JEMC said:
The problem is that we don't know how big of an improvement will Lovelace be to Ampere in RT so, even if AMD manages to double or triple its performance, things could end in the same awkward situation. I think (and I know that it's easy to say something like this for someone that doesn't know shit), that they should double thei RT hardware per CU. With RDNA2 they have 1 RT core per CU, and that has proved to not be enough. They should increase that to 2 RT cores per CU, plus the extra CUs from the new chiplet design. Pack in the extra refinements and improvement that come from a second iteration, and then we would be onto something. |
Yea it's also quite hard to tell where the bottleneck is for AMD. The main difference at a high level between Nvidia and AMD's implementation is that Nvidia's RT cores does both Intersections and BVH traversal where as RDNA 2's Ray Accelerators only does Intersections and leaves the BVH traversal to the CUs. Now the main benefit with leaving it up to the CUs is that on consoles, developers can fine tune it but on PC as both Direct X Ray Tracing and Vulkan Ray Tracing doesn't allow such fine tuning as of yet, there doesn't seem to be any benefit. But no one actually knows what benefit it would even have even if the PC Apis would allow it since even on consoles, you have ps5/series X which are similar to 2070 Super and 2080 performance in Raster but in heavy RT games, they are more like 2060 Super even though consoles have low level advantages.
But regardless of which, I think the next jump for PC hardware is gonna be one fap worthy jump. Lets just hope that the price doesn't increase by much and the miners/scalpers won't interfere to this extent.
PC Specs: CPU: 7800X3D || GPU: Strix 4090 || RAM: 32GB DDR5 6000 || Main SSD: WD 2TB SN850