This reviewer received promo considerations or sweepstakes entry for writing a review.
I picked this card up during a price match that brought it down to MSRP, mainly for running local large language models. On paper it’s slightly faster than the 5090 Founders Edition, and my own benchmarks back that up. Using ONNX with DirectML, I saw single precision scores around 42,000, half precision around 61,000, and quantized around 31,000. For comparison, my 3090 scored about half of that with single precision in the 21,000 range and half precision around 29,000. The improvement in AI workloads is clear and consistent.
The only real issue is the size. This card is enormous. It barely fit in my full tower case, and the four way power adapter made cable management difficult. The adapter is bulky enough to press against fans and block airflow. I strongly recommend using a single cable that can handle the full power draw instead.
I also tested a 5090 Founders Edition and found performance only slightly lower, but it’s far smaller and much easier to install. If you can find this card on sale or with a price match, it delivers amazing results, but if case space or airflow matters, the Founders Edition offers nearly the same performance in a more practical design.
