Thought you guys might like to see another one I finished up for a good friend of mine. He had some really tough space requirements and we had to trade off a bit of performance for 2 of the GPUs for thermal reasons but it still puts out very impressive numbers, far better than I was expecting with the RTX flow through chimney design.
The top card steps down to 300-325watts, the 2nd card steps down to 400-450watts and the bottom two usually stay at 550-600watts when the system is at max load on GPUs. For normal day to day usage with small bursts of inference they all remain at 600watt without a problem which surprised me I thought for sure the top two would step down but they don’t for regular LLM usage. Training, or full bore parallel GPU usage is when the top ones heat up faster and they do throttle but again it faired much better than I was expecting.
I had to make a custom RAM duct as the RAM was getting way too hot and throttling, and now they stay in the high 40c range and are perfect.
Enjoy 
