Yeah, there comes pretty good time for GPU fans .
About AMD, lets look at what AMD staff says about Polaris.
First things first: Efficiency. So we look at die shrink that brings 50% better power consumption. So a GPU that had 250W TDP will have 125W TDP, thats to shrink itself. But, there is another bit of information...So not only die shrink brings efficiency, but architecture itself. We are looking at 65% better efficiency overall for the GPUs. Lets think about it for a second. 200W R9 280X on 28 nm from TSMC, but with new architecture would draw around 170W at max. Without the shrink. After shrink it would be 85W, if we take 50% lower power consumption, and not account best case scenario.
Here is a link: http://www.overclock3d.net/articles/gpu_displays/amd_has_two_polaris_gpus_coming_this_year/1
Now density: https://www.semiwiki.com/forum/content/3884-who-will-lead-10nm.html On the bottom there is table which shows that density ration between TSMC 28 and GloFo/Samsung 14 nm FF is not 2x. It is 2.2x. So the same R9 280X without advancements in architecture on 14nm would be... 144mm2 die(60% smaller die). And we do not know how will new architecture affect the die sizes, it can be optimized for it, from ground up, and it looks like that is the case.
Lets get back for a second to previous rumor about the die size of one of GPUs from AMD: 232mm2. If Fiji would be ported to 14 nm without any advancements in architecture it would be 250mm2 die. Coincidence? I may be wrong here, of course, but what better way to bring VR into much lower price/performance brackets than by bringing that kind of performance here?
Third thing: Performance. If Mahigan from Anandtech forum is right it looks like we might be pretty surprised with performance of Polaris GPUs. http://forums.anandtech.com/showpost.php?p=38026437&postcount=357
http://forums.anandtech.com/showpost.php?p=38026062&postcount=355
I will not be surprised if my calculations and best case scenario are true, and we are looking at Titan X performance at a fraction of power consumption and cost.
Unfortunatelly I cannot bring anything for Pascal, because... there is no rumors. There is no silicon, we do not even know if the new arch from Nvidia will bring improved Asynchronous Compute(second engine) or even Hardware Scheduling. If the slides from Nvidia are true, and Pascal is only Maxwell on 14 nm with FP64 - the chances for both of them, which would make GIGANTIC difference, are almost null.
What purpose would that serve for the public?Ok, I'll buy it. This 232 mm2 GPU (most likely Polaris 11) probably has performance better than Hawaii with lower power consumption. This would be a suitable GPU for the mac pro, but it doesn't make me as excited if a > 300 mm2 GPU was going in there. Then you would see something that would hands down beat Fiji.
If all this is true and we are getting ~100mm2 and 232 mm2 GPUs from AMD this year then its pretty clear what their strategy is. Get to market first with 14/16 nm GPUs and make sure you own the GPU space that retails for < $400. Whether this succeeds or not probably depends on how long it takes Nvidia to get a consumer version of Pascal out there, how much they charge for it and how fast it is.
Apple was all "OpenGl for the Win!" until they fell laughably behind, at which point it became "OpenCl for the Win" and we all know that they are still laughably behind there. So, instead of competing, they switched courts again.
Ok, I'll buy it. This 232 mm2 GPU (most likely Polaris 11) probably has performance better than Hawaii with lower power consumption. This would be a suitable GPU for the mac pro, but it doesn't make me as excited if a > 300 mm2 GPU was going in there. Then you would see something that would hands down beat Fiji.
I have one another info. It looks like Polaris 10 and 11 are GDDR parts.
Keeping in mind that Apple often gets early access to components what are the odds of any of this showing up March 15?
Or are we still stuck with WWDC on June 6 as a potential release date, which would be fairly horrendous.
Keeping in mind that Apple often gets early access to components what are the odds of any of this showing up March 15?
Or are we still stuck with WWDC on June 6 as a potential release date, which would be fairly horrendous.
If Polaris 10 and 11 are not HBM and top out at 232 mm2 then I think it points to another high end GPU coming from AMD.
Despite the shrink and some architectural gains in Polaris, its tough to imagine a 14 nm 232 mm2 GPU beating out a 28 nm 600 mm2 GPU (Fiji). The performance gains should be good but not that good.
Perhaps this high end chip is the rumored Greenland/Vega. Its tough to imagine AMD not being the first company with a HBM2 chip out the door. The question of course is when would such a chip be ready. Rumors also point to HBM2 not being quite ready yet so its tough to tell.
I think WWDC is probably the earliest possible. Intel's Broadwell-EP isn't supposed to launch until early June and AMD's GPUs are rumored for the summer.
HBM2 already is in mass production, only at Samsung. Micron will start Mass production in April if I remember correctly, and GDDR5X which will go to GP104 will start in summer. So no High-End HBM2 chips before late Q3, the same goes for every chip that will use GDDR5X.If Polaris 10 and 11 are not HBM and top out at 232 mm2 then I think it points to another high end GPU coming from AMD.
Despite the shrink and some architectural gains in Polaris, its tough to imagine a 14 nm 232 mm2 GPU beating out a 28 nm 600 mm2 GPU (Fiji). The performance gains should be good but not that good.
Perhaps this high end chip is the rumored Greenland/Vega. Its tough to imagine AMD not being the first company with a HBM2 chip out the door. The question of course is when would such a chip be ready. Rumors also point to HBM2 not being quite ready yet so its tough to tell.
I think WWDC is probably the earliest possible. Intel's Broadwell-EP isn't supposed to launch until early June and AMD's GPUs are rumored for the summer.
Think rather about that Polaris 11 will be most likely the next D310 FirePro in Mac Pro .
D310 - 6 GB of VRAM
D510 - 8 GB of VRAM
D710 - 16 GB of VRAM.
Makes sense?
I meant, all that is coming but how much of it is available just now?
BDW-EP is to be announced yet this quarter, Polaris/Vega might be coming after the summer, the rest is available of course but the bulk of the upgrade (CPU and GPU) are still not here yet. I wouldn't count on a new nMP rolling out before late Q3 or even Q4.
Hope I'm wrong though.
Even more, TB3 displays are not abundant, much on the contrary. Apple hasn't provided one yet, if they will ever. Should one be counting on HSMI only, if it still remains, for display output? Or depend on adapters? Really?
Doesn't sound like Apple.