- Home
- Datacenters
- Mistral Large Model Implied Cluster
Existing Unlikely
Mistral Large Model Implied Cluster
• November 26, 2023
Total GPUs
4.0K
H100 Equiv
4.0K
Power
5.8 MW
FP16
4.0K PF
System Equivalents
This datacenter's 4.0K GPUs equal:
NVIDIA DGX H100
8 GPUs/system
500
NVIDIA HGX H100 (8-GPU)
8 GPUs/system
500
NVIDIA HGX H100 (4-GPU)
4 GPUs/system
1.0K
Specifications
| Primary GPU | NVIDIA H100 SXM5 80GB |
| GPU Count | 4.0K |
| Sector | Private |
Additional Information
"Mensch said his new model costs less than 20 million [Euros] to train"
Taking H100 price as $2/hr, and assuming they trained for 3 months max, this implies that the cluster they used was >4k GPUs