What's going on with AMD funding a CUDA translation layer, then nuking it? AMD's legal team appears to have clawed back control of much of the ZLUDA project's code base. The open source project, for which the House of Zen pulled support earlier this year, enabled compiled CUDA code to run natively on non-Nvidia GPUs. ZLUDA was introduced as a way to run unadulterated CUDA binaries on Intel GPUs. …
I don't think that was a broad/general statement about whether or not emails are legally binding, but rather a comment about this specific situation. Whether or not the email is legally binding matters less than "would you actually get enough (if any) money from winning the lawsuit to cover the legal costs." If there's any significant chance that the answer to the latter is "no", then yeah the former really doesn't matter. The problem here being that the codebase was FOSS. It has no monetary value so it would be difficult, if not impossible, to argue he suffered any losses he would deserve to be compensated for.
He was compensated by AMD for the work that he did prior to ending their contract with him, so what else could possibly be argued for damages?
It's more the case that he could *probably* get away with simply ignoring the takedown request, but that runs the risk of AMD filing a suit. Even if it's unlikely AMD would win assuming there really is a clause in the original contract, it would still most likely be a pyrrhic victory. He may get to keep the code online, but would be unlikely to make back all that he'd have spent in legal fees.
Nvidia's terms of service are not enforceable in many jurisdictions and I think they'll be reluctant to try and enforce it. We may also see test cases asserting copyright not only over the source code, but also the compiled form, at which the game will be over. Courts would generally, and CPU history is full of examples, prefer some kind of cross-licensing agreement between competitors. After all, even transpiled code is most likely to still run fastest on Nvidia's chips, but the market is currently more than big enough for both, with customers probably keen on a mix of interoperabillity and specialisation.
Not saying AMD is entirely the good guy here and I would hope for more transparency, but it would not surprise me if NVIDIA leaned on AMD to 'deal with' the 'problem' of ZLUDA. By the take-down originating from AMD it saves NVIDIA (some) of the PR backlash and potential vigilantism.
ZLUDA was licensed under a dual MIT & Apache license. I noticed on issues page the suggestion that GPL/LGPL would have been better, presumably because of GPL's "viral" nature. ;) I guess the idea is that any party that took issue with any of the content of the project would have to do their own "dirty work" and take the consequences of any negative publicity rather than the "back passage" strategy of indirectly coercing another contributing party.
I don't know whether CUDA is much used in AI/LLM to leverage NVidia GPUs but if it is, then a pretty obvious explanation comes to mind in that either, or both, of the two corporate parties imagine ZLUDA is giving away a valuable shovel in the AI gold rush.
It is said that the US is a Godless place but in reality it is extremely devout but unfortunately their god is Mammon and before whose altar all enthusiastically bow down in avaricious supplication.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
