NVIDIA GeForce driver deployment in datacenters is forbidden now

This is a great time to remind everyone that AMD does not have this restriction for their Radeon graphics cards. Also, AMD has always been very supportive of the community and they&#x27;ve always respected their customers.<p>After decades of being the principled underdog, hopefully everyone can rally around them and make sure their open source projects work well with AMD products and contribute to their new open source initiatives.

So, my summary:<p>Unsourced Japanese news outlet published an article claiming this clause was added to the EULA. (Edit: located here: <a href="https:&#x2F;&#x2F;wirelesswire.jp&#x2F;2017&#x2F;12&#x2F;62708&#x2F;" rel="nofollow">https:&#x2F;&#x2F;wirelesswire.jp&#x2F;2017&#x2F;12&#x2F;62708&#x2F;</a> )<p>This version of the EULA, located at: <a href="http:&#x2F;&#x2F;www.nvidia.com&#x2F;content&#x2F;DriverDownload-March2009&#x2F;licence.php?lang=us&amp;type=GeForce" rel="nofollow">http:&#x2F;&#x2F;www.nvidia.com&#x2F;content&#x2F;DriverDownload-March2009&#x2F;licen...</a>, has the no-data center clause. Note the &quot;2009&quot;.<p>The version linked from the actual driver download page (at <a href="https:&#x2F;&#x2F;www.geforce.com&#x2F;drivers&#x2F;license" rel="nofollow">https:&#x2F;&#x2F;www.geforce.com&#x2F;drivers&#x2F;license</a> ), has no such clause.<p>I think I&#x27;ll postpone my outrage until the clause appears on the EULA that I actually have to agree to when I download GeForce drivers.<p>Alright, edit: For me, downloading drivers through <a href="https:&#x2F;&#x2F;www.geforce.com&#x2F;drivers" rel="nofollow">https:&#x2F;&#x2F;www.geforce.com&#x2F;drivers</a> gets me the second EULA that I linked to. However, downloading drivers through <a href="http:&#x2F;&#x2F;www.nvidia.com&#x2F;Download&#x2F;index.aspx?lang=en-us" rel="nofollow">http:&#x2F;&#x2F;www.nvidia.com&#x2F;Download&#x2F;index.aspx?lang=en-us</a> gets me a EULA with this data center &quot;limitation&quot;. This seems to me to be pretty problematic and an ineffective update.

<a href="https:&#x2F;&#x2F;github.com&#x2F;ROCmSoftwarePlatform&#x2F;hiptensorflow" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;ROCmSoftwarePlatform&#x2F;hiptensorflow</a><p>Everyone on HN knows to always go with free, open software. Without competition any company (Intel, nvidia, etc.) starts to exploit the consumer.

1. Write a Tensorflow wrapper that mints a new private cryptocurrency where the proof of work is training your deep learning model.<p>2. Sell it to companies who bought racks full of GeForce GPUs for deep learning.<p>3. Profit!

Is this even at all legal? Wouldn&#x27;t the First Sale Doctrine and the extremely, extremely limited rights to use software that courts have defended (the implied license to use software necessary for a purchased device HAS been defended in court) protect the purchaser? I really don&#x27;t think nVidia would be able to go after anyone legally for deploying the software in a datacenter. Companies can put whatever they want in software licenses and most of it is totally unenforceable bunk that just hasn&#x27;t been tested in court. And if it comes to it, they will normally drop the case or do whatever they can to avoid it ever being tested.

Linus tried to warn us! <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=_36yNWw_07g" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=_36yNWw_07g</a>

&gt;&gt; No Datacenter Deployment. The SOFTWARE is not licensed for datacenter deployment, except that blockchain processing in a datacenter is permitted.<p>I thought they were trying to mitigate the diversion of gaming gpus towards mining, apparently not.

Quick question: Why? What does NVIDIA have to gain by forbidding people from using GeForce for ML in datacenters?

If you want to know what Nvidia are afraid of, look at the last figure in this O&#x27;Reilly blog on distributed tensorflow.<p><a href="https:&#x2F;&#x2F;www.oreilly.com&#x2F;ideas&#x2F;distributed-tensorflow" rel="nofollow">https:&#x2F;&#x2F;www.oreilly.com&#x2F;ideas&#x2F;distributed-tensorflow</a><p>On a DeepLearning11 server (cost $15K), you get about 60-75% DL training performance compared to a DGX-1 (cost $150k)

What surprised me in the first place is why Tensorflow an &quot;OpenSource&quot; initiative by Google chose proprietary CUDA over &quot;OpenSource&quot; OpenCL.<p>Check Tensorflow issue 22 for more info.<p>Just sayin.

[Merry Xmas Nvidia! - Torvalds](<a href="https:&#x2F;&#x2F;4bds6hergc-flywheel.netdna-ssl.com&#x2F;wp-content&#x2F;uploads&#x2F;2012&#x2F;09&#x2F;Linus-Torvalds-Fuck-You-Nvidia.jpg" rel="nofollow">https:&#x2F;&#x2F;4bds6hergc-flywheel.netdna-ssl.com&#x2F;wp-content&#x2F;upload...</a>)

discussion here<p><a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=15983587" rel="nofollow">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=15983587</a>

What is a &quot;datacenter&quot;? I try to come to these things with an open mind but EULAs are really the dregs of legal writing :&#x2F;

This is one way to help the competition immensely. The premium nvidia charges for their tesla cards just isn&#x27;t worth it. It&#x27;s not enough of a performance advantage to warrant the price increase, and reliability wise you can by 6-10 1080ti&#x27;s for the cost of one p100.

So it will be profitable for a third party company to develop a GeForce driver.<p>This will also not be enforceable in all countries. It is probably not enforceable at sea where maritime law rules.

The NVIDIA website has a whole section on Data Center graphics cards: <a href="https:&#x2F;&#x2F;www.nvidia.com&#x2F;en-us&#x2F;data-center&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.nvidia.com&#x2F;en-us&#x2F;data-center&#x2F;</a><p>Won&#x27;t this just be hurting their own sales possibilities? I can only guess they&#x27;ll be announcing some datacenter-specific version of the software soon, which is essentially the same but mysteriously more expensive.

Was this actually an executive-level decision? The announcement of this change (if you can call it that) was so bungled it creates the impression some low-level person in legal did this without coordinating with the rest of nVidia.

Obligatory Linus Torvalds response, re: NVIDIA: <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=iYWzMvlj2RQ" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=iYWzMvlj2RQ</a>

If I understand the underlying issue here, it&#x27;s that Nvidia presumably doesn&#x27;t want it&#x27;s consumer graphics cards used in Datacenters, and instead wants to sell more expensive cards.<p>Isn&#x27;t this similar to how movie studios used to require Video Rental Stores pay more money for a film, vs the retail cost?

Somebody should write an adapter that hooks into their GPL shim and forwards requests over the network to a machine not in the data center that runs their binary blob. Trying to limit use of certain hardware via stupid legal means just begs for stupid technical solutions.

Eeh I have quite a few of these in data centers for Deep Learning work. What am I supposed to do now. Swap them out for Tesla&#x2F;Quadro cards? Switch to AMD or Xeon Phi? Or just ignore it and possibly violate ToS. Nice move Nvidia ...

Anyone else bothered by the 2009 in the url?<p>Edit: Some driver download pages currently link to this page.

“No Datacenter Deployment. The SOFTWARE is not licensed for datacenter deployment, except that blockchain processing in a datacenter is permitted.”

Just laugh and do it anyway.

This is why open source drivers are important.

&quot;Datacenter&quot; undefined.

That&#x27;s a great example of why we need open source hardware. 10-20 years ago this was a norm in software field. Now, you can license world class products with complete source for free.

They&#x27;re nuts and I&#x27;d love to see them try to enforce this. Coming soon to a DC near you: The Nvidia inspectors, picking locks on cages and racks looking for humping.

Sounds like one of those hilarious EULA clauses that is void in many countries with laws around companies not having the right to control the use of their product after sale.

Between this and the high Tesla prices, they really want to move people to use Google&#x27;s TPUs...

And THAT&#x27;S WHY it&#x27;s good for corporations to have closed non-libre drivers :) They can make an easy buck by selling &#x27;more expensive&#x27; licenses. You are the buyer, you decided for a closed solution, so pay more for it!

Nowhere in the document does it define &quot;datacenter,&quot; very nebulous term. Perhaps too nebulous to enforce.<p>I have 2 racks in my basement, if I had an NVIDIA GPU in one of my servers for playing games, am I violating their EULA?

Is there some legal reason why AMD can&#x27;t implement a CUDA compiler for their architecture? Seems like the obvious move right now.

I think the card itself takes more power and wears out faster compared to the data center versions. Is this true in practice?

I think this is designed to deal with the low prices offered by ovh and heizner. And in favor of Google and Amazon.

Does that blockchain clause target miners? Aren&#x27;t they called Mining Farms, not Data Centers?

How does NVIDIA define a datacenter exactly?

It&#x27;s beyond sickening to see one more product you buy attempt to control you via license. when will this shit end for hardware or is it just the start?