A list that appeared in Geekbench’s OpenCL performance test has allowed us to identify the Radeon Pro W7600, a new AMD graphics card with which the Sunnyvale giant is aiming for the mid-range, and which will have a configuration of 16 WGPs (Workgroup Processors) and 8 GB of graphics memory.
These data may not seem like much, but since we clearly know the distribution of the RDNA3 architecture depending on the number of computing units we can establish a very complete and totally reliable breakdown of their specifications. In total, the Radeon Pro W7600 will have 32 compute units, as each WGP has two compute units.
The 32 computing units equals 2,048 shaders (64 shaders per compute unit), and will have 32 ray tracing acceleration units, 128 texturing units (four per compute unit), and 64 raster units. Its memory subsystem will consist of a 128-bit bus and 8 GB of GDDR6 memory at an unspecified speed, although it will most likely reach 18 GHz.
Otherwise it will have 32MB L3 cache, also known as infinite cache, which, unlike what we have seen in its older sisters, will not be externalized in chiplets, but will be integrated into the same chip as the GPU. Its maximum working speed is 1940MHz according to this test, a fairly low value that suggests that we could be dealing with an engineering sample.
Those 8 GB of GDDR6 memory will have the ECC (error correction) technology, something that is not present in general consumer graphics cards, and that makes it a viable option to carry out projects and investigations that require working with advanced simulations where total precision is required.
We do not have details about its possible release date and the price has not been revealed either, but it is rumored that it could be presented sometime next month, taking advantage of the SIGGRAPH celebration. Its price should be similar to that of the Radeon Pro W6600 at its launch, which would lead us to situate ourselves in the range of 500-600 euros approximately.
