In the past few days, there has been an extremely heated discussion on a ban on the use of Nvidia, artificial intelligence and science and technology in the country. However, as the Nvidia announced official opinions yesterday, this incident has finally come to an end.
Only one conclusion - NVIDIA really 'wronged' this time.
Re-examine the new NVIDIA rules, the question is exactly where?
First of all, let's start with the "new deal" and the outbreak of the whole incident.
The new rules actually come from the NVIDIA GeForce driver download page, and users must check the appropriate license agreement before downloading the driver (similar to the one that must be checked before installing the software).
This also means that - all users of the GeForce driver software actually agree to this license by default.
Controversial content appears in the 'Restrictions' section of the 'Licensing' section, specifically The saying is 'The data center is not licensed to use this software unless the data center is doing blockchain.'
This content was first discovered and reported in real time by the German technology magazine golem.de and was posted on the Reddit forum by the user 'booooomba' but quietly changed in the press's description -
'The new NVIDIA EULA prohibits Deep Learning applications to run on GeForce GPUs. (The new NVIDIA EULA prohibits running deep learning applications on GeForce GPUs.)
It is unclear why exactly "booooomba" would choose this point of view to explain the new rules. However, the new rules do raise the following questions:
What does the term "data center" mean in this rule? And why the blockchain operation is excluded?
The other three items in the 'Restrictions' clause are common product rights. Why does one such specific provision appear in Article 4?
Next, Lei Feng network on these three major questions for further analysis.
Key question 1: What NVIDIA wants to limit what?
Yesterday, NVIDIA official reply to the Lei Feng network, you can clearly that one point - the new rules targeted at not the ordinary GeForce video card users, but large data centers, and in possession of these large data centers, often The major cloud service providers.
The NVIDIA official statement states: 'Users are not encouraged to improperly use our GeForce and TITAN products in harsh, large-scale enterprise environments.'
At the same time, NVIDIA also emphasizes: 'Researchers will typically use GeForce and TITAN products for non-commercial use or other non-data center-scale research purposes, and NVIDIA has no intention of banning these uses.'
Through these two statements , We can easily see that two of them need to meet the key criteria at the same time: whether commercial use, data center scale.
Combining these two conditions, all we can get is medium to large cloud service providers that are currently providing users with 'GPU Cloud Computing' services.
The key question 2: Why do you want to put the block chain a horse?
In many previous reports, many media have added a more or less 'description':
NVIDIA this time limits the data center, but let go of bitcoin mining !, and some friends even in the title of the big 'new NVIDIA ban: can not run GeForce graphics depth learning (mining can) 'But is it really so?
The first point to clarify first is that NVIDIA has never supported GeForce products for mining.
In the middle of this year, NVIDIA said directly on the activities of a video card brand: "NVIDIA does not encourage GeForce graphics card mining, users are not encouraged to do so. To mine, please use a special mining video card model."
Although so far this sentence has not been written in any regulation, but the NVIDIA in the middle of this year dedicated mining GPU is the best proof.
This year NVIDIA introduced the graphics card 'P106-100' for mining, which is 'modified' from the consumer market by the GeForce GTX1060 graphics card. Not only is there no hardware image output interface, but even the Dirext function (Microsoft's display API software, currently Windows All kinds of games on the need for this software to work) are also castrated off the physical level.
There is only one such purpose with great care - protect the consumer market as much as possible.
In the bitcoin's once-lucrative 2013-2014, AMD's HD7000 series had been out of stock for a long time.
But as bitcoin cool down dramatically in 2014 and 2015, more and more AMD graphics, once used in mining, are 're-entering' the consumer market and these batches of 'abusive' mining products have greatly affected AMD's spending Grade product market, the issue posed by the sale is to make AMD complacent.
NVIDIA has always been a vigilant in the wake of the digging boom. As ICO-driven 'mining boom' started earlier this year, many domestic NVIDIA graphics OEMs posted on the outside of their product packaging:
'If the product is used for mining, we reserve the right to refuse warranty and maintenance.'
From AMD's lesson, to the life of OEM manufacturers, and then to the special mining card P106-100 launch, NVIDIA hide mining can not hide it, how could it expressed its support for the attitude?
P106 mining card driver is actually modified by the GeForce driver
The question is, why the rules of this block will be operating on the block it?
In a video card industry professionals finally tell the mystery of the fruit -Not NVIDIA supports blockchain, but GeForce drivers are actually 'indirection' also used on mining dedicated graphics cards.
As for why not specifically get a series of drivers Well, NVIDIA no effort, do not want to spend that effort.
Key Question 3: Why does NVIDIA 'highlight' this requirement?
Although driven user agreement usually point everyone will agree, but NVIDIA will put the most basic provisions of numerous products together, the importance of this provision visible.
Why should NVIDIA 'highlight' this requirement? The immediate cause is NVIDIA official mentioned in the statement of product suitability concerns, more importantly, NVIDIA's desire to clarify the consumer market, business market differentiation.
Must be emphasized in advance that this is related to money, but it is not a simple question of money.
This is a common occurrence in the semiconductor market, from NVIDIA rival AMD to CPU boss Intel, which in fact has its own market segment, but of course the difference between consumer and enterprise markets.
Consumer-grade market users tend to be relatively simple to use, while demand for products is also more concentrated.
For example, GPU, as long as you have good graphics performance and stable operation on it.
But for enterprises, providing hardware is not enough. You also need to develop various hardware and software interfaces for them. You need to adapt various application scenarios. You need to provide various enterprise services and even engineers To the scene for technical assistance.
These wool should naturally be in the body of the sheep, also created a price gap.
NVIDIA solutions for different GPU acceleration applications
How to separately charge the two groups of sheep in the consumer-level market and the enterprise-level market separately and ultimately become a problem that the semiconductor industry enterprises must face.
It is common sense to charge more for a cloud service that generates commercial benefits and at the same time requires more product service.
On the other hand, 'consumer-grade products are cheaper than enterprise-level' can even be understood as 'NVIDIA collects enterprise-class marketing fees reasonably, which in turn reduces the cost of the consumer market.'
This is not only the choice of the company's overall strategy, but also the result of the final choice of market competition rules.
Additional instructions: consumer products really do not apply to the enterprise market it?
As mentioned above, one of the best explanations of the difference between the consumer market and the enterprise market is the difference between the two:
The consumer market GPU is usually installed in the chassis, you need to complete the entire cooling process, while using a shorter duration, lower frequency of use.
In the enterprise market, GPUs are often "plugged" into server racks, often with heatsinks and external power (wind, liquid) to complete the cooling process. The use of the GPU is very long, with very high frequency of use and stability requirements forever priority.
This also makes the products of both markets very different in all respects, such as appearance.
Take the GeForce GTX1080Ti and Tesla P40, which also use the GP102 core (different models) and the key processor parameters are very close. The former and the latter's appearance is completely different:
GTX1080Ti comes with a turbofan, while the tail of the graphics cooler does not allow air circulation, this cooling mode in the consumer PC market environment is very efficient.
The Tesla P40 uses a fanless, head-to-tail design, which can be arranged several times closer to the server without compromising heat dissipation.
In addition to the appearance of the difference, NVIDIA actually will take some potential effort.
Although the same materials and processes are used in the GPU production process, there are often some nuances in the finished product.
For example, some GPUs can work at lower voltages and some GPUs have higher overclocking performances. These subtle differences, combined with NVIDIA's differences in product parameters, eventually lead to the redundancy of server products .
Commercial mainstream cloud service is enterprise-class products
After considering many factors, I believe you may already have their own judgments, at least in the Lei Feng network seems:
This new rule is only a clear NVIDIA for their own products business, and the 'unfortunate' in the business, are not the cloud services industry rules' company.
Lei Feng Wang also specifically interviewed the domestic cloud services giant Ali cloud and Tencent cloud, Ali cloud that did not comment.
However, according to Lei Feng network query, Ali cloud has in fact maintained close cooperation with NVIDIA, NVIDIA's latest Tesla V100 series GPU has also been in the hands of Ali cloud, is being deployed.
Tencent cloud response is more direct:
'Building GeForce GPUs with Enterprise IaaS (Infrastructure as a Service) and then selling them is not a recommendation, and they are more focused on stability from the needs of enterprise-class users. The stability of the Telsa series, indeed the energy efficiency There are advantages. '
Asked about the value of NVIDIA's cooperation with them, Tencent Cloud Engineer said: 'NVIDIA is very supportive of our hardware.
Tencent cloud on this basis, through the depth of development, introduced a variety of products to meet the needs of users.
Interestingly, Amazon AWS introduced the NVIDIA 8-way Tesla V100 GPU accelerator in October this year and is offering it as a completely new cloud service to end users.
In this way, although using GeForce is more profitable on the books, the cloud service providers that build enterprise-class services in this way are in fact only 'heterogeneous' in the industry.
Written in the last: draft also feed a well
As the main 'power source' for parallel computing and even for the AI era, NVIDIA's several GPUs are driving the industry with their ever-evolving performance.
And NVIDIA does more than just create a product, develop hardware and software interfaces, provide application solutions, incubate start-up companies, hold technology summits, and more.
Nvidia's "monopoly" of Nvidia is not as good as saying NVIDIA's own technology continues to leap forward, creating a huge advantage in its products and services.
However, because an enterprise is restricting it rapidly in terms of technology and market and preventing it from being 'monopolized', is it in violation of the intention of technological development and market development or is it a genuine abuse of right?
