Head-to-head local inference test: 2x RTX 5090 vs 2x RTX PRO 6000 vs Max-Q
Street-Buyer-2428 · reddit · 2026-11-05
A Reddit user is running a rare controlled comparison for local inference: 2x RTX 5090 vs 2x RTX PRO 6000 vs 2x RTX PRO 6000 Max-Q, all swapped through the same pair of PCIe slots in one Threadripper PRO 9975WX / WRX90E / 128GB ECC / 1700W PSU box to eliminate confounds.
All three use the GB202 chip with 1.79 TB/s bandwidth; differences are VRAM (32GB vs 96GB ECC), power (575W / 600W / 300W) and cooling. Testing plan: Qwen3.8-27B (FP8 and NVFP4, TP=2), plain baseline then MTP and DFlash2, measuring prefill, decode and power draw at multiple context lengths, plus a P2P driver unlock run on the 5090s. The author is taking requests for additional benchmarks — a useful decision dataset for anyone building a multi-GPU local inference rig.
More from Infra
- citrini: being part of AI supply chain no longer guarantees 100% yearly stock gains — citrini · 2026-11-05
- Building orbital data centers to dodge regulators is 'completely unserious' — zacharynado · 2026-11-05
- Verda Launches Inference Preview with Day-0 Access to Kimi K3.1 Pro — Kimi_Moonshot · 2026-11-05
- Banach: open-source engine streams 1T-parameter Kimi K2 from NVMe on a 64GB desktop — gautamrbharadwaj · 2026-11-05
- Google unveils $10B, 200-hectare datacenter project in Châteauroux, France with up to 1.2GW — QuintinPope5 · 2026-11-05
- Corebrum OS adds built-in VNC screen share, demoed remote-viewing an NVIDIA Thor robot from a Mac — chrismatthieu · 2026-11-05