Shunt-modded RTX 6000 PRO MaxQ outperforms RTX 5090 in Anima and LLM benchmarks

A user compares the full compute performance of an RTX 5090 against shunt-modded and water-cooled RTX 6000 PRO MaxQ cards, as well as a rented RTX 6000 PRO WS edition. The test evaluates both Anima benchmark scores and LLM prompt processing speeds across various power limits.

• Shunt-modded RTX 6000 PRO MaxQ at 600W achieved the highest Anima score, finishing in 32.7s (12.8% faster than the baseline RTX 5090 at 600W).
• The same MaxQ configuration at 475W remained faster than the RTX 5090 at 600W, completing the benchmark in 35.3s.
• RTX 6000 PRO WS at 600W performed nearly identically to the baseline RTX 5090 at 600W (37.3s vs 37.5s).
• LLM tests using Kimi K2 and GLM 5.1 models were conducted locally with llamacpp, identifying the primary GPU as the bottleneck.

The results demonstrate that modifying the power limit of the RTX 6000 PRO MaxQ allows it to exceed the performance of a high-power consumer RTX 5090 in both compute and inference tasks.