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.

Workbench / Online

Model / Llama 1BMode / EstimateUnits / GB (10⁹ B)

A / Model configuration

ModelMeta

Parameters (billions)

Editing the count switches to a custom model with an estimated architecture.

Precision0.5 B / param

Context length8,192 tokens

Batch sizeConcurrent sequences

Workload

Target hardwarePinned

Architecture / published config

Layers: 16
Hidden: 2,048
KV heads: 8 × 64
KV / token: 32 KB

B / Estimated VRAM

Inference

0.95GB

Serving: weights + KV cache + runtime overhead. Estimate — not a guarantee.

Model weights1.24B params × 0.5 B0.62 GB
KV cache8,192 tokens × batch 10.27 GB
Runtime overhead≈10% of weights0.06 GB

Memory headroom

Free after estimated load

Utilization

Target ≤ 90% of 94.0 GB

Decode ceiling

~6,290 tok/s

Theoretical, batch 1, bandwidth-bound

VRAM usage

0.95 GB / 94.0 GB

0 GB94 GB

Pinned configuration

1 × NVIDIA H100 NVL 94GB

Total VRAM: 94.0 GB
Topology: Single GPU
Interconnect: NVLink bridge · 600 GB/s
Board power: 400 W

Compatibility / 1 × H100 NVL

Inference
0.95 GB · 1% of 1 × H100 NVL
Ready
Fine-tuning
6.6 GB · 7% of 1 × H100 NVL
Ready
Training
27.7 GB · 29% of 1 × H100 NVL
Ready

Alternative configurations

MuseBoard gives first-order estimates for planning conversations. It does not replace profiling, capacity testing or detailed ML infrastructure planning.