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 / DeepSeek 32BUnits / GB (10⁹ B)

A / Model configuration

DeepSeek
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
64
Hidden
5,120
KV heads
8 × 128
KV / token
256 KB

B / Estimated VRAM

Inference

20.2GB

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

  • Model weights16.4 GB
  • KV cache2.1 GB
  • Runtime overhead1.6 GB

Memory headroom

11.8 GB

Free after estimated load

Utilization

63%

Target ≤ 90% of 32.0 GB

Decode ceiling

~109 tok/s

Theoretical, batch 1, bandwidth-bound

VRAM usage

20.2 GB / 32.0 GB

0 GB32 GB

Pinned configuration

1 × NVIDIA GeForce RTX 5090

Total VRAM
32.0 GB
Topology
Single GPU
Interconnect
PCIe 5.0
Board power
575 W

Compatibility / 1 × RTX 5090

  • Inference

    20.2 GB · 63% of 1 × RTX 5090

    Ready
  • Fine-tuning

    33.9 GB · Needs 2 × RTX 5090

    Limited
  • Training

    590.2 GB · Needs 24 GPUs — exceeds one consumer node

    Cluster

Alternative configurations