NVIDIA May Be Trying To Prevent GeForce GPUs From Being Used In Data Centers

[coder]

I take it you have never run a business of any kind. The purpose of a business it to, wait for it, MAKE MONEY and as much of it as it can. That's how they have money to pay salaries, bonuses, benefits, hire more people, R&D, etc.

I know it's become very fashionable to hate on the wealthy but it's the wealthy among us, those with the disposable income, that create jobs, bring new products and services to market and innovate.

Wow, someone sure has a thin skin. By calling it greed, I don't think he was necessarily making a moral judgement. I didn't interpret it as such.

Whether greed is good or not depends on context. Greed is merely a motivation that can be harnessed for the positive benefits you cite ...or can run amok and create much misery. More typically, it's not strictly one or the other, but that's getting OT.

[Brophen]

You forget that America is not the whole world.

Imho it's all to the good. Anyone remember what cryptomining did to the discrete card's retail prices? Inflated the fuck out of them and made getting cards hard. Now it's bit harder for corporate entities to grab gaming cards by the truckloads.. Real consumer wanting to buy 1-2 cards for upgrade wins from this.

Isn't that "ends justify the means?". What if Nvidia decided it liked cryptominers more than you and made a new driver update that had the license "This is not to be used for gaming"? Who is Nvidia to tell you what you can do with the card you BOUGHT from them? It's the principle

[DrYak]

1) Organizations that have need for GPU powered data centers could potentially just hire someone to create a custom driver based on the open source frame work currently available and 2) consumer video cards are probably poor choices for data centers

The counter argument to both points boils down to :
- not every company working with clusters and GPGPU is a huge billion-USD-rich oil company using the cards to do geological models simulation or whatever. Some of them are academics doing some life-science analysis (genomics, etc.)
- these might not have the expertise or even the resources to support the development of some nouveau-based cuda-compatible opencl clone.
- the point of GPGPU is to have increased computations throughput - meaning that nouveau-based new cuda stuff needs to be quite good to be of any use on Nvidia gfx cards.
- given their meagre financial resource, and the price point of consumer vs. pro cards - and depending on the actual GPGPU work-loads - using a general consumer gaming gfx card might actually might sense (slightly lower specs, much lower price)

So yeah, in short : some poeple do have needs that benefits from gaming gfx card, but not enough have the capability to go make their own drivers.

Anyone remember what cryptomining did to the discrete card's retail prices? Inflated the fuck out of them and made getting cards hard.

And the same crypto-miner are currently being exempted from the restrictions. So no escape from inflated prices.

Now, the question is : is this exemption due to AMD cards being a viable alternative and Nvidia not wanting to risk losing these customers ?
(A few years ago, AMD gfx card happened to have the most impressive hashing speed. If that's still the case today, Nvidia making the miners angry is at risk of throwing them in the arms of the competition).

[coder]

That's not true at all. Every capitalist market in the world is regulated by it's government. For the -very- reason to prevent both greed -and- communism. Government regulation is the trick discovered more than 2500 years ago.... It's not new....

Relax, I'm pretty sure @cl33r is being sarcastic. That or trolling.

[oiaohm]

Really why is any one surprised the writing was on the wall with vgpu support. In fact switching to AMD is no the only option.

https://www.top500.org/news/intel-launches-fpga-accelerator-aimed-at-hpc-and-hpda-applications/

We are seeing faster fpga and future massive number of core risc-v.


Its not possible to make a decent open source Nvidia driver that is legal while Nvidia refuses to hand over the power management firmware. It does have me asking the question is Nvidia driver in fact slower than the open source one and that is why Nvidia does not want to compete on the same clockspeeds?

Reality here is a bad time for Nvidia to attempting to be too profit grabbing it might end up their cards are not profitable against the fpga or risc-v options.

[duby229]

Really why is any one surprised the writing was on the wall with vgpu support. In fact switching to AMD is no the only option.

https://www.top500.org/news/intel-launches-fpga-accelerator-aimed-at-hpc-and-hpda-applications/

We are seeing faster fpga and future massive number of core risc-v.


Its not possible to make a decent open source Nvidia driver that is legal while Nvidia refuses to hand over the power management firmware. It does have me asking the question is Nvidia driver in fact slower than the open source one and that is why Nvidia does not want to compete on the same clockspeeds?

Reality here is a bad time for Nvidia to attempting to be too profit grabbing it might end up their cards are not profitable against the fpga or risc-v options.

I just read your post and wanted to give my opinion on the matter. I'd say it depends on your usage scenario whether to choose Nouveau or the nVidia driver. If you are strictly a desktop user the open source display stack is really good, Nouveau really does give a much better desktop experience. On the other hand if you need any kind of 3d engine performance you're gonna be better off choosing the nVidia driver. I don't know how performance per clock relates if at all, but the overall "feel" is much better with the nVidia driver.

[anarki2]

Meh. There's no "may be trying". They've 100% achieved this years ago. Not de juro, but de facto.

You can’t use consumer GPUs in datacenters

[theriddick]

Nvidia's strangle hold on their consumers tightens and continues, if only another company was able to compete at their level.. (1080ti++++) IF ONLY <stares hopefully at AMD>

[edwaleni]

Just use GPUBox. Community edition is free for up to 4 GPU's and uses distributed CUDA over Windows or Linux.

Works great over Infiniband.

[TheBlackCat]

Relax, I'm pretty sure @cl33r is being sarcastic. That or trolling.

I would like to think so, but on the other hand @cl33r appears to just be summarizing the apparently totally serious argument Spooktra made in the post directly above.