Benchmarking Legacy NVIDIA Enterprise GPUs on Modern Workloads
eso_logic · reddit · 2026-07-13
Using custom GPU coolers and benchmarking tools developed over the past year, the author evaluates whether a batch of retired NVIDIA enterprise cards still holds value for modern workloads.
Background
- Budget cards like the P100 (16GB) cost around $75, while the V100 (16GB) is under $200.
- Paired with cheap X99 Xeon motherboards, you can amass a massive pool of "idle VRAM".
- While acknowledging issues like end-of-life software and poor power efficiency, the author considers this acceptable for a homelab setup; simply turning them off when idle saves power.
Testing Setup
Using a Dockerized benchmark suite, the author tested various Tesla cards on:
- LLMs
- Computer Vision
- Blender
- Whisper
- Other modern workloads
Key Findings
- The V100 is the sweet spot: The 16GB V100 performs surprisingly close to the more expensive T40.
- The P40 beats the P100 for LLMs: This aligns with community consensus.
- The M60 is great for Whisper: It delivers astonishing audio transcription performance, occasionally beating the V100 at a mere $50.
- Scaling is mostly linear: Stuffing cards into a 4U chassis shows no severe diminishing returns; however, mixing generations means slower cards will bottleneck LLM workloads.
- Low CPU/Motherboard requirements: Faster single-core CPUs only marginally benefit Whisper and Vision Transformers; a cheap X99 motherboard and a Xeon with ample PCIe lanes are generally sufficient.
The author plans to test additional workloads in the future.
More from Infra
- NVIDIA publishes Vera CPU architecture details before AMD’s AI event — ryanshrout · 2026-07-22
- oMLX 0.5.2 adds Mac menu-bar stats, low-bit decode kernels, and faster downloads — awnihannun · 2026-07-22
- Strangeworks launches Aura to turn enterprise ops into production optimization systems — whurley · 2026-07-22
- Graph workload 854.graph500 enters SPEC CPU 2026 as a new CPU benchmark — Prof_DavidBader · 2026-07-22
- HilbertRaum open-sources a fully local AI chat and document analysis app for private use — Vladowski · 2026-07-22
- Hybrid and local inference are emerging as a response to AI energy and token costs — dmitry140 · 2026-07-22