I'm building and fine-tuning my own AI models and need some advice on which GPU to purchase. I'm leaning towards either the RX 9070XT or the RTX 5070 Ti. While both are similarly powerful, the RTX 5070 Ti comes at a much higher price. A big factor for me is CUDA, which is crucial for AI training, and as many of you know, AMD's ROCm, while supported by PyTorch, might not perform as well as CUDA. Given this, do you think the RX 9070XT can handle the workload effectively, or should I invest in the more expensive RTX 5070 Ti?
2 Answers
From my experience, CUDA is definitely the way to go if you're serious about AI training. It has better optimization for most frameworks. Although the RX 9070XT might seem like a good deal, the performance increase with the RTX 5070 Ti can be worth the extra cost, especially during heavy training tasks. If you can afford it, I'd say go for the RTX.
While the RX 9070XT can handle AI tasks, the RTX 5070 Ti has the edge, especially since CUDA is more established for these workloads. If you spend a bit more on the 5070 Ti, you’ll likely save time in the long run with faster training. Just think about how much you value that time versus the cost!
Thanks for the insights! It looks like I might just go with the 5070 Ti after all.
That makes sense! I've seen similar results with my setup too, CUDA just runs smoother.