@tormentos said:
So not even winning lemmings drop the secret sauce..lol
PS4 Pro's ROPS doesn't have X1X's 2MB rendering cache.
1MB L2 cache in Hawaii GPU has ~1TB/s memory bandwidth at 1Ghz clock speed, but stupid AMD didn't connect it to ROPS, hence gaming workloads are largely bound by non-DCC external memory bandwidth.
Hawaii is fine with GpGPU (e.g. crypto currency) when TMUs are used with 1MB L2 cache. Polaris 10 has 2MB L2 cache and it's very competitive with GP104 in crypto currency GpGPU workloads, but graphics workloads needs ROPS which is not connect to hyper fast L2 cache.
If X1X ROPS's 2MB render cache has ~1TB/s memory bandwidth and they used micro-tile cache render methods, the potential X1X's advantage is 4X over PS4 Pro's ROPS with 218 GB/s memory bandwidth.
Most AMD examples are in compute shader path which is connected to L2 cache which narrows X1X's advantage.
TrueSky is picking on PS4 Pro's known weakness.
Log in to comment