@AzatiS said:
None can beat Playstation 3 on that.
... Epic on paper , disaster in action.
Programming SPE is similar to programming AMD ACE units.
The major differences are;
1. AMD's 8 ACE units has one big (2 MB for PS4 Pro) and fast unified L2 cache while 7 SPEs has 256 KB separate local memory pool per SPE. Sharing data with AMD GPU's L2 cache is easier.
2. AMD's L2 cache has both manual and automatic cache mode while SPE's local memory pool only has manual mode.
3. Each AMD CU has local memory data storage (LDS) and L1 instruction/ data cache. SPE only has 256 KB separate local memory pool as mentioned in point 1.
--------
On NVIDIA Maxwell/Pascal, tiling render to L2 cache is automatic i.e. NVIDIA's "smart" driver converts normal immediate mode rendering into tile cache render. NVIDIA's tiling render to L2 cache is applicable for both Compute Engine and Pixel Engine while current AMD GPUs only has Compute Engine to L2 cache feature.
Vega has both Compute Engine and Pixel Engine connected to L2 cache just like NVIDIA's Maxwell/Pascal.
Log in to comment