I put a datacenter GPU in my gaming PC
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Tesla V100 SXM2 – bought used on eBay for ~£150, a 2017 NVIDIA data‑center GPU with 16 GB HBM2 and 5120 CUDA cores.
- SXM2‑to‑PCIe adapter – a third‑party PCB costing ~£50 that lets the V100 connect to a standard PCIe slot alongside an RTX 4080.
- V100 – delivers 900 GB/s memory bandwidth, 22 % higher than the RTX 4080’s 736 GB/s and outperforms Apple’s M5 Max (614 GB/s) despite its 2017 launch.
- Dual‑GPU setup – combining the V100 with the RTX 4080 yields 32 GB total VRAM and runs a 27‑billion‑parameter LLM at ~32 tokens per second.
- Cost comparison – the V100 + adapter combo costs ~£200, a fraction of the £2,000+ price of a consumer RTX 5090 with comparable VRAM.
Why it matters: DIY AI enthusiasts gain high‑performance LLM inference at a fraction of consumer GPU cost, while Nvidia’s data‑center GPU market sees extended resale value, pressuring consumer GPU pricing and encouraging hobbyist repurposing.
