The Aivres KR6288 is the first NVIDIA HGX H200 server we are reviewing. The H200 features a new generation of Hopper GPUs. As a result, this 6U server is designed to house two processors, 32 DIMMs, nine or more 400G NICs, and eight GPUs with over 1.1TB of combined HBM3e memory. We have been reviewing 8-GPU servers with every generation since 2016-2017, so we are excited to kick off our H200 series.
Since this is the first NVIDIA HGX H200 8 GPU of three we are reviewing, we have a video for this one. We also have something a bit different, as I wanted to go into the “why” for these platforms in many cases. Treat this video as a primer on how things are evolving with this class of AI server:
As always, we suggest watching this in its own tab, browser, or app for the best viewing experience. We also did not buy this server because it is too expensive for us to purchase and review. Instead, Aivres is loaning us the server so that we can do the review. We need to say it is sponsored. You also might notice that we did the photos and B-roll on the second set instead of our regular photography set because of how large this system is. Although I am a fan of having either glass or wood tables for our photos, we bring out the moving blanket when a system weighs this much.
Aivres KR6288 External Overview
The front of the 2U system is almost all about airflow.
On the left rack ear, we have the power button and status LEDs.
Next to that, we have eight 2.5″ U.2 SSD slots.
Here is the storage backplane that is installed in the server.
We have a VGA port and two USB 3 ports on the right rack ear. These servers generate a lot of heat, so having cold aisle KVM hookups is nice from a service perspective.
The bottom section is designed to house the NVIDIA HGX H200 8 GPU subsystem. We are going to focus on that after our external overview. Aivres has a midplane designed to provide data and power connections to the massive GPU compute subsystem. You can also see that this server has a tray design that one can directly access the GPUs. GPUs can fail and require replacement. The leading designs for these servers have directly accessible GPUs via trays to make service much faster, an important metric when these systems cost so much.
Another key aspect of this design is providing clear airflow paths since the HGX H200 8-GPU assembly can use well over 5.5kW without cooling. Here is a Kioxia CM7 that you can see clearly through the airflow holes in the midplane.
Since many have yet to see this, we will do an entire section on the NVIDIA HGX H200 8 GPU chassis. It sits on rails to help it slide easily into the main chassis.
Next, let us get to the rear of the server.
“designed to house two processors, 32 DIMMs, nine or more 400G NICs, and eight GPUs with over 1.1GB of combined HBM3e memory.” Minor typo at the beginning, I think that you mean tb ;)
Aivres seems as another brand for Chinese Inspur – same as Kaytus.
Inspur was forced to exit since if they owned a company they couldn’t have a server with H200’s. Kaytus was the one who went to Singapore? Aivres was spun out as the US operations and sales as its own OEM. If they were a Chinese brand owned by Inspur they couldn’t get the H200’s for this server. I’m seeing a H200 server from them, and their business addresses are all in CA, so I don’t think it’s Inspur