we’re working extraordinarily arduous to carry you a graphics card that you simply’ll be extremely proud to personal. Growing merchandise with billions of transistors and forward-thinking structure is extraordinarily tough — however extraordinarily rewarding — work. And a few of Vega’s options, like our Excessive Bandwidth Cache Controller, HBM2, Fast-Packed Math, or the brand new geometry pipeline, have the potential to essentially break new floor and basically enhance sport growth…
On HBM2, we’re successfully placing a know-how that’s been restricted to tremendous costly, out-of-reach GPUs right into a client product. Proper now solely insanely priced graphics playing cards from our rivals that aren’t inside attain of any gamer or client make use of it. We wish to carry all of that goodness to you. And that’s not simple! It’s not like you possibly can run all the way down to the nook retailer to get HBM2. The excellent news is that in contrast to HBM1, HBM2 is obtainable from a number of reminiscence distributors – together with Samsung and Hynix – and manufacturing is ramping to satisfy the extent of demand that we imagine Radeon Vega merchandise will see out there.
For the previous few months, I’ve been questioning if HBM2 has run into extra issues than initially anticipated. When AMD launched HBM, it gave the impression to be taking a management place on reminiscence know-how, a lot because it did with its adoption of GDDR4 and GDDR5 (Nvidia skipped GDDR4 and didn’t undertake GDDR5 till after AMD had executed so). On the time, it appeared that AMD can be the one firm to area HBM, however that Nvidia and AMD would each use HBM2 for his or her subsequent era of 14nm graphics playing cards. As a substitute, we noticed Nvidia go for GDDR5X, whereas AMD had no high-end resolution in any respect.
Quick ahead to 2017, and we’ve obtained AMD nonetheless caught on GDDR5, GDDR6 waiting in the wings, and no signal of HBM2 on anybody’s client graphics playing cards. Nvidia has rolled HBM2 out on its highest-end Quadro line, however these GPUs nonetheless price hundreds of greater than your typical client components. No one is pushing HBM2 to client but, and Raja’s particular willingness to name out HBM2 because the tech that’s been tough to get into client components as “not simple,” appears to spotlight what the hold-up is.
A number of years in the past, it seemed as if HBM and HBM2 would turn out to be the de facto substitute for high-end graphics playing cards. Decrease-end playing cards would nonetheless use GDDR5, however even this is able to finally turn out to be much less widespread, as RAM loadouts rose above 4GB for lower-end playing cards. That forecast seems to be a lot much less sure now. If AMD and Nvidia can’t get HBM2’s prices beneath management, it’ll stay a fringe know-how, restricted to the highest-end SKUs. As soon as GDDR6 is broadly obtainable, each corporations would possibly transfer away from it altogether.
After all, it’s nonetheless attainable that these are rising pains and that HBM2 will kind itself out, however the longer it takes to get the into market, the much less probably that appears. AMD is already engaged on follow-up designs to Vega and Ryzen, and acknowledge throughout its FAD this week that it’s concurrently constructing the follow-up components to these follow-up components on the similar time. HBM2 merely hasn’t emerged because the inexpensive next-generation know-how we have been hoping it might be, and I think — although once more, it is a suspicion, not a declare — that HBM2 is the explanation why Vega’s client launch is operating late.*
* — AMD would undoubtedly quibble with whether or not Vega is late, for the reason that chip will ship in an expert card inside the first half of 2017, however let’s be real looking right here. Shopper Vega was anticipated in-market by Q2 2017, and AMD nonetheless isn’t saying when the buyer playing cards will launch.