From AMD, we learn that the company is launching its Ryzen AI Max 400 processors, known as “Gorgon Point”. Broadly speaking, the company is using the same base as the 300 series, but with a few welcome improvements, such as better RAM support and higher frequencies. Clearly, these are CPUs designed to run AI locally.
Ryzen AI Max 400: faster and with more RAM!
Three references so far:
Broadly speaking, AMD has adopted the characteristics of the Ryzen AI Max 300 in terms of number of cores, CPU and GPU architecture. The CPU uses Zen 5 cores, while theiGPU uses RDNA 3.5. Of course, we mustn’t forget to mention the presence of an XDNA2 NPU with 50 TOPs.
The largest model is the Ryzen AI Max+ 495, with 16 cores and 32 threads. It boasts a maximum operating frequency of 5.2 GHz and 80 MB of cache. Graphics are provided by a RADEON 8065S with 40 compute units.
The second model, the Ryzen AI Max 490, boasts 12 cores and 24 threads, all running at a maximum of 5 GHz. Cache is reduced to 76 MB, while the graphics section features 32 compute units.
Finally, we have the Ryzen AI Max 485, the smallest of the three. Here, we’re back to 8 cores and 16 threads at 5 GHz. The cache is increased to 40 MB, while the same 32-compute unit RADEON 8050S of the Ryzen Max 490 is used.
In all cases, TDPs are configurable from 45 to 120W, although the base TDP is 55W.
But what’s new?
Finally, these CPUs boast improved RAM support, with AMD announcing that they can accommodate up to 192 GB of unified RAM and up to 160 GB of memory for graphics. This is quite a change from the 112 GB max of the previous generation. Otherwise, we’d like to mention the 100 MHz increase in CPU base frequency, while the GPU also gains 100 MHz.
In short, all these small improvements make it possible to manage large LLMs locally, since we’re talking about model support for over 300 billion parameters. As for agentic AI, AMD’s communication indicates that these CPUs will enable several models to run simultaneously.
