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 / Phi-3 MiniMode / EstimateUnits / GB (10⁹ B)

A / Model configuration

ModelMicrosoft

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: 32
Hidden: 3,072
KV heads: 32 × 96
KV / token: 384 KB

B / Estimated VRAM

Inference

5.3GB

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

Model weights3.82B params × 0.5 B1.9 GB
KV cache8,192 tokens × batch 13.2 GB
Runtime overhead≈10% of weights0.19 GB

Memory headroom

Free after estimated load

Utilization

Target ≤ 90% of 24.0 GB

Decode ceiling

~490 tok/s

Theoretical, batch 1, bandwidth-bound

VRAM usage

5.3 GB / 24.0 GB

0 GB24 GB

Pinned configuration

1 × NVIDIA GeForce RTX 3090

Total VRAM: 24.0 GB
Topology: Single GPU
Interconnect: PCIe 4.0 · NVLink bridge (2-way)
Board power: 350 W

Compatibility / 1 × RTX 3090

Inference
5.3 GB · 22% of 1 × RTX 3090
Ready
Fine-tuning
6.3 GB · 26% of 1 × RTX 3090
Ready
Training
71.1 GB · Needs 4 × RTX 3090
Limited

Alternative configurations

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