Hes right about one thing though, if anything saves AMD its going to be the consoles, as they just dont have the PC GPU market share to push the devs to code for it on PC. There only hope is that the consoles push the devs to use the new features, and that has a trickle down effect to the PC ports.
That is normally how competition works, if you got 2 companies fighting to sell, they usually fight with prices, thus keeping costs lower.Remember how there was a large contingent on here insistent that AMD helps lower prices?
Thanks to the inflated Vega prices, all of a sudden Nvidia now has "issues" and has to raise prices.
In reality, AMD GPU shortages during launches are so bad that retailers price the GPUs really high, and then they realize that people will still actually buy them at those levels and next year, we get higher priced GPUs. Really, you can make a case for AMD helping to push up GPU prices a ton by showing the prices people will pay for subpar performance.
After the last 3 high end launches, there really isn't anything positive that happens at the initial launch of a high end AMD product.
I think the success of Raven Ridge in low budget systems, utilizing Vega features as part of the APU, will help drive developers to adopt performance enhancing features of the architecture.
If millions and millions of systems are added that run Zen + an RX550-560 equivalent GPU as part of the APU exist, that's a whole market unto itself that's worth looking into. Especially since those systems will benefit the most from squeezing every bit out of the Vega arch.
True of course, as i suspect the next consoles will likely be some form of Raven ridge.
But i still dont see how thats going to help out current vega users in time to have a meaningful boost in performance before vega is outdated and volta/navi are the cards of the day.
I think most people generally keep a video card for 2 years minimum, honestly 3-4 years sounds more likely even for gamers. That's plenty of time to enjoy the benefits.
Additionally, if the performance improves relative to Pascal with newer games in the future, that will keep the resale value up as well.
Eh, we'll see.LOL, resale value is going to be non existent after volta releases. This is the elephant in the room all the AMD supporters seem keen on ignoring, AMD released this so late they have no time to make improvements its now or never. When volta brings 1070/1080 performance down into the xx60 price range, which they will, then you are going to be able to buy a card that outperforms vega for under $300. I dont see that in any way shape or form helping Vegas resale value, performance enhancements or not.
Eh, we'll see.
RAM pricing is cocking up the whole market, so MSRP's will probably creep up noticeably from the GTX1060's launch prices. I wouldn't be surprised at a $275-300 baseline "MSRP" for the XX60 with it closing in on $350-400 for aftermarket parts.
This had me laughing pretty hard. So you're insinuating that everyone waiting for Vega were not playing games? Every time I read that I can't help but imagine a 5 year old's voice stomping his/her feet. I'm pretty sure most people probably had a video card to do something with.
I think most people generally keep a video card for 2 years minimum, honestly 3-4 years sounds more likely even for gamers. That's plenty of time to enjoy the benefits.
Something tells me AMD is not interested in repeating Hawaii. People are still rocking their 290/290X because of how well it aged. Skipping 300/Fury/400/500 cards along the way.
AMD needs to give its user base a reason to upgrade too.
I wouldn't be surprised if AMD takes a page from NV and Vega's refresh gets an uptick in performance. Something that won't trickle down to current Vega owners. It creates an upgrade path while using similar hardware and thus revenue they desperately need.
Really, and it saddens me to say this because i hate mining. Its all going to come down to how well volta mines.
Something tells me AMD is not interested in repeating Hawaii. People are still rocking their 290/290X because of how well it aged. Skipping 300/Fury/400/500 cards along the way.
AMD needs to give its user base a reason to upgrade too.
I'm sure the primitive shader and new fast path for geometry isn't out yet because it's really, really, really hard to program a smart driver / compiler to take instruction streams from the older programming model and re-jigger them to not only work in the new fast path, but to have no regression cases and to have a substantial number of cases where there is noticeable speedup. Probably less than 1% of all programmers in the world could even come close to knowing how to write code like that. This is really hardcore computer science and engineering stuff. So no doubt its taking a long time because its really hard and AMD doesn't have a massive high end dev team (but certainly a respectable one).
And use 450w for a 96 rop part?The problem is, AMD should and would have known that and betting on such features that require hardcore software / CompSci investment in itself is a dumb choice. AMD doesn't have the capacity or market share to bet on software features. GCN obviously has the 4 shader engine limit or else using the 3 billion transistors for more engines (traditional hardware) instead of clock speed and difficult to program for specialized hardware would have been the obvious much more sane choice.
Hot Chips 2017: A Closer Look At Google's TPU v2Memory bandwidth ended up being the primary limitation to the TPU design -- if Google had outfitted the processor with GDDR5, it could have ended up with far more performance per watt. The company addressed the memory deficiency with HBM memory on the second-generation Cloud TPUs.
If anything its about time they change to something far more smart and less brutal.
What you are writing just says the same as i wrote as i can tellExcept they are already "brutal" in shader count department and get consistently beaten by 64 ROP parts with far less shaders and buried by 96 ROP parts @2Ghz clocks that AMD parts can't touch. I think in the end disparity in R&D budget shines through no matter if going "smart" or brute force route. There is no doubt that big Volta will have >64 ROPs, cause NV knows you can't beat scaling of 50% more.
This had me laughing pretty hard. So you're insinuating that everyone waiting for Vega were not playing games? Every time I read that I can't help but imagine a 5 year old's voice stomping his/her feet. I'm pretty sure most people probably had a video card to do something with.