Skip to content
MUSEBOARD

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

Workbench / Online
Model / Qwen 3BUnits / GB (10⁹ B)

A / Model configuration

Alibaba
B

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

Precision0.5 B / param
Context length8,192 tokens
Batch sizeConcurrent sequences
Workload
Pinned

Architecture / published config

Layers
36
Hidden
2,048
KV heads
2 × 128
KV / token
36 KB

B / Estimated VRAM

Inference

2.0GB

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

  • Model weights1.5 GB
  • KV cache0.30 GB
  • Runtime overhead0.15 GB

Memory headroom

38.0 GB

Free after estimated load

Utilization

5%

Target ≤ 90% of 40.0 GB

Decode ceiling

~1,006 tok/s

Theoretical, batch 1, bandwidth-bound

VRAM usage

2.0 GB / 40.0 GB

0 GB40 GB

Pinned configuration

1 × NVIDIA A100 40GB

Total VRAM
40.0 GB
Topology
Single GPU
Interconnect
NVLink 3 · 600 GB/s
Board power
400 W

Compatibility / 1 × A100 40GB

  • Inference

    2.0 GB · 5% of 1 × A100 40GB

    Ready

  • Fine-tuning

    9.4 GB · 24% of 1 × A100 40GB

    Ready

  • Training

    61.8 GB · Needs 2 × A100 40GB

    Limited

Alternative configurations

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