Skip to content
Tech News
← Back to articles

Lawyer expands homemade AI server to 12 used Nvidia V100s for full DeepSeek V4.1 Flash support

read original get NVIDIA Tesla V100 32GB GPU → more articles
GoKawiil Brief

Nick Mermiges added a third board of four Nvidia V100 GPUs to his private AI server, bringing the total to 12 cards and 384GB of memory. This upgrade lets the system hold all 15,360 experts of DeepSeek V4.1 Flash at full 4-bit NVFP4 precision without compression, reaching up to 184 tokens per second on code tasks. The full build reportedly cost $9,450.

Why It Matters

GoKawiil's interpretation of the reporting above, not reported fact.

The project suggests that older, discontinued GPUs like the V100 can still be repurposed to run cutting-edge AI models by using software workarounds to unpack newer quantization formats like NVFP4, which the hardware wasn't originally designed to support. Mermiges claims his setup can outperform commercial DGX Spark configurations at single-request decoding for roughly half the cost of three such units, which could hint at a cheaper path for hobbyists or small firms to run large models locally rather than relying on costly new hardware.

Key Takeaways
Worth a Look

NVIDIA Tesla V100 32GB GPU — This build is the perfect example of how used datacenter GPUs like the V100 can be stacked affordably to run massive local LLMs at full precision. If you're into DIY AI server builds, sourcing used V100 cards is a budget-friendly way to get into high-VRAM multi-GPU inference setups like this one.3

See NVIDIA Tesla V100 32GB 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, 2026-11-20

Published there as: “Lawyer’s $9,450 private AI server of 12 used Nvidia V100s now runs every DeepSeek V4.1 Flash expert at 4-bit NVFP4”

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.
Keep reading
dgx spark 19 briefs → deepseek v4.1 flash 16 briefs → ai server 5 briefs → nvidia v100 5 briefs →