The Nvidia Tesla V100 graphics accelerator, released eight years ago, still demonstrates high efficiency in running AI models locally. Tests conducted by the Hardware Haven channel show that this legacy server card performs faster than modern models like the RTX 3060 12 GB and Radeon RX 7800 XT 16 GB. This is reported by Ixbt.com .
Currently available on the secondary market for around $100, the Tesla V100 requires an SXM-to-PCIe adapter, a separate power supply, and a 3D-printed cooling system for installation in a standard PC. The total cost of the system, including all additional components, was $235.
According to test results, the Tesla V100 generated 130 tokens per second on the GPT-oss 20B model, significantly higher than the Radeon RX 7800 XT (90 tokens/s). It also outperformed the RTX 3060 12 GB on the Gemma4:e4b model with a ratio of 108 to 76 tokens/s.
The main advantage of this card lies in the high bandwidth of its HBM2 memory and its optimization for computational tasks. Although the installation process is complex, server GPUs remain a very useful and affordable solution for AI enthusiasts.
Experts also recommend considering the 32 GB versions of the Tesla V100. Selling for around $400–$500, these models can be a worthy alternative to expensive modern RTX cards for running heavier LLMs locally.
Read “Zamin” on Telegram!
