Interview with Raja Kaduri

This is a brilliantly informative interview with Raja Kaduri; courtesy of PC Perspective. It just came out and has less than 50 views right now as I am posting it. It's around 12 Minutes long. Enjoy and discuss away.

https://youtu.be/4qJj1ViyyPY

It talks about both the software & hardware side of AMD. I feel it belongs more in this (driver) section than in the (hardware) section.

 

So no HBM2 on polaris? That is all i found interesting.

 

Love the talk about multi-gpu. I was worried about this aspect but I guess I will continue to build kickass systems!

 

I'm remaining hopeful for the future of multi-GPU after listening to the discussion. If the whole VR thing kicks off surely this will force better integration into game engines currently on the market.

 

Translation:

No big Polaris in 2016.

 

I think you can easily read between lines that AMD is planning to release multiple multigpu systems, even from smaller dies. In order to get multigpu really adopted by developers, is to make it mainstream.

Let's be realistic, if this new process tech gives us 2.5x performance on biggest dies. How far is it gonna carry us. 60 fps 4k; 5k? Yeah sure for current genre of games, but future games will have even more demanding graphics. So I think aiming for multigpu mainstreaming is a nice goal, and just like it happened on cpu's, we need it happening on gpu's aswell. Performance requirements are raising so fast compared how "slowly" new process nodes evolve.

 

Translation: HBM 2.0 is not ready

 

polaris and vega re not the same hing

 

As a current CF user this interview sounds both good and bad. Good, because investing in multi-GPU for DX12/Vulkan is absolutely great, especially when the APIs are this young since it could lead to mGPU becoming more common in engines/games. On the other hand, actual CF support in actual, existing, DX9/10/11 games has most definitely been quite poor for over a year now and this focus on DX12/Vulkan may very well mean it won't be getting any amazing attention in the future while DX11 is still relevant either.

 

Are you going to buy and use a multi-GPU setup on your PC after you saw the support we got until now from GPU makers and game devs?

CFX 4 GB VRAM 14/16 nm GPUs in 2016?

No thanks.

In DX12 era GPU makers can sell multi-GPU cards or promise amazing performance from CFX or SLI and point finger to game devs for all the multi-card support (or lack of it) we will get.

Next time i buy it will be a new 14/16 nm single GPU with a bare minimum of 8 GB VRAM (16 GB better).

 

Oh I won't be getting another 4GB card for sure.

Naturally transition period will take time. But all developers know that if they are gonna hit 8K resolution in couple of years, there won't be a single gpu solution that can do that. Besides, if at the point when you have no single gpu cards at market anymore, developers have no other options than invest in multigpu support. This will take time, but this needs to happen. Process nodes just don't advance fast enough.

 

Whatever he mentioned was common sense actually. In smaller process nodes big dies are even more expensive. I bet that NVIDIA is pushing something similar too.

 

Can we expect GDDR5 X in some polaris cards? if we get double bandwidth of a r9 390X so 384Gb/s x 2 = 768Gb/s is not that bad, just in the middle of HBM1 and HBM2 bandwidths.

 

I fear dx12/Vulkan is another wall in practise for multi GPU, because I don't have much faith many devs. putting effort in a niché market.......
I'm also far from impressed from any current implementation of dx12 which seems to favor CPU limited area's to begin with and not GPU limited area's, offers no real iq features, although that could change later on. as things, offcourse, have to mature over time.
I wonder if ROTT and Hitman (GOW and Talos(Vulkan) are wrapped?) are really developed from the ground up for dx12, you cann't go parallel if it's some kind of wrapper or post thing.

But if there come a GPU which renders every game in 5K @ 60 fps at max iq I would be happy

 

What he said in the interview is that game developers have to adjust to multiGPU like they adjusted to multi CPU, and that the node economics demand that.

 

I'm not going to be a guinea pig in this multi-gpu transition/training process of 2(3?) years for game devs in DX12.

I had enough in DX11.

 

thats kinda what i got out of it as well.

If it tracks the same way muli-core cpu did, its going ot be a looooonnnngggg time beofre its standard to take advantage of multi-gpu in that manner.

Christ, up until fairly recently there were still games being released that are single core cpu.

 

we are guinea pigs all of our lives. no escaping that

 

The reality:

Talos Vulkan: no CF/SLI;
GOW dx12:never mind;
Hitman dx12: "Hitman has stopped working and need to close" ie. no CF
(no CF either for dx11, but no one seem to bother:bang
ROTT dx12, no CF;
UE4 dx12: no CF/SLI;

But you can mix GPU's and running AOTS benchmarks the whole day :wanker:

 

That's why he said that they send the CryEngine and the cards to a lot of Universities, because it's developers that need to change. It's quite similar to multicore with CPUs.