SupermicroSYS-822GS-NB3RT-01-G2

Eight NVIDIA Blackwell Ultra GPUs on an HGX B300 board, two Xeon 6768P and 2 TB of memory in an 8U chassis, in the configuration Supermicro already has assembled and tested.

Ask for price and lead timeSee the configuration
Supermicro SYS-822GS-NB3RT-01-G2Rack 8U · Gold Series G2

The exact configuration

Official datasheet

Front view diagram: hover over each component.
Processor2× Intel Xeon 6768P64 cores · 2.4 GHz
GPUNVIDIA HGX B300 8-GPU
Memory2 TBDDR5-6400
Storage1× 960 GB M.2 NVMe
Network8-port 800 GbE / XDR800
Form factorRack 8U
Warranty
3 years parts and labour
On-site engineer
Optional: next business day, for 3 years
Software
Supermicro Data Center Management Suite, licence per node

For training large models: 2.3 TB of GPU memory joined by NVLink

This is the machine for training and fine-tuning large models, and for serving them to many users at once. The eight GPUs add up to around 2.3 TB of HBM3e and are joined by NVLink inside the chassis, so a model too big for one GPU's memory is spread across all of them without touching the network, while the eight 800 GbE ports, which also run as InfiniBand XDR (800 Gb/s), are there to join several nodes into a cluster when one is not enough.

Compared with the RTX PRO 6000 machines, whose GPUs communicate over PCIe, here traffic between cards goes over NVLink and NVSwitch, and that is what counts in training, because at every step the eight GPUs have to synchronise their results. For inference on models that fit in 768 GB, a 4U SYS-422GA-NRT-02-G2 is usually enough.

What we do

We deliver it with the operating system and the GPU stack installed and tested with your workload: RHEL or Ubuntu, the drivers and the NVIDIA GPU Operator on Kubernetes or OpenShift AI, and vLLM or another inference server on top. If your training data lives on Ceph or other storage, we also build the network between the two. Support afterwards comes from the same people who installed it.

More on how we build AI inference on Kubernetes and Ceph.

What your data centre needs

It takes 8U and, according to the official datasheet, carries six 6,600 W Titanium power supplies in 3+3 redundancy, which means six sockets split across two independent circuits if the redundancy is to mean anything. Eight GPUs of this generation, the two processors and the fans go well past 10 kW per chassis, and that decides how many fit in a rack. Cooling is by air, with up to twelve high-performance fans, so before the order we go through rack power, PDUs and cold-aisle containment with you.

Compared with the rest of the family

ModelWhat it hasWhen to choose it
AS-8126GS-NB3RT-01-G2HGX B300 8-GPU · 3 TBSame HGX B300 with AMD, 3 TB and a separate storage network.
AS-5126GS-TNRT-01-G22x RTX PRO 6000 · 1.5 TBTwo GPUs today, space for eight and memory to spare.
SYS-212GB-FNR-01-G22x RTX PRO 6000 · 512 GBThe most compact: 2U for a two-GPU service.
SYS-422GA-NRT-01-G24x RTX PRO 6000 · 1 TBFour GPUs that MIG splits into up to sixteen instances for several teams.
SYS-422GA-NRT-02-G28x RTX PRO 6000 · 1.5 TBThe most PCIe can do: eight GPUs and 400 GbE networking.
SYS-C542i-11302URTX 50-ready · 32 GBDevelopment tower without a standard GPU; you pick the card.
ARS-511GD-NB-LCC-01-G2Grace + B300 · 748 GBA GB300 node in a tower for teams that fine-tune models without booking a cluster.

See all AI and GPU servers

What the delivery includes

From the factory

  • Configuration assembled and tested by Supermicro
  • Warranty: 3 years parts and labour

From us

  • Rack mounting
  • Remote management (BMC) and up-to-date firmware
  • Operating system and the stack that runs on top
  • Testing with your workload before it goes into production

If you need it

  • 24×7 on-call support with SLA
  • On-site engineer next business day for the full 3 years

How we build it

What runs on top of these servers, explained in detail:

AI inference with Ceph, OpenStack and KubernetesAll our AI services

Support and training for your team

Independent support from the people who installed itLocal LLM and GPU infrastructure courseDeep learning course with PyTorch and TensorFlowRed Hat OpenShift training

Questions about the SYS-822GS-NB3RT-01-G2

How does it differ from the AS-8126GS-NB3RT-01-G2?

They carry the same eight-GPU HGX B300 board. The AS-8126GS swaps the Xeons for two AMD EPYC 9575F, raises memory to 3 TB and adds two dual-port 200 GbE cards alongside the eight 800 GbE ports, which helps keep storage traffic separate. The SYS-822GS is the one to choose if the rest of your estate runs on Intel.

What does the data centre need?

An 8U chassis per server, six sockets for its 6,600 W supplies and cooling ready for more than 10 kW per machine. Before quoting we go through rack power, PDUs and room temperature with you, because that is what most often delays an installation of this kind.

Can I start with one node and grow?

Yes. That is what the eight 800 GbE ports are for: joining HGX nodes into a cluster where the GPUs in one machine talk to those in another over the network. With two or three nodes a single switch is enough, and beyond that we design the network and storage with you.

Is the SYS-822GS-NB3RT-01-G2 right for what you're building?

Tell us about the workload and the data centre it's going into, and we'll come back with the price, a confirmed lead time and what it would take to get it running.

Ask for price and lead timeWhatsApp+34 91 198 02 43