During a recent earnings call, AMD revealed its plans to bolster its line of EPYC processors by infusing them with Xilinx’s FPGA-powered AI inference engine.
While the first of these new AI-infused CPUs are slated to arrive next year, the move shows that the chip giant is working to incorporate the technologies from its $54bn Xilinx acquisition into its existing products.
Still though, recent patents from AMD show that the company is actually working on several methods to connect AI accelerators to its processors including using 3D chip stacking tech to do so.
AMD isn’t the first chipmaker to pair its CPUs with in-built FPGAs as Intel tried something similar after acquiring Altera back in 2015. Despite the fact that the company announced and even demoed a combined CPU+FPGA chip, the final product ended up being more of an experiment that it hasn’t revisited since.
FPGA-infused CPUs
Although AMD hasn’t yet gone into the details regarding its future FPGA-infused products, connecting Xilinx FPGA silicon to its processors will likely require some sophistication.
Instead of using standard PCIe lanes and QPI interconnect to connect FGPA chips to a CPU like Intel did, AMD’s patents suggest that the company is working on an accelerator port to do so in a variety of ways.
AMD could use 3D stacking chip tech like it does with its Milan-X processors to provide performance, power and memory throughput advantages. However, this approach can present thermal challenges that would reduce performance if the chiplet is placed too close to compute dies.
Fortunately, we’ll likely learn more about AMD’s upcoming AI-infused CPUs at the company’s upcoming Financial Analyst Day next month.
Via Tom's Hardware
