Intel had a cool technology on display at Hot Chips 2023, beyond just server chips. It had a direct mesh-to-mesh optical fabric. What might also be interesting is the 8-core processor with 66 threads per core.
Again, please excuse typos, these are being done live.
Intel Shows the First Direct Mesh-to-Mesh Optical Fabric
The key motivation behind this was the DARPA HIVE program for hyper-sparse data.
When Intel profiled the workloads that DARPA was looking at, they found they were massively parallel. Still, they had poor cache line utilization and things like big long out-of-order pipelines were not well utilized.
Here is an interesting one. Intel has a 66-thread-per-core processor with 8 cores in a socket (528 threads?) The cache apparently is not well used due to the workload. This is a RISC ISA not x86.
Intel is packing these into 16 sockets in a single OCP compute thread and using optical networking.
Here is the die architecture. Each core has multi-threaded pipelines.
The high-speed I/O chips bridge the electrical to optical capabilities of the chip.
Here is the 10-port cut-through router being used.
Here is the on-die network where the routers are placed. Half of the 16 routers are there just to provide more bandwidth to the high-speed I/O. On-packaged EMIBs are being used for the physical connection layer.
Going off-die, each chip uses silicon photonics to drive its optical networking. With this, the connections between cores can happen directly between chips even if they are not in the same chassis without adding switches and NICs.
These chips are being packaged as a multi-chip package with EMIB. Having silicon photonics engines added a few other challenges of going from package to strands of fiber.
Here is the optical performance.
In terms of power, this was done in an 8-core 75W CPU. More than half of the power here is being used by silicon photonics.
Here is the simulated to measured workload performance scaling.
Here is the actual die photograph and confirmation that this is being done on TSMC 7nm.
Here is what the package and test board looks like:
This was done in 7nm and work is still happening on this in the lab.
It was interesting to see that Intel did not use the pluggable connector it showed off at Innovation 2022. It seems like this might have been built before that project was ready. This was assisted by Ayar Labs on the optical side.
Perhaps the big item is the 66 threads per core! That is a huge figure. I think folks will enjoy that stat.
Just as a heads-up, we are going to have a video later this week on the Intel Xeon Max (Sapphire Rapids with HBM2e onboard) and will even show it booting a hypervisor and running a VM all from HBM, without DDR5 installed. Intel has a lot of exotic chips either as projects or in production. Subscribe to our 250K+ YouTube channel to see all the fun things we get to do with chips like Xeon Max there.