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 / DeepSeek V3Mode / EstimateUnits / GB (10⁹ B)

A / Model configuration

ModelDeepSeek

Parameters (billions)37B active / token

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: 61
Hidden: 7,168
KV heads: MLA
KV / token: 69 KB

Uses multi-head latent attention (MLA), which compresses the KV cache substantially.

B / Estimated VRAM

Inference

369.6GB

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

Model weights671B params × 0.5 B335.5 GB
KV cache8,192 tokens × batch 10.58 GB
Runtime overhead≈10% of weights33.6 GB

Memory headroom

Does not fit this configuration

Utilization

Target ≤ 90% of 512.0 GB

Decode ceiling

~1,550 tok/s

Theoretical, batch 1, bandwidth-bound

VRAM usage

369.6 GB / 512.0 GB

0 GB16 × 32 GB

Pinned configuration

16 × NVIDIA GeForce RTX 5090

Total VRAM: 512.0 GB
Topology: 2 nodes × 8
Interconnect: PCIe 5.0
Board power: 9,200 W

Compatibility / 16 × RTX 5090

Inference
369.6 GB · 72% of 16 × RTX 5090
Ready
Fine-tuning
442.8 GB · 86% of 16 × RTX 5090
Ready
Training
11,824 GB · Needs 416 GPUs — exceeds one consumer node
Cluster

Alternative configurations

Needs 16 GPUs — exceeds one consumer node
Mixture-of-experts: all 671B parameters stay resident in memory; only ~37B are active per token.

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