- Joined
- May 10, 2023
- Messages
- 390 (0.65/day)
- Location
- Brazil
Processor | 5950x |
---|---|
Motherboard | B550 ProArt |
Cooling | Fuma 2 |
Memory | 4x32GB 3200MHz Corsair LPX |
Video Card(s) | 2x RTX 3090 |
Display(s) | LG 42" C2 4k OLED |
Power Supply | XPG Core Reactor 850W |
Software | I use Arch btw |
It's just describing where the WMMA ops run at. Also:Tom's harsware has more that just the wmma listed.
This goes in hand with what I said. It makes use of the units found in the compute units, it's not a new unit by itself.and the AI Matrix Accelerators share many of the shader execution resources.
The article from TechSpot that I linked also mentions it:
Whereas on CDNA you have different register banks and ALUs to work on their MFMA instructions.Another significant new feature is the appearance of what AMD calls AI Matrix Accelerators.
Unlike Intel's and Nvidia's architecture, which we'll see shortly, these don't act as separate units – all matrix operations utilize the SIMD units and any such calculations (called Wave Matrix Multiply Accumulate, WMMA) will use the full bank of 64 ALUs.