[AMD] Difference of DirectML vs ZLUDA:
DirectML: Its Microsofts backend for Machine Learning (ML) on Windows. DirectML provides GPU acceleration for common machine learning tasks across a broad range of supported hardware and drivers, including all DirectX 12-capable GPUs.
Its slow and uses a lot of VRAM Resources but it works on any AMD GPU and even IGPUs.
It can be tuned in performance by using Tools such as MS Olive and ONNX. But this requires Model conversion and a limited feature set.
GPUs Supported by DirectML: Nearly every AMD GPU and IGPU.
ZLUDA: ZLUDA lets you run unmodified CUDA applications with near-native performance on AMD GPUs.
Its like an emulation layer to get Cuda tasks working on AMD GPUs on Windows or Linux.
Its faster and uses alot less VRAM compared to DirectML.
ZLUDA is the recommended Version for anyone with a supported card.
GPUs Supported by ZLUDA in this Guide:
(gfx1035) - AMD 680M-APU
(gfx1103) - AMD 780M-APU
(gfx803) - RX 570, 580, 590 8GB,
(gfx900) - RX VEGA 56, Vega 64, (gfx906) Radeon Vii,
(gfx1012) - RX 5500, XT,
(gfx1010) - RX 5600, XT,
(gfx1010) - RX 5700, XT,
(gfx1032) - RX 6600, 6650, XT, S, M, 6800S, 6700S,
(gfx1031) - RX 6700, 6750, XT, S, M, 6800M,
(gfx1030) - RX 6800, XT,
(gfx1030) - RX 6900, 6950, XT, S, M,
(gfx1102) - RX 7600, XT, S, M,
(gfx1101) - RX 7700, XT, S, M,
(gfx1101) - RX 7800, XT, M,
(gfx1100) - RX 7900, XTX, XT, GRE, M,
(gfx1201) - RX 9070, XT,
这是官网的支持列表