Silverforce11
Lifer
- Feb 19, 2009
- 10,457
- 10
- 76
A lot of people say AMD is more forward looking / creates GPUs that do well for the future, but it seems like AMD is a company that is creating their own goddamn future. They invented Async compute, pushed Mantle/DX12/Vulkan down everyone's throat with the consoles (which wouldn't have even been possible without AMD's efforts in APUs, which began all the way back when they bought ATi), invented HBM, and now who knows what they're planning to do with multi-GPU in the future.
I am somewhat concerned for the new APIs though, what if AMD decides to do a huge arch change and suddenly GCN is left in the dust like Kepler is today given the low level nature of Vulkan and DX12?
Architecturally as long as GPUs have separate engines, such as Compute Units/ALUs, Rasterizers and DMAs, it will be fine with DX12/Vulkan. Each of these subunits form the core of rendering because we're dealing with pixels afterall and have been for decades. Iterations improve them, they become more efficient, faster, but they are still there.
The biggest limit on AMD's ability to change GCN will be to keep the ISA backwards compatible. It's required of them now due to the console compatibility being a core feature of MS & Sony's ecosystem.
And yes, you are very right. AMD is creating a future that best suits their IP. It's been years in the work and will have a few more years to go for AMD to truly prosper. But it hinges on them not self sabotaging. The RX power issue shouldn't have even been an issue.
Also, If NV & Intel are not pushing hard for small multi-die on an interposer approach, they will be caught with their pants down in a few years time. Especially on 7nm and below where difficulties with large dies will be a major restriction for single monolithic architectures.