Power-limit vs TG/s for 2x3090

<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1t4b203/for_those_wondering_about_the_power_consumption/"> <img alt="For those wondering about the power

<!-- SC_OFF --><div class="md"><p>Got Qwen 3.5 27B to 1,103,941 tok/s on 12 nodes with 96 B200 GPUs.</p> <p>At that rate you process 50,000 insurance policy documents in hours

<!-- SC_OFF --><div class="md"><p>Hey everyone, I've been working on getting Multi-Token Prediction (MTP) working with quantized GGUFs for Qwen3-27B and the results are pretty

<!-- SC_OFF --><div class="md"><p>Every output token the model generates requires reading the entire KV cache from High Bandwidth Memory.</p> <p>Here's what that actually looks

<!-- SC_OFF --><div class="md"><p><strong>Geezuz, its fast!</strong></p> <p>Qwen3.8-27B, two RTX 3090s power-limited to 280 W, plain PCIe, no NVLink, at 0 Context:<br /> (tests

<!-- SC_OFF --><div class="md"><p>Wrote a deep dive on <strong>FlashAttention-4 (03/05/2026)</strong> that's relevant for anyone thinking about inference performance.</p>

<!-- SC_OFF --><div class="md"><p>Noob here, Running Qwen3.6 35B A3B in LM Studio on a 3080 10GB + Ryzen 5 3600 on Windows 10.</p> <p>Tried some unsloth quants with identical

<!-- SC_OFF --><div class="md"><p>I am sure many of you already know this, but using MSI Afterburner, you can change the voltage your single or multiple GPUs can draw, which can