For any AI workload that demand the highest levels of inference performance and cost efficiency, Jotunn8 delivers a purpose-built solution at every scale.
TFLOPS
TB/S
GB HBM3e
MODELS
Lanes x PCIe Gen5
MB SRAM
VSORA’s software ecosystem is designed to support leading and frontier AI models, including Qwen, Llama, GPT-OSS and Mixtral, alongside industry-standard frameworks such as PyTorch, ONNX, OpenVINO and Hugging Face. With more than a hundred AI models already imported, its CUDA-free software stack and vLLM integration simplify deployment and enable high-performance AI inference with minimal development effort.
Generate more tokens from every watt, maximizing compute efficiency for demanding inference workloads.
Deliver higher AI performance within the same power envelope, using less power per chip.
Reduce the cost of every AI inference by improving economics as workloads and deployments scale.
Leading GPU (fp4) vs VSORA Jotunn8 (fp8).
Leading GPU (fp4) vs VSORA Jotunn8 (fp8).
VSORA works with OEM partners to deliver complete, validated inference systems.
Form factor
4U (8xVSORA Jotunn8)
CPU
Host processor agnostic
Compute
25.6 PFlops
Memory
2.3 TB memory (8x288GB HBM3e)
TDP
Up to 9kW
Cooling
DLC
Compute nodes
8 nodes
Compute
205 PFLOPS
Memory
18.4 TB HBM3e
TDP
80 kW
Liquid cooling
On rack CDU
All-in-one
Turnkey solution available for deployment
