Workbench / Online
BUILD YOUR COMPUTE STACK
Start with the model. MuseBoard maps the memory and hardware. Settings are mirrored to the URL, so any configuration can be shared.
Compute workbench
B / Estimated VRAM
Inference227.5GB
Serving: weights + KV cache + runtime overhead. Estimate — not a guarantee.
- Model weights202.9 GB
- KV cache4.2 GB
- Runtime overhead20.3 GB
Memory headroom
Does not fit this configuration
Utilization
Target ≤ 90% of 384.0 GB
Decode ceiling
Theoretical, batch 1, bandwidth-bound
VRAM usage
227.5 GB / 384.0 GB
Pinned configuration
16 × NVIDIA GeForce RTX 4090
- Total VRAM
- 384.0 GB
- Topology
- 2 nodes × 8
- Interconnect
- PCIe 4.0
- Board power
- 7,200 W
Compatibility / 16 × RTX 4090
- Ready
Inference
227.5 GB · 59% of 16 × RTX 4090
- Ready
Fine-tuning
305.8 GB · 80% of 16 × RTX 4090
- Cluster
Training
7,191 GB · Needs 336 GPUs — exceeds one consumer node
Alternative configurations
- Needs 16 GPUs — exceeds one consumer node
- 01SERVE / READY
Inference
Estimate memory requirements for serving models.
weights + KV cache + overhead
Open in workbench - 02ADAPT / LORA
Fine-tuning
Explore memory requirements for adapting existing models.
frozen base + adapters + activations
Open in workbench - 03TRAIN / ADAM
Training
Understand large-scale compute requirements.
weights + grads + optimizer + activations
Open in workbench
MuseBoard gives first-order estimates for planning conversations. It does not replace profiling, capacity testing or detailed ML infrastructure planning.