AMD is clearly stating it's GCN based. So maybe they change the name, like pascal will be mostly maxwell based. But GCN 2.0 would be probably also ok.
Good point. It's possibly they'll just call it an all new GPU architecture since consumers love hearing something is new and revolutionary.
For AMD specifically, I am not worried about their upcoming GPUs in terms of performance.
I am. AMD used up a big advantage of HBM1 with its reduced memory controller and big benefit in reduced power usage. AMD also reduced DP capability on Fury X. Despite both of those moves, Fury X cannot overclock as well as the 980Ti and has inferior perf/watt. That means if AMD has an improved architecture + node shrink, NV has both of those + the move to HBM2. That to me gives a big advantage to NV. In other words either NV has to mess up with Pascal or AMD has to make up a massive gap in perf/watt to catch up.
What I am worried about is the twin devil that has tormented them in the last two major GPU release cycles; namely supply and reference design.
True, but you forgot 3 other key factors that matter as well imo:
- Price/performance
- Driver support with more AMD GE titles to combat GWs
- Overclocking - with Maxwell overclocking so well, it's not enough for AMD's cards to just match Pascal. If Pascal overclocks better in all price segments, most gamers will buy a card that's as fast at stock but overclocks better. Higher overclocking headroom also allows AIBs to be more aggressive with factory pre-overclocked SKUs; and we know how reviewers LOVE to put factory pre-overclocked NV cards against stock/reference AMD cards.
I'd prefer AMD to wait a little bit and not fall for the bait of rushing as fast as possible and get the supply in order(without waiting too long of course). Take that extra time and get really good reference designs across the board. Not just heat/noise but also stuff like coil whine, QC issues in general.
This one is tricky. AMD waited way too long to release 380X/390/390X and Fiji which gave NV a big window to capture so many consumers because AMD didn't even show up.
Early adopters have shown that they are willing to pay premiums for the latest products -- 7950/7970, GTX780, 750/750Ti, 960, etc.
Look at the Titan and Titan X. NV released 780Ghz and 980Ti very shortly after their releases and the after-market cards even had faster performance out of the box for $350 less. I think what hurt 7970 more was not $550 initial price but the loud reference blower. Had 7950/7970 showed up as an after-market pre-overclocked card from day 1, NV's 670/680 wouldn't have done much damage at all. AMD would have just lowered prices on 7950/7970 and still be ahead.
If AMD can execute the same HD7000 strategy but with proper coolers, that would be better than launching after nV. I think for next generation launching first is critical esp. for AMD. Think about it, imagine how a $550 Fury and $650 Fury X would have looked in March 2015 against the $999 Titan X and how those cards looked after their launch after the 980Ti showed up?
Launching first will create a big halo factor due to perf/watt (marketing) and latest features. Just look at 750/750Ti that had very mediocre price/performance at launch and yet sold tens of millions of units.
We all know that getting an AMD GPU 3-4 months after launch is the best. You get the second revision and the (inevitable) major driver performance improvements. That needs to change. Most people shop on first impressions and if AMD should get more market share back, they need to be A1 from day 1.
The driver improvements in the first 12 months happen for both though. It's not just an AMD thing. Also, what you are suggesting is more complicated than that.
Imagine if AMD delayed the launch of HD7850-7970 by 3-5 months. Would they have been better off? No, because it would have meant selling older HD6000 series all that time and worse getting slaughtered by March 2012 released 670/680 cards. Similarly, look at the launch of the 780. AMD was 5-6 months behind with 290/290X and NV maximized profits while taking market share without any effort since 780 had no competition.
Honestly, I think launching first, even if you are slightly slower is better. Launching 5-6 months behind for 5-10% more performance isn't a strategy that works for AMD. It works for NV since customers who buy NV will wait.
Facebook
Twitter