RTX A6000
A professional workstation graphics card released by NVIDIA in 2020, based on the Ampere architecture with 48 GB of GDDR6 memory, belonging to the RTX A series.
Timeline (last 90 days)
There are at least four NVIDIA professional graphics cards named "RTX 6000" on the market: Quadro RTX 6000 (Turing, 2018, 24GB GDDR6, 672 GB/s), RTX A6000 (Ampere, 2020, 48GB GDDR6, 768 GB/s), RTX 6000 Ada Generation (Ada Lovelace, 2022, 48GB GDDR6, 960 GB/s), and RTX PRO 6000 Blackwell (Blackwell, 2025, 96GB GDDR7, 1,597–1,792 GB/s).
The RTX PRO 6000 Blackwell has a compute capability of 12.0, the RTX 6000 Ada 8.9, and the RTX A6000 8.6.
The RTX A6000 is Ampere architecture and has no native FP8 hardware, sglang unpacks FP8 weights on the fly via Marlin kernels, the base Qwen3.8-27B decodes at 24 tok/s on this card, while RedHatAI's INT4 W4A16 quantization reaches 41 tok/s on the same card and the same 1M context.
Based on measurements with chunks of about 250 words on an RTX A6000, all-MiniLM-L6-v2 processes 1,074.7 chunks per second, taking about 16 minutes for 1 million chunks; Qwen3-Embedding-8B processes 26.1 chunks per second, taking about 10.6 hours for 1 million chunks.
All Facts (4)
There are at least four NVIDIA professional graphics cards named "RTX 6000" on the market: Quadro RTX 6000 (Turing, 2018, 24GB GDDR6, 672 GB/s), RTX A6000 (Ampere, 2020, 48GB GDDR6, 768 GB/s), RTX 6000 Ada Generation (Ada Lovelace, 2022, 48GB GDDR6, 960 GB/s), and RTX PRO 6000 Blackwell (Blackwell, 2025, 96GB GDDR7, 1,597–1,792 GB/s).
50%UnverifiedThe RTX PRO 6000 Blackwell has a compute capability of 12.0, the RTX 6000 Ada 8.9, and the RTX A6000 8.6.
50%UnverifiedThe RTX A6000 is Ampere architecture and has no native FP8 hardware, sglang unpacks FP8 weights on the fly via Marlin kernels, the base Qwen3.8-27B decodes at 24 tok/s on this card, while RedHatAI's INT4 W4A16 quantization reaches 41 tok/s on the same card and the same 1M context.
50%UnverifiedBased on measurements with chunks of about 250 words on an RTX A6000, all-MiniLM-L6-v2 processes 1,074.7 chunks per second, taking about 16 minutes for 1 million chunks; Qwen3-Embedding-8B processes 26.1 chunks per second, taking about 10.6 hours for 1 million chunks.
50%