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 70BUnits / 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
80
Hidden
8,192
KV heads
8 × 128
KV / token
320 KB

B / Estimated VRAM

Inference

41.5GB

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

  • Model weights35.3 GB
  • KV cache2.7 GB
  • Runtime overhead3.5 GB

Memory headroom

214.5 GB

Free after estimated load

Utilization

16%

Target ≤ 90% of 256.0 GB

Decode ceiling

~170 tok/s

Theoretical, batch 1, bandwidth-bound

VRAM usage

41.5 GB / 256.0 GB

0 GB256 GB

Pinned configuration

1 × AMD Instinct MI325X 256GB

Total VRAM
256.0 GB
Topology
Single GPU
Interconnect
Infinity Fabric
Board power
1,000 W

Compatibility / 1 × MI325X

  • Inference

    41.5 GB · 16% of 1 × MI325X

    Ready
  • Fine-tuning

    64.0 GB · 25% of 1 × MI325X

    Ready
  • Training

    1,262 GB · Needs 8 × MI325X

    Limited

Alternative configurations