Overview
AbyssCore V800 cluster is 2U rack mounted GPU chasis that can be hooked to any computer with four PCIe 4.0 (PCIe 3.0) x16 slots. V800 has total of eight NVidia Tesla V100 GPU’s hooked in pairs with each pair having super fast NVLink communication between each other. Total available video memory available in V100 is either 128GB or 256GB depending on GPU configuration. It has sufficient memory capacity to run large language models with fairly large amount of parameters. V100 is mature GPU architecture that has Tensor cores and still support for latest Cuda versions. This makes it the most cost efficient GPU in the market to start runing LLMs with.
Features
The main highlights of the V800 GPU cluster are
- 8pcs of either 16GB or 32GB NVidia V100 SXM2 GPU
- SFF-8654 link to host system with four x16 PCIe busses (1 for each GPU pair)
- Watercooling for each GPU with common inlet/outlet for the cluster
- Total of 2800W PSU (2x 1400W Dell Platinium PSUs)
- Maximum power consumption is 2.4kW
Technical specification
Technical parameters of the cluster are
Disclaimer
Technical details and availability is subject to change. Non stock item lead time is 30-60 days depending on part availability.
Arviot
Tuotearvioita ei vielä ole.