Very useful info. Thanks! Also, I would to know about Cuda cores and Steam processors. They're essentially unified shaders, but I have read they cannot be compared apples to apples. I read on enthusiast websites that Nvidias shaders are larger, more complex, so they do more per cycle. So if the AMD card has 5,000 shaders, the nVidia card can achieve the same with a lot less shaders, is this correct? Someone did some conversion with Maxwell and said for every Cuda core it takes like 2.25 steam processors.
No, it doesn't.
Take Radeon Pro 555X with 768 GCN cores, 128 bit memory bus of GDDR5, and compare it to similarly clocked(or - decklocked) GTX 1050 with 768 CUDA cores, 128 bit memory bus of GDDR5. Both will perform the same - because ALUs are ALUs. You can even use Intel GPU, with 768 Cores, if it would exist - it would perform exactly the same as those two GPUs.
1280 GCN core chip, with 192 GB/s of memory bandwidth GPU, will be exactly the same level of performance as 1280 CUDA core GPU with 192 GB/s if both are clocked at around 1.3 GHz.
Vega Pro 20 vs Quadro P3000.
AMD GPUs with higher core counts are not fed properly, and are unable to clock high enough. Nvidia GPUs are bigger, because Nvidia increased the pipeline length(Nvidia burned quite a lot of transistors on physical design level), so the GPUs can clock higher.
AMD to get the same level of scalability as Nvidia, has to add Operand Cache to their pipeline, which will save bandwidth, register file size, power, and allow those GPUs to clock higher. This will also allow AMD to scale beyond 4 Shader/Geometry Engines, which will allow AMD to get much higher Geometry throughput, if they will also add Primitive Shaders, properly working, as a feature of Navi.
Until GP100 chip, GV100 - it was AMD which GPUs were able to do more stuff, with each cycle, because the GPUs were wider(64 Cores/256 Register File Size vs 128 cores/256 KB register file size). Nvidia has just recently got parity with AMD on the front of "width" of the GPUs, and it hasn't really made their GPUs faster in gaming, because of very simple reason. ALUs are most important here.
And some of those early RTX adopters aren't really impressed with their ridiculously expensive cards
Hopefully, developers will also support AMD's ProRender. It looks impressive. Too bad, it took Nvidia to look into Raytracing to get developers to get serious about it. I would also like to see Freesync being a standard for all monitors rather than Nvidia's expensive G-Sync. AMD is making some good moves the past few years with open source technologies. I'm really hoping 7nm Vega and Navi compete with Nvidia. Hopefully in the mobile market as well, since I don't think Nvidia will be able to come out with anything like their 10 series for RTX that will have decent TDP.
Do not count on this. Want to know why, which is also apparent on this very forum?
People want AMD to be competitive with Nvidia, not because they want to buy their products. They want to buy Nvidia GPUs cheaper!
That is very reason why I genuinely hate desktop, and professional market. Pure stupidity.
