It just goes to show that openness is more agile - if your code is open source then it can be repurposed quickly, by the people that need it _now_, to provide the results they want.
Whereas if your code is a massive blob then your developers are the bottleneck to any new functionality or usage.
If your code is your product, then that's manageable. But when your product is the hardware, and the code is there purely to make it functional, then retaining that level of control is counterproductive.
Not really. They may not have seen that support for other chip architectures would be a money maker. They obviously didn't think it would be enough of a money maker to spend money doing it.
The case at hand shows that if you have open source drivers though, that you gain access to an army of folks who get paid by other people to further develop your product. Here's where they were wrong (though perhaps not moronically so) and why they missed a huge order.
This is great. The more governments, companies, and other organizations demand open-source drivers from their hardware vendors, the greater the benefit will be to society. Virtually everyone benefits from more hardware devices being more compatible and easier to integrate with each other.
The didn't demand openness. The article is sensational.
> The problem is that the GeForce / Quadro driver from NVIDIA is only available for Linux x86 and x86_64 architectures, not MIPS or even ARM (only the Tegra driver is for ARMv7).
Openness is great, other things being equal, but in the graphics card market other things certainly aren't equal. A huge amount of R&D goes into the drivers rather than the cards themselves, because better drivers offer a substantial commercial advantage over the competition.
If that advantage is reduced -- or even becomes a liability -- because you gave away the secrets that you spent a lot of time and money learning, then there is no longer any reason to spend that time and money in the first place, and not even your own drivers will benefit.
This is the problem with almost any Internet debate about intellectual property or open development processes. It's easy to see the benefit tomorrow if you just force everything open today, but you have to consider the implications for next week as well.
In this particular case, a lot of people are calling NVIDIA names and assuming they're being stupid. It's entirely possible that they have competent management who have looked at the loss for this one deal, balanced it against the ongoing loss of disclosing their proprietary driver techniques without an adequate level of control, and simply decided that it wasn't a good deal. Sure, it's 10,000,000 "lost" sales, but the global PC market alone ships upwards of 300,000,000 units every year, so this deal isn't even close to big enough to reshape the industry.
Because it should be true! Ignoring for the moment the veracity or falsity of the rumor, the fact that open-source code is a better choice, even from the economic angle, should be the main point for discussion here, tangentially spurred by this rumor.
Well, actually, it feels like Nvidia deliberately refused to participate in this, as the price, which they requested for porting the driver sounds ridiculous.
I think they named the sum knowing that chinese will refuse to pay.
Every time I read this story about the Chinese government ordering 10,000,000 GPU's I think of cryptography and espionage - not Linus and Linux and open source software.
Don't get me wrong, I know that the Chinese government probably has a lot of computers. It's just that the driver issue isn't a deal breaker for a computer on a clerk's desk in the interior ministry. I find the assumption that Beijing has been taken over by Stallmanists dubious. If the selection of GPU's by the Chinese government has an underlying motive, odds are the motive is based on China's strategic interests.
I love nvidia, compared to ATI cards they are marvelous for what we use them for (video and graphics). But, my main gripe with them is Quadro cards - main difference (apart from minimum x5 markup on prices) is driver which is different from Geforce cards and sole reason for performance boost. It would be REALLY good for them to have competition in that space, but sadly ATI/AMD is not up to the par.
Now that Intel has announced Xeon Phi, I dare to hope that "gpu programming" will become an artifact of the past, so "open-source drivers" will become a moot point anyway. (Xeon Phi runs linux as a local OS.)
> Now that Intel has announced Xeon Phi, I dare to hope that "gpu programming" will become an artifact of the past
Your hope makes no sense, MIC is still SIMD and requires programming for SIMD, it's not fairies and rainbow magically turning code into SIMD-compatible instructions. Not only that, it's completely and utterly irrelevant to TFA because:
> so "open-source drivers" will become a moot point anyway
TFA is about GPUs, not SIMD GPGPU. GPUs need drivers. Even IGPs require drivers.
> Xeon Phi runs linux as a local OS
Xeon Phi does not run anything any "local OS", Xeon Phi is a CPU arch. A computer based on Phi may run linux as a local OS, that does not magically mean the OS needs not be aware of the architecture and how to communicate with it.
> Xeon Phi does not run anything any "local OS", Xeon Phi is a CPU arch. A computer based on Phi may run linux as a local OS, that does not magically mean the OS needs not be aware of the architecture and how to communicate with it.
I think that zvrba's post was referring to the fact that Xeon Phi cards are internally running an embedded version of Linux. The host system doesn't need to know about that technical detail due to drivers on the host system hiding that away, but he is still correct [edit: in that there is a Linux subsystem present.]
"Meanwhile on the software side of things in an interesting move Intel is going to be equipping Xeon Phi co-processors with their own OS, in effect making them stand-alone computers (despite the co-processor designation) and significantly deviating from what we’ve seen on similar products (i.e. Tesla). Xeon Phis will be independently running an embedded form of Linux, which Intel has said will be of particular benefit for cluster users."
Whereas if your code is a massive blob then your developers are the bottleneck to any new functionality or usage.
If your code is your product, then that's manageable. But when your product is the hardware, and the code is there purely to make it functional, then retaining that level of control is counterproductive.