Apertus 70B — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
The flagship Apertus. At Q4 it needs a 48 GB card or a large-memory unified machine.
Apertus 70B needs about 43 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
VRAM and speed by quantization
Quoted against NVIDIA RTX 4090 (24 GB). Weights plus overhead only: this model's architecture is unpublished, so no KV cache is added.
| Quant | Memory | VRAM | Speed (est.) | Fit |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 23.8 GB | ~29 tok/s | Tight | |
| Q3_K_M 3.41 bpw | 30.6 GB | ~3 tok/s | Offload | |
| Q4_K_M 4.83 bpw | 43.1 GB | ~3 tok/s | Offload | |
| Q5_K_M 5.67 bpw | 50.4 GB | ~2 tok/s | Offload | |
| Q6_K 6.56 bpw | 58.2 GB | — | Too big | |
| Q8_0 8.50 bpw | 75.2 GB | — | Too big | |
| F16 16.00 bpw | 140.8 GB | — | Too big |
Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. Apertus 70B VRAM calculator →
Get Apertus 70B running
The cheapest catalogued GPU that runs Apertus 70B is the NVIDIA DGX Spark (128 GB).
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
How to run Apertus 70B
Install Ollama, then run:
ollama run apertusSpecifications
Corroborated — Two or more independent sources agree on these figures, but the model card itself was not retrieved. Treat the numbers as good rather than confirmed. Still unconfirmed: context, releaseDate, hfModelId.
- Parameters
- 70 Billion
- Context window
- 65,536
- Architecture
- Dense
- Provider
- EPFL / ETH Zurich / CSCS
- Licence
- Apache 2.0
- Practical deployment
- Workstation (large unified memory or pro GPU)
- Specified at
- Q4_K_M
- System RAM
- 64 GB
- Record updated
- 2026-09-19
Commercial use permitted. No usage restrictions beyond attribution.
Quality and use cases
Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.
Other Apertus sizes
Apertus 70B — frequently asked questions
How much VRAM does Apertus 70B need?
About 43 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does Apertus 70B run on an RTX 4090 (24 GB)?
No. Apertus 70B needs about 43 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.
How do I run Apertus 70B locally?
Install Ollama and run `ollama run apertus`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.
What other sizes does Apertus come in?
Apertus 8B (6 GB), Apertus 70B (43 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.