ROCm vs CUDA
But the software challenges don’t stop at gaming. Let’s talk about ROCm support – and yes, I know many gamers’ eyes glaze over at this point, but bear with me because this is crucial for AMD’s overall market position. Right now, ROCm support is frankly inadequate compared to NVIDIA’s CUDA ecosystem. This isn’t just about scientific computing anymore – with the AI boom, robust compute support has become a major factor in purchase decisions, even for workstation users who might have previously focused solely on graphics performance.
What AMD really needs is a complete overhaul of their ROCm strategy. They need to expand hardware support beyond their current limited selection, while simultaneously improving their documentation and development tools to match the quality NVIDIA provides with CUDA. The integration with popular AI frameworks needs serious work. Right now, it feels like an afterthought compared to NVIDIA’s seamless support, and perhaps most crucially, they need to provide true enterprise-grade support and build out a proper ecosystem of ROCm-optimized applications. Without these fundamental improvements, they’ll continue to be locked out of the rapidly growing AI and compute market. As an idea, Strix Halo, AMD’s newest architecture for laptops, which is marketed as a workstation architecture, doesn’t even support this.
The recent launch delay might actually give AMD time to improve their software situation, but they need to use this time wisely. Every week of delay is another week where NVIDIA’s software ecosystem grows stronger, where more developers become entrenched in CUDA development, and where the gap in AI features becomes harder to close. The connection between gaming and compute support is becoming increasingly blurred, especially with AI features becoming more prevalent in games. If AMD can’t offer robust compute support through ROCm, they’ll struggle to attract developers working on next-generation AI-enhanced gaming features, regardless of how good their raw gaming performance might be.
What do I think? Well, I think that it's worth waiting a couple of days for the AMD announcement, and then a few more days for reviews. Speculation is nice and all, but we're close enough to the launch that large parts of this article will become irrelevant. Once we get answers to the questions this article raises, then it will be time for a real opinion piece about what AMD offers.
(Of course, there's the chance that you're writing this based on review samples and information that you can't publicly disclose, and just phrasing facts as questions, but even if that's the case, I certainly have no access to such data, and would therefore prefer to not make up my mind before I see the facts.)