Inventec Artemis Rack
Inventec's integrated rack build of the NVIDIA GB200 NVL72, and the first of its Artemis family. Eighteen 1RU compute trays carrying four GPUs each, plus nine 1RU NVLink switch trays, fill a 48RU MGX rack, pairing 36 Grace CPUs with 72 Blackwell GPUs. GPU to GPU traffic runs over passive copper NVLink cable cartridges rather than optics, and the rack accepts rack-mounted, row-based or sidecar liquid cooling with an Inventec-designed rack-mounted CDU. Blind-mate manifolds are used throughout for serviceability. Inventec rates the rack at a maximum of 125 kW for the GB200 NVL72 building blocks; note NVIDIA quotes 120 kW for the reference design, and this entry keeps Inventec's figure since Inventec is the one publishing it. No throughput or memory figures are published for the rack; those are on the linked NVIDIA reference entry.
We will point you at suppliers who have it. Free, and no signup.
Systems built on this design
1 configurationBuilt on the NVIDIA GB200 NVL72 baseboard, so accelerator performance matches the reference. Chassis, CPU, memory and cooling are this vendor's own.
System Details
GPU Configuration
System Specifications
Precision Performance Breakdown
| Precision | System Performance | Per GPU | Efficiency |
|---|---|---|---|
| 2.88 PFLOPS | 40 TFLOPS | 27.0 TFLOPS/kW | |
| 5.76 PFLOPS | 80 TFLOPS | 54.0 TFLOPS/kW | |
| 90.0 PFLOPS | 1,250 TFLOPS | 1500.0 TFLOPS/kW | |
| 180 PFLOPS | 2,500 TFLOPS | 2700.0 TFLOPS/kW | |
| 180 PFLOPS | 2,500 TFLOPS | 2700.0 TFLOPS/kW | |
| 360 PFLOPS | 5,000 TFLOPS | 5400.0 TFLOPS/kW | |
| 720 PFLOPS | 10,000 TFLOPS | 10800.0 TFLOPS/kW | |
| 360 POPS | 5,000 TOPS | 5400.0 TFLOPS/kW | |
| FP6 | 360 PFLOPS | 5,000 TFLOPS | — |
Where a vendor states a basis, the figure shown is the dense one, and any figure it publishes is listed separately. Vendors commonly headline the sparse number instead: for NVIDIA's 2:4 structured sparsity that is exactly twice the dense figure, though other vendors' sparse modes do not all follow that ratio.
Powered by NVIDIA GB200
This system utilizes 72 × NVIDIA GB200 NVL72 GPU 186GB GPUs, each delivering exceptional performance for AI and HPC workloads.
Per GPU TDP
—
Per GPU Memory
186 GB
Process Node
4nm
Architecture
Blackwell
Documentation & Resources
Flopper Spec Sheet
Inventec Artemis Rack specifications, generated from our database. Printable.
Vendor product page
Inventec Artemis Rack documentation
Inventec Artemis Rack
Inventec • Latest version
Typical Use Cases
The Inventec Artemis Rack runs 72× GB200 GPUs over NVLink across all 72 GPUs over passive copper cable cartridges, with nine 1RU NVLink switch trays; scale-out is ConnectX-7 at 400 Gb/s InfiniBand east-west and a BlueField-3 DPU at 400 Gb/s north-south, delivering 360 PFLOPS FP8 dense.