Server with RTX A6000: 48 GB on one card and 96 GB as a pair
The NVIDIA RTX A6000 is the top Ampere professional card. According to the NVIDIA specification, it has 48 GB of GDDR6 with error correction (ECC) and 768 GB/s of bandwidth, 10,752 CUDA cores, active cooling and a power draw of up to 300 W. Two cards can be joined with an NVLink bridge (up to 112 GB/s), giving 96 GB as a pair. In essence it is the same card as the A40, but with a fan and faster memory.
What it is good for
- Language models of 30–70B parameters: 30–34B in 8-bit, 70B in 4-bit quantization on one card, or with room for context on an NVLink pair.
- Fine-tuning mid-sized models with LoRA and QLoRA.
- Rendering heavy scenes and visualization that do not fit in 24 GB.
A newer card in the 48 GB class is the L40S with FP8; if you need more memory on a single card, look at the RTX PRO 6000 with 96 GB. How much memory a model takes is covered in the article on running your own language model on a server.
What to check in a plan
- Model name. The RTX A6000 (Ampere) and RTX 6000 Ada are different cards: the latter also has 48 GB but is newer and faster. Check plans with "A6000 Ada" in the name with the host.
- Cloud or dedicated server. This card comes both in cloud plans with 1–4 cards and in dedicated servers; check whether a cloud plan gives you the whole card.
- NVLink in a two-card build: ask whether the bridge is installed.
Monthly and hourly cost
In the catalogue as of 17 November 2026, the median price of a server with one RTX A6000 is about $520 a month, and a single card costs from $420 to $700. Over 730 hours that comes to about $0.71 an hour, less than most other 48 GB cards. None of the hosts with this card in the catalogue offer hourly billing; all plans are monthly. For short jobs, compare hourly plans with the L40S and read how to work out the cost of an hour.