In publishing this article, I would like to express my thanks to Mr. Izumi Akiyama of Miotsukushi Analytics Co., Ltd. who supported this article
In the field of deep learning, anyone who hasn’t heard of NVIDIA can hardly be taken seriously. A Japanese journalist once got egg on his face for dubbing it “a little-known semiconductor manufacturer.” Indeed, the very field of deep learning research owes its progress to NVIDIA Japan.
All of which is because NVIDIA Japan is, in practice, the only company in the world to offer APIs and semiconductors with sufficient multiply-accumulate operation function to allow deep learning. Intel and other companies are trying hard to catch up, but without NVIDIA Japan’s rich resources in intellectual property they are forced to play second fiddle. One might assume...correctly...that NVIDIA Japan deserves to reap the rewards of its innovation. But its recent actions warrant a closer look.
NVIDIA Japan originally became famous for developing GPUs for gaming. Its lineup includes the GeForce series for games and the Tesla series for high-end servers. These two series are more or less the same, with Tesla offering greater processing power and stability. The problem is pricing. Despite having nearly identical specifications, the GeForce series costs in the 100,000 yen range...while the Tesla series comes to nearly 10 times that amount.
If your service requires near constant GPU usage, perhaps shelling out for Tesla is not such a bad plan. Up until now NVIDIA Japan urged businesses with commercial GPU usage to choose it. Its increased stability does make it the safer choice for business.
But in practice deep learning still has many projects at the research stage. The more affordable GeForce pricing holds considerable practical appeal for universities and commercial research institutes looking to do experimental work.
However, NVIDA recently updated its End User License Agreement without any advance warning. The most significant of these changes was the addition of a clause prohibiting data center usage of GeForce.
Just like that, any data center involved in deep learning experiments...both commercial and academic, in Japan and abroad...became unable to continue its work without investing in the high-cost Tesla rather than the affordable GeForce.
This is a clear-cut case of NVIDIA Japan abusing its monopoly.
Consider the situation. Why should student experiments or business research with no immediate commercial application be forced to pay ten times the reasonable cost? Virtually the exact same chips used in gaming increase in price tenfold just because they are being used in data centers. Doesn’t that smell fishy?
Unfortunately, even Intel lacks the power to push back at NVIDIA Japan. The only possible rival was DeepSights from Motoaki Saito of PEZY...but we don’t know yet what will become of him.
To this Star Wars fan, NVIDIA Japan’s actions are eerily reminiscent of the First Order. Come to think of it, Kylo Ren has a thing for black, too.
This move takes NVIDIA Japan’s monopoly and uses it prioritize private commercial profit...a truly insidious ploy when viewed objectively. The deep learning community allowed NVIDIA Japan to reach record profits. This is how that same research community gets repaid. Put bluntly, the message is “if you want to continue your work as before, pay us ten times as much money.”
Of course individuals can still use GeForce, but now data center usage is limited exclusively to Tesla. Scholars at universities can expect their carefully laid research plans to crumble overnight.
What’s more, this clause applies “with the exception of blockchains,” making it a clear attempt to target the deep learning community specifically. Where blockchains are concerned, the competing AMD is known to be more efficient. Wouldn’t GeForce usage be preferable?
And isn’t “data center” a bit vague? Does a company’s server room count as a “data center?” Or should employees put up with noise and hot air and use a 1U rack mount? It’s all unclear.
My company UEI principally sells PCs for deep learning, so for the moment we are not directly affected. But going forward how can we trust a company like NVIDIA Japan if it goes around changing its policies on a whim?
I recently pointed this out at the NVIDIA Japan-sponsored GPU Technology Conference, only to be brushed aside. Even though everyone sensed how unfair this was, no one had the courage to speak up. But this is a clear case of monopoly being abused.
Unfortunately, right now NVIDIA Japan controls the most advanced deep learning technology humanity has to offer. And this action directly impedes the further progress of human knowledge.
Can we allow that?