Lawyer upgrades to eight-card V100 private AI server
Understand this story
A lawyer shared on Reddit that he built a private AI server using eight second-hand 2018 Nvidia Tesla V100 data center GPUs, running DeepSeek V4.1 Flash for local inference, with the goal of keeping case files from leaving the office. The lawyer said that at the time of purchase, each four-GPU board cost about $2,400. The lawyer later updated that he raised about 80% of the workload from 2-bit to 4-bit NVFP4 precision, and added 1TB of memory and an RTX 3090. The trade-off was that code output speed dropped from 160–215 tokens/s to 150–180 tokens/s, but the crashes that had previously occurred with the new configuration disappeared.
AI overview based on source reports · Updated 14 min ago
Developments
2 developments
- Nov 15 14:19 · 1 reportsLawyer upgrades eight-card V100 private server to run DeepSeekHacker News AI：Lawyer upgrades eight-V100 private AI server to run 80% of DeepSeek v4.1 Flash
- Nov 10 03:58 · 1 reportsLawyer builds private AI server with eight second-hand V100sHacker News AI：Lawyer builds private AI server with eight used Nvidia V100 GPUs
Coverage timeline
Follow the coverage to explore different perspectives.
- Hacker News AILawyer upgrades eight-V100 private AI server to run 80% of DeepSeek v4.1 Flash
A lawyer running a private AI server on eight used Nvidia V100 GPUs moved about 80% of its work from 2-bit to 4-bit NVFP4 precision, adding 1TB of RAM and an RTX 3090. The server runs DeepSeek V4.1 Flash with a legal harness for his law practice, and code output dropped from 160–215 to 150–180 tokens per second, though the crashes that hit the new build are gone. The crashes, originally blamed on PCIe, were likely caused by a bad PSU.
- Hacker News AILawyer builds private AI server with eight used Nvidia V100 GPUs
A lawyer on Reddit described a private AI server built from eight used 2018-era Nvidia Tesla V100 data center GPUs so case files never leave the office. The build uses two 4-card SXM2 NVLink boards, four PEX switches and SlimSAS cables, and had one burnt power connector; it now runs the open DeepSeek V4.1 Flash model with every expert in GPU memory at up to 215 tokens per second. The four-GPU boards cost about $2,400 apiece when purchased.
Activity for this story
- Current comparable activity
- 3
- Peak comparable activity
- 10Nov 10 05:00
- Change in the past 24 hours
- -49%
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.