Workroom
Datacenter fleet
Workroom is an agent computer for managing datacenter servers and GPUs. Open any machine to see its health and what the agent can do on it. Every change waits for your ok.
- Servers online
- 996 of 1,000
- GPUs
- 1,280
- Power draw
- 2.91 MW
- Cooling load
- 2.83 MW
- Open incidents
- 3
Fleet floor
hover for specs · click to inspect- healthy
- agent working
- failed part
- maintenance
- PDU and cooling units
MCL-1 floor plan, 1,000 servers.
Inspection
needs attentiongpu-027
NVIDIA DGX B300 · rack A07, slot 3 · inlet 31 °C · 14.1 kW
- gpu-025
- gpu-026
- gpu-027
- gpu-028
What the agent is doing now
- Rack inlet at 31 °C, migrating jobs to rack A12
- Health check: 8 of 8 GPUs responding, memory errors clean
- Scheduling: placed training job tj-4471 on GPUs 0-7
- GPUs
- 8x NVIDIA B300
- GPU memory
- 2.1 TB
- Training
- 72 PFLOPS FP8
- NVLink
- 14.4 TB/s
- Networking
- ConnectX-8 800 Gb/s
- DPUs
- BlueField-3
- CPUs
- 2x Intel Xeon 6776P
- Size
- 10U
- Power
- ~14.5 kW
Operations the agent can run
Every operation waits for your ok. Power actions go through the server's BMC.
Agent activity across the fleet
- running
Rack A07 inlet reached 31 °C. Migrating 3 training jobs to rack A12.
- attempted
PSU 2 failed on stor-118. Load moved to PSU 1, ticket INC-2304 opened.
- running
Provisioning new node cpu-488. Burn-in checks running.
- verified
Closed incident INC-2291 on gpu-031: fan replaced, checks passed.