Those are very worrying statistics. I am planning to upgade to a discrete GPU from the 5600G’s iGPU, I hope this doesn’t become a problem in the future.
It’s a problem already, right now. Prices are already ridiculous and I’m sure the Nvidia 5000 series will be even more so before AMDs 8000 series add fuel to the fire in an effort to retain their fledgling market share.
It’s the main reason I haven’t upgraded yet: I just don’t want to drop ~$1K USD on another GPU that can handle 4K 120fps displays
The problem is that NVIDIA isn’t resting on their laurels, they improve by a large margin from architecture to architecture and continue to innovate features. AMD can barely keep up imitating some of these features (upscaling, RT, frame generation, heck even NVENC is superior to what Radeon offers) and the results are often worse (RT performance, DLSS vs FSR).
AMD only barely undercuts NVIDIA’s pricing based on raster performance, so this is essentially the easiest upsell ever. Pay 15 % more but get better versions of features, new features early, broader compatibility also in terms of compute and more efficiency? Sure, most people will pay 15 % more for that.
AMD needs to be way more aggressive on pricing and try to innovate useful features first on Radeon. That being said, I think NVIDIA would simply price-match as soon as AMD gains any traction.
At this point I have more faith in Intel to be competitive in a few generations. They seem to be able to almost match RT performance, already putting AMD to shame with their first generation of Arc GPUs. Their upscaling tech is way closer to DLSS, Intel QSV is a pretty solid hardware encoder and let’s hope they do a better job competing at compute.
The problem is that Nvidia pulls proprietary nonsense out of their ass, specifically to say “AMD can’t do [blank]!”
Physics. Sound. Compute. Hair. Raytracing. Upscaling. And on and on and on, aggressively poised to be incompatible. CUDA is fifteen years old and Nvidia still threatens anyone who tries cross-compiling it to SPIR-V or OpenCL.
This is anti-competitive behavior from a blatant monopoly. They have supermajority market share - they are abusing it. The right answer is to shatter this corporation.
AMD needs a suitable alternative to CUDA cores.
AMD’s compute capabilities are fine. CUDA is an artificially restricted API that is not permitted to recompile into anything else.
Then it’s on AMD to develop a competing API, and ultimately on them for not being competitive enough to develop an alternative alongside the development of CUDA.
It’s not Nvidia’s fault they developed a better product. And I say this as someone who prefers AMDs products, although I had to get a 3060 for my 3D modeling. I would have loved to go AMD, but I really couldn’t.
Nvidia turning your own code into an anti-competitive lock-in is absogoddamnlutely their fault. Again: they actively prevent efforts to recompile CUDA into the competing APIs that already exist and work just fine - because they have to maintain their vice-grip on a market they got to first and turned into a weapon.
I would have loved to go AMD, but I really couldn’t.
THAT’S WHAT ANTI-COMPETITIVE MEANS.
The FTC needs to invalidate Nvidia’s monopoly on CUDA. AMD can’t compete because AMD isn’t allowed to compete because of copyright law.
You compete by having your own product, not by using what your competitors developed. It’s on AMD for not making their own product to compete with CUDA
What part of “programming languages are standards” do you not understand?
What you wrote is like suggesting that Internet Explorer shouldn’t have been allowed to implement JavaScript and that all websites should’ve been forced to decide between being compatible with only Netscape or only IE.
They have ROCm and ZLUDA but both are inferior to CUDA. They completely overslept the whole AI hype cycle and they could have been valued now to over 1 trillion.
