Slashdot Mirror
Nvidia Wants To Prohibit Consumer GPU Use In Datacenters (theregister.co.uk)
The Register reports:
Nvidia has banned the use of its GeForce and Titan gaming graphics cards in data centers -- forcing organizations to fork out for more expensive gear, like its latest Tesla V100 chips. The chip-design giant updated its GeForce and Titan software licensing in the past few days, adding a new clause that reads: "No Datacenter Deployment. The SOFTWARE is not licensed for datacenter deployment, except that blockchain processing in a datacenter is permitted."
Long-time Slashdot reader Xesdeeni has a few questions: Is this really even legal? First, because it changes use of existing hardware, already purchased, by changing software (with potentially required bug fixes) agreements retroactively. Second, because how can a customer (at least in the U.S.) be told they can't use a product in a particular place, unless it's a genuine safety or security concern (i.e. government regulation)!?
Nvidia expects that "working together with our user base on a case-by-case basis, we will be able to resolve any customer concerns," they told CNBC, adding that "those who don't download new drivers won't be held to the new terms."
Long-time Slashdot reader Xesdeeni has a few questions: Is this really even legal? First, because it changes use of existing hardware, already purchased, by changing software (with potentially required bug fixes) agreements retroactively. Second, because how can a customer (at least in the U.S.) be told they can't use a product in a particular place, unless it's a genuine safety or security concern (i.e. government regulation)!?
Nvidia expects that "working together with our user base on a case-by-case basis, we will be able to resolve any customer concerns," they told CNBC, adding that "those who don't download new drivers won't be held to the new terms."
It's about pricing.
For roughly equivalent GeForce versus Quadro/Tesla, nVidia charges an arm and a leg more for the professional model. They have long been forbidding their partners from selling 'professional' (workstation/server) products and allowing to order GeForce with them.
Particularly this has been a sore spot for academia, where they always want the cheaper GeForce model and they inexplicably can't get it.. I work at a vendor and customers always assume it is us being rent seeking assholes. I'm happy nVidia is making it very clear they are the ones being assholes, not us.
Enforcement of this will be pretty much impossible without some tie-in to the OS or drivers. First of all, what's a datacenter? Cloud-based infrastructure, or a room of servers only running internally on a local network? How about a mining operation in someone's basement? A grad student running a small network of GPUs for some sort of academic research? Etc. Now, if they really want to enforce it, it can be done -- you'd have to tie the software and drivers to server-class platforms that people typically have to pay for. E.g., I've seen Chelsio do that with some of their iWarp NICs where iWarp is disabled on anything but Windows 10 Enterprise and the Microsoft server OSes (though in that case, Chelsio claimed that Microsoft forced them to do it). On the Linux side, that might not be a realistic option.
There are quite a few cases where you need access to the massive parallelism of a graphics card and don't care about the actual graphics part, and integrated graphics give less value for money. Render farms, machine learning, certain types of modeling (e.g. weather), etc.
and end up having some sort of issue running the hardware where NVIDIA could be held liable.
This. The amount of time I had to spend debugging application issues because someone had the great idea to run a rig stuffed with gaming cards 24/7 instead of using hardware that is certified to last is insane. You have penny pinchers that see better performance for less money on the gaming cards and fail to notice just how much corners they cut. I had some randomly hang after eight hours of constant use, a known issue that affects nobody using gaming cards as intended.
Worse I have coworkers who try to sell customers on gaming cards since they can't be bothered to optimize their shaders and the performance difference between those cards is nontrivial. I hope that puts a stop to that.
I recently tried to add an nvidia card to my workstation for a virtual machine, and it turned out that nvidia breaks the driver when they detect the card is in a virtual machine.
Specifically you get an unexplained "Code 43" error, and nvidia's excuse is that there is a bug which they will not fix. However if you spent some time to hide the VM, like removing hypervisor drivers, it would have magically worked. Unlucky as I am, it turns out nvidia also broke that workaround (at least it did not work for me).
There are 3rd party patchers for this thing: https://github.com/sk1080/nvid... which require a lot of involvement, and will probably break at the next update. Given so much effort by nvidia to make sure I would be unable to use the hardware I purchased, I gave up, and removed the nvidia card from the workstation.