In recent years, euphemisms artificial intelligence, Deep Learning and other magic words have not been another way to call the addition to the hardware of a new type of unit capable of performing mathematical operations with matrices to great speed. Well, against all odds, AMD has hidden a detail in its graphics processors with RDNA architecture 3 that has to do with it. are the calls WMMA instructions and they bring a very significant change, never before seen in AMD gaming graphics cards.
Those of you who are more veteran will remember how in the 90s the word multimedia resonated as much as artificial intelligence does today. The reason? The addition of SIMD units in processors of yesteryear. Today, the same thing happens with the calculation of matrices and artificial intelligence. Well, AMD is said to have finally adopted these units in its home GPUs, after four years since NVIDIA launched its first RTX with them.
A unit to operate with matrices
One of NVIDIA’s spearheads in its GPUs is the addition in the RTX of the so-called Tensor cores, which are nothing more than a type of run called systolic array. Its main asset? The ability to work with high-speed data tables. These types of mathematical structures are crucial for artificial intelligence algorithms and for being able to store large amounts of data. What is an image for example? A pixel array.
Since a matrix is nothing more than an N-dimensional vector, until now vector or SIMD units have been used to perform these operations, which can execute the same instruction in parallel with several different operands. However, while they can execute such instructions, they are much slower to do so than what we call a systolic array.
Well, with NVIDIA having their Tensor drives and Intel having added their XMX, which are just trade names for the same concept, the one that was missing was AMD. In any case, the absence was rather in its GPUs for personal computers and workstations. Since in the high performance computing or HPC world, their CDNA-based AMD Instinct has been with these drives for a long time.
WMMA, Matrix CPUs for the RX 7000
AMD’s name for its systolic arrays is Matrix Core Units.Who have been unpublished in RDNA and RDNA 2. Well, after several rumors in which it was said that AMD would not bet on this type of units in its GPUs, it turns out that in the LLVM repository there are references to the inclusion of WMMA instructions for GFX11this is what AMD calls the RDNA 3 architecture of its future GPUs like the RX 7000 and processors like the Phoenix
In the repository the acronyms WMMA are described as Wave Matrix Multiply Accumulate and are described as:
These operations perform matrix multiplication and addition in the form: D = A*B+C.
This recalls the structure defined by NVIDIA to describe its Tensor Cores. Moreover, as you can see, AMD’s rival company also used the WMMA concept. In addition, the adoption in AMD GPUs of this type of unit is revealed with a diagonal look at the source code.
- The resulting matrix (D) given in 32-bit or 16-bit floating point and the 3 matrices operating in 16-bit floating point or BFLOAT16 format.
- The final value of D being in 32 bit integers, and A, B and C in 4 or 8 bit integers.
So in summary: we find that AMD added WMMA instructions in RDNA, which refer to adding a unit to operate with matrices, i.e. finally AMD GPUs will have their own Tensor Cores. Which are widely used in the world of artificial intelligence and have been one of the elements