What are you most excited for from the next generation of cards?

[FileTrekker]

With NVIDIA likely to release their next generation Pascal-based cards this year, what kind of features would you be looking for to make you consider an upgrade in the next generation, Red or Green?

Obviously smaller and more power efficient would be nice, so I can get a higher end card (960 currently, would like to get a 980 level card in but my case is a mid-tower and it won't fit, don't want a bulky case...)

I think NVIDIA have already said they're going to push boundries when it comes to the ammount of VRAM and how memory is used, but I'm not sure these are pretty critical things for the current generation of games to make an immediate upgrade worth it.

So, what standout features would you want from a card?

 

[Barbas]

I just hope they work on day one with a lot more games, to be honest. Too many unfortunate people are having to slave away in crunch time for that. Smaller cards would also be nice, obviously, because it's always nice not to cut your hands up whilst rummaging around inside the computer's guts.

 

[Elvis Starburst]

I just want the Canadian prices to drop again so I can afford one and reap its benefits ;3; A 970 should not be $500...

 

[TotalerKrieger]

We will very likely have 4K/60FPS capable GPUs on the market in early 2017. AMD and Nvidia have a "perfect storm" scenario where several leaps in performance will be introduced in a single GPU generation (at least for the flagship models).

Nvidia's GP100 (Big Pascal) and AMD's Vega 10/11 will be the first designs to use HBM2 VRAM. As I understand it, HBM technology involves stacking VRAM chips on top of each other to get much higher bandwidths. Leaked specs suggest that the AMD and Nvidia flagship models will be released with 16GB of HBM2. The next performance tier down will probably be released with 8-12GB of HBM2 VRAM but this will still be much faster than 8-12GB of GDDR5 memory. This jump in memory performance will be coupled with one of the largest die shrinks in GPU history (28nm to 16/14nm). Die shrinks allow for more transistors to be placed on the same surface area significantly increasing processing capabilities. Upping the stakes even further, DX12 and Vulkan may offer significantly improved GPU performance (10-20%) by reducing CPU overhead in games.

Based on a recent demonstration by AMD, the mid to high end cards in the Polaris and Pascal series (R9 480/490 and GTX 1070/1080) will be easily capable of 1440p/60FPS gaming (ie. 10-20% better than the GTX 980 Ti and Fury X). These cards may only have GDDR5 or GDDR5X memory as they will be released before mass production of HBM2 starts in late 2016.

 

[Sic Transit Gloria]

I'm excited for a memlogic optronic matrix which achieves sentience and takes over my computer to binge on porn.

 

[Saulkar]

when it comes to the ammount of VRAM and how memory is used, but I'm not sure these are pretty critical things for the current generation of games to make an immediate upgrade worth it.

But it would really be worth it for GPU rendering. When a 3D animation is rendering on your GPU the entire scene is loaded onto the VRAM and when you run out the render crashes. The only renderer that does not suffer from this issue is Redshift but it is far from an industry standard like VRay which does not do out of core memory.

I tried to render a single frame from this 3D animation made in Blender using my GPUs instead of my CPU and the RAM requirements jumped to 17 gigs before it crashed my Titan Blacks.
My computer has enough RAM but not my cards. I would kill for an Nvidia card with 32GB of Memory like the latest AMD Firepro card. Videocards can render animation 10-50 times faster than a CPU but the VRAM limitation is the primary bottleneck and OpenCL's kernal is not sufficient for GPU rendering at the moment! The Blender Foundation is working with AMD to change this but it is still a long ways off. Look at the feature comparison page!

https://www.blender.org/manual/render/cycles/features.html