Intel is doing a mini launch of its next-generation server CPU (or something thereabouts at SC26). There are many systems on the floor, but one, the Bull Sequana XH31600, actually showed Diamond Rapids packages.
Bull Shows Intel Xeon Diamond Rapids Package at SC26
Starting off, here is the Bull compute blade with blind mate power and liquid at the rear, networking in front, and four Intel Xeon Diamond Rapids packages up front.

Just ot be clear about what you are seeing, these are two dual-socket compute nodes with Intel Diamond Rapids, each with 16-channel memory.

Here is a view of the package. We have already seen the LGA9324 socket covered in theĀ ASRock Rack DMRD16-2L+ from OCP Summit 2026.

You can see that in the Bull platform, the CPUs are on fairly wide motherboards because of all of the memory on either side.

The idea is that these blades are liquid cooled, so you can see how everything has heatpipes and liquid cooling blocks.

The other node had the liquid cooling blocks as well as the cabled connections to the networking.

On the networking, Bull had its BXI technology for interconnect. In its booth, Bull showed a BXI switch with 64x 800G OSFP ports, but the BXI NICs were shown as QSFP-DD and PCIe Gen5 x16. It actually looks like Bull is doing something similar to what we did with theĀ NVIDIA ConnectX-8 C8240 800G Dual 400G NIC Review This is a SuperNIC when we managed to use two PCIe Gen5 x16 links to get 800Gbps from the ConnectX-8.

What stood out a bit is that there was an NVIDIA ConnectX-8 SuperNIC as well in an OCP NIC 3.0 form factor. It’s interesting that Bull has its own interconnect, with a 51.2T switch and a PCIe Gen5 NIC in a platform with a Gen6 CPU, but right next to it is NVIDIA’s high-volume PCIe Gen6 NIC competitor.

What is very cool about this platform is that it is incredibly bespoke.

For years, Bull has had one of the better-looking blade designs, largely because they liquid-cool everything. Probably worth pondering, however, is that with NVIDIA pushing the industry with ORv3, and AMD now with the double-wide. It feels like there is a high-volume standard for high-density racks in the industry if you want power and cooling. Maybe the more interesting part is that Bull’s design does not have a copper-cabled backplane in the rear.

Overall, this was very neat to see.
Final Words
What has been interesting at SC26 is that this type of platform would have been mind-blowing years ago. Now, with the NVIDIA Vera Rubin designs in ORv3 racks, the Bull systems fit in with the higher-volume AI infrastructure bits. What Bull has, however, is SiPearl, Intel Xeon Diamond Rapids, NVIDIA, and AMD Venice nodes on display in a common form factor.

Many other vendors are showing off Diamond Rapids systems with neat designs, including 32DIMM/ socket 2DPC configurations at SC26. For its part, Bull is the only one we have found that has a package. Diamond Rapids seems to be in the sampling phase. The best guess is maybe we start to see a launch in Q2 2027, but if that slips past June (or it becomes a split launch with 1P platforms and 2P platforms), the next real launch window for Intel is early September 2027.



