For the previous six months, the patron GPU market has grappled with three fundamental questions: When would GPUs be inexpensive, when would Nvidia launch a refresh to its standard Pascal structure, and when would AMD launch new elements to hopefully higher compete with Nvidia? To date this yr, we’ve gotten decrease GPU costs , an Nvidia launch occasion that may or may not occur this fall (rumors on timing and plans and doable stock hold-over are operating rampant), and from AMD’s nook… not a lot in any respect. Publicly, the one factor the corporate has stated is that it plans to launch a 7nm Vega refresh for the machine studying market later this yr.
However there’s a rumor from Chiphell person WJM47196, who has a reasonably good monitor report on these kinds of bulletins, that’s making the rounds on what AMD may be planning for this yr — and it’ll most likely increase some eyebrows.
First off, there’s a brand new Polaris household supposedly being prepped for a This fall 2018 introduction. Constructed on the identical 12nm course of enchancment as AMD’s second-generation Ryzen CPUs, it might provide a 15 p.c efficiency enchancment over present playing cards.
There are solely two methods for AMD to ship this sort of efficiency enchancment, and solely certainly one of them is sensible. To enhance GPU efficiency you can also make a GPU wider or smaller whereas protecting the structure the identical, or you may enhance the GPU’s clock pace. With architectural modifications unlikely, this implies that AMD would as a substitute merely attempt to ratchet clocks larger. And who is aware of? On condition that Polaris was the primary 14nm discrete GPU GlobalFoundries had ever constructed, it’s doable that AMD discovered some low-hanging fruit that may permit it to hit larger clocks. The bigger drawback is that whereas a 15 p.c soar would give AMD a lot stronger footing towards Nvidia as we speak, it may not evaluate properly towards any next-generation Nvidia launches — at the very least not with out value cuts.
Subsequent up, Navi. There’s a relatively confused suggestion that Navi can be each a mainstream and high-end half arriving someday within the 2019 timeframe, and that it’s going to debut within the price range phase first earlier than ultimately launching as a high-end, HBM2 geared up half someday “a lot later.” The recommended timeframe is:
This fall 2018: Polaris 30 (efficiency up 15 p.c).
H1 2019: Navi 10 (price range half, and timing on this introduction is unclear, with further reference to a Q1 launch)
H2 2020: A brand new, high-end Navi half, as a “true” successor to Vega
That is relatively muddled, and I’m undecided in any respect how a lot weight I’d placed on it. AMD has been extensively reported to be engaged on Navi for Sony’s PlayStation 5 and we all know the corporate has usually aligned its console and non-console launches. The concept that the work AMD is doing for Sony might result in a price range GPU constructed on an early 7nm course of as a pipe-cleaner? Not loopy. Equally, the concept AMD would port Polaris to 12nm and spin it for a little bit of further headroom isn’t nuts, both.
If we needed to make a guess, right here’s the place we’d land. The Navi 10 chip, if it exists, is a customized 7nm GPU constructed for a buyer like Apple. This chip offers AMD a greater foot into ultra-low-power markets, the place its comparatively power-hungry GCN structure can use all the method node enhancements it might get. The overhauled Polaris lineup, in the meantime, offers AMD a 15 p.c efficiency kick in desktop, the place it may be mixed with value cuts to spice up AMD’s total place. This may be additional strengthened if Nvidia focuses on refreshing its top-end GPUs first, as the corporate is more likely to do. It wouldn’t be uncommon for the introduction of recent midrange playing cards to lag the launch of a GTX 1180/1170 for some months.
The concept that Navi is a single GPU design that stretches throughout each GDDR6 and HBM2 is just a little odd, and sure displays the truth that Navi is getting used to consult with a brand new structure relatively than a single card. There’s treasured little motive why AMD would try and take a single GPU throughout each reminiscence requirements when the prices of doing so are so excessive — the complete GPU reminiscence subsystem and reminiscence controller each should be rearchitected when shifting from GDDR6 to HBM2 or vice-versa. This is probably not the one motive why Nvidia has stored its non-Pascal GPUs nearly completely within the skilled house, but it surely’s undoubtedly a part of it. However the concept it might take AMD till Q3/This fall 2020 to exchange Vega had higher be mistaken.
It took AMD ~26 months to maneuver from Fury X to Vega. On the time, this was the longest high-end refresh cycle in historical past, although Nvidia has truly taken again that report. If the GTX 1180 launches in August, it’ll have taken Nvidia 27 months to tug off a high-end refresh. However Vega is already nearly a yr outdated, it didn’t cleanly put AMD again in competitors with Nvidia on the highest degree, it isn’t going to face up properly towards the GTX 1180 or hypothetical 1180 Ti with out value cuts to reposition it (not except Nvidia merely raises its personal costs, which, hey, it might do), and the concept of leaning on it as a high-end resolution for one more two years after an Nvidia refresh is a poor one. It could successfully suggest that AMD, which is already working a GPU cycle behind Nvidia, could be prepared to take a seat out on the excessive finish completely for one more two-year interval.
However as we’ve stated, this can be a rumor — and a relatively disjointed rumor at that. Take it with a small mountain of the white stuff.
Now learn: PCMag’s Greatest Graphics Playing cards of 2018