NVIDIA RTX A6000
NVIDIA · Enthusiast
Ampere-generation professional workstation card with 48GB of ECC memory, a long-standing favourite for single-card local inference of large models.
Specifications
| VRAM | 48GB |
|---|---|
| Memory bandwidth | 768 GB/s |
| Power draw | 300W |
| Type | gpu |
A used-market favourite among local-AI builders for cheap 48GB. Verify current pricing.
NVIDIA RTX A6000: common questions
- What models can the NVIDIA RTX A6000 run?
- With 48GB of VRAM it can run models up to roughly 77B parameters at a 4-bit quantisation, or smaller models with more context. Use the hardware matrix for specifics; these figures are approximate.
- How much power does the NVIDIA RTX A6000 draw?
- About 300W under load, so pair it with a power supply that has real headroom.
Strengths
- 48GB on one card runs 70B-class models at 4-bit without splitting across GPUs
- A modest 300W draw and single connector make it easy to run in a workstation
- A blower cooler and dual-slot form factor allow multi-card builds for even more memory
Weaknesses
- 768 GB/s bandwidth is well below newer data-centre parts, capping throughput
- Ampere is now two generations old, with no FP8 and slower tensor cores than Ada
- Professional pricing means poor cost-per-GB versus consumer cards
Featured in builds
Entry last verified 18 August 2026. Specifications and especially prices change; verify before buying.