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 / Llama 70BMode / EstimateUnits / GB (10⁹ B)

A / Model configuration

ModelMeta

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 hardwareBest fit

Architecture / published config

Layers: 80
Hidden: 8,192
KV heads: 8 × 128
KV / token: 320 KB

B / Estimated VRAM

Fine-tuning

64.0GB

QLoRA-style: quantized frozen base + LoRA adapters. Estimate — not a guarantee.

Frozen base weightsINT4 base model35.3 GB
Adapters + optimizer~0.5% params · FP32 Adam5.6 GB
ActivationsCheckpointed, per step17.2 GB
Runtime overhead≈10% of subtotal5.8 GB

Memory headroom

Free after estimated load

Utilization

Target ≤ 90% of 80.0 GB

Aggregate bandwidth

3.35 TB/s

1 × 3.35 TB/s

VRAM usage

64.0 GB / 80.0 GB

0 GB80 GB

Recommended configuration

1 × NVIDIA H100 SXM 80GB

Total VRAM: 80.0 GB
Topology: Single GPU
Interconnect: NVLink 4 · 900 GB/s
Board power: 700 W

Compatibility / 1 × H100 SXM

Inference
41.5 GB · 52% of 1 × H100 SXM
Ready
Fine-tuning
64.0 GB · 80% of 1 × H100 SXM
Ready
Training
1,262 GB · Needs 24 × H100 SXM (3 nodes)
Cluster

Alternative configurations

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