Intel Files a Patent For “Disaggregated GPU” Design, Are We Going To See “True” Multi-Chiplet GPUs Soon?

wccftech.com:

Intel is reportedly eying toward utilizing disaggregated GPU architectures, as Team Blue filed a patent that allows them to leverage dedicated logic "chiplets."

Intel's Newest Patent Filing Shows The Future of GPU Designs, Moving Away From Monolithic Structures To Specialized Chiplets

Well, for enthusiasts, the leap towards moving away from monolithic designs towards adopting smaller, specialized chiplets is an ambition that they certainly see the future of the markets in, and with Intel filing a new patent relevant to this, it is indeed exciting to say that the firm will adopt disaggregated GPU designs moving into the future, although we don't know how far away we are from the actual implementation.

Related Story Intel Might Merge Memory Controller Onto The Compute Die With Panther Lake, Attempting To Fix Deep-Rooted CPU Latency Issues

Based on a patent filing (via @Underfox3), Intel is now exploring the route of using logic chiplets for GPU workloads, and we are hoping to see market adoption for now.

For those unaware, disaggregated GPU architectures is a rather innovative approach towards designing GPUs, which involves moving from monolithic configurations to small, specialized chiplets, which are then interconnected using relevant technology. Dividing the GPU design into chiplets allows manufacturers to fine-tune each chiplet for specific use cases, such as compute, graphics, or AI, allowing their usage for larger applications.

Another huge benefit of disaggregated GPU architectures is their ability to be much more power-efficient compared to existing techniques, given that individual chiplets allow power-gating, which means that when they are not in use, they can be powered down to conserve energy. This design technique brings in several other benefits, such as workload customization, modularity, and flexibility, which is why in the realm of GPU designs, this technique is seen as a benchmark of the future.

Interestingly, we saw AMD file a similar patent earlier