GoKawiilLawyer builds 8-GPU V100 server for in-house AI case review
A lawyer posted on Reddit detailing a homebuilt AI server using eight used 2018-era Nvidia Tesla V100 SXM2 32GB GPUs, paired via two NVLinked four-card baseboards and four PEX switches on a Threadripper Pro system with 2,800W of power. The machine runs the DeepSeek V4.1 Flash model entirely in GPU memory at up to 215 tokens per second, letting the lawyer process case files locally without sending data to outside servers.
GoKawiil's interpretation of the reporting above, not reported fact.
The build illustrates how secondhand enterprise GPUs can offer large pools of fast VRAM at lower cost than new hardware, though eBay listings cited in the post show V100 prices have already more than doubled since the builder's original purchase. This suggests growing demand for used AI accelerators among individuals and small firms seeking private, on-premises AI tools for sensitive work like legal document review.
- Server uses eight used Nvidia V100 32GB GPUs across two NVLinked four-card SXM2 boards
- Runs DeepSeek V4.1 Flash locally at up to 215 tokens/second for first-pass legal document review
- Used V100 prices have more than doubled since the builder's purchase, per current eBay listings
NVIDIA Tesla V100 32GB SXM2 GPU — If you're inspired by this lawyer's DIY private AI server, the V100 SXM2 32GB is the exact card driving the build, offering huge HBM2 bandwidth and NVLink support for multi-GPU inference setups. It's a proven workhorse for running large open-weight models locally without exposing sensitive data to the cloud.
See NVIDIA Tesla V100 32GB SXM2 GPU on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: tomshardware.com — Shane Downing, 2026-11-09
Published there as: “Lawyer builds private AI server with eight used Nvidia V100 GPUs”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.