Tuesday, December 26th 2017

NVIDIA Forbids GeForce Driver Deployment in Data Centers

NVIDIA recently updated the end-user license agreement (EULA) for their GeForce Software. There's one particular statement in the limitations section that caught our eye. And it reads: No Datacenter Deployment. The SOFTWARE is not licensed for datacenter deployment, except that blockchain processing in a datacenter is permitted. It seems that NVIDIA isn't too happy with data centers that utilize GeForce and TITAN graphics cards instead of the more expensive Quadro or Tesla cards. With this prohibition in place, data centers are forced to either invest in NVIDIA's pricier offerings or completely switch over to AMD. Data centers that are using GeForce products for cryptocoin mining are unaffected by this change in the EULA.
Clearly, NVIDIA isn't wasting any time and has already started to enforce their new EULA. Sakura Internet, one of the largest data centers in Japan, was the first to receive a notice from NVIDIA to stop providing servers with TITAN X products.

Here's the Google-translated press release from Sakura:

Sakura's dedicated servers High-firing series Quad GPU new provision temporary suspension
December 21, 2017 Dear customers, Sakura Internet Inc. Thank you very much for your continued patronage of Sakura Internet.

On November 30, 2017, the licensing terms for the use of NVIDIA Corporation's driver software have been revised and the license terms for the latest GeForce driver software. The provision of "prohibition of introduction to the data center" has been added. For details, refer to Article 2.1.3 from the following URL.
  • Japanese:
http://www.nvidia.co.jp/content/DriverDownload-March2009/licence.php?lang=jp&type=geforcem
  • English:
http://www.nvidia.com/content/DriverDownload-March2009/licence.php?lang=us&type=geforcem

In addition, we received written notice from NVIDIA Corporation. According to this notice, NVIDIA Corporation agrees to the above license terms on the GPU server service (Sakura's dedicated server high-fire series Quad GPU model) equipped with TITAN X provided by the Company, Based on the view that downloading the driver software for GeForce on the server is an infringement of copyright (reproduction right). We urge customers who have downloaded it on or after December 7, 2017 to stop offering the Quad GPU model.

We are currently considering NVIDIA Corporation's notice content with experts as well, but considering the possibility of inconvenience to our customers, we are considering the following "Sakura's dedicated server. We will temporarily suspend the new provision of the high-fire series Quad GPU model ".
  • Quad GPU (Pascal) model: TITAN X (Pascal architecture) installed
  • Quad GPU (Maxwell) model: TITAN X (Maxwell architecture) installed
TITAN X non- loading models (TESLA V100 model, TESLA P100 model, TESLA P40 model). We will continue to offer.

We are sorry to cause inconvenience, but we will do our utmost to make it possible for our customers to use our services with confidence. We sincerely appreciate your continued patronage. Sources: NVIDIA, Sakura Internet
Add your own comment

90 Comments on NVIDIA Forbids GeForce Driver Deployment in Data Centers

#2
NdMk2o1o
Translated: We don't want you using consumer grade GPU's that are just as good as 5/10x more expensive workstation cards at certain workloads.

Time to use modded/3rd party drivers as it seems it's a software agreement that's being breached and not a hardware one, basically you can't use standard consumer Geforce drivers in an enterprise environment
Posted on Reply
#3
infrared
Ouch, that's gonna be a real curb stomp for data centers that already spent their budget on GF/Titan gpus :eek:

I wonder if this would encourage them to gear up with AMD gpus. Idk, I'm fond of the green team but this is a bit nasty.
Posted on Reply
#4
ensabrenoir
....not Pleased with any of these companies...from Intel to AMD and Nvdia.
Posted on Reply
#5
eidairaman1
The Exiled Airman
They want companies to spend more is why. Just another trash Tactic of Team Greedy Envy.
Posted on Reply
#6
Fluffmeister
I thought GeForce and Titan products were "gimped"? I wonder what these Data Centers are using them for, how many they buy and why AMD get no love in said market.
Posted on Reply
#7
OSdevr
Fluffmeister said:
I thought GeForce and Titan products were "gimped"? I wonder what these Data Centers are using them for, how many they buy and why AMD get no love in said market.
I think Geforce have handicapped double precision compute but Titans don't.

As for AMD, my guess would be CUDA.
Posted on Reply
#8
close
Wouldn't his affect only future versions of the driver? Usually EULAs aren't updated retrospectively.
Posted on Reply
#9
Fluffmeister
OSdevr said:
I think Geforce have handicapped double precision compute but Titans don't.

As for AMD, my guess would be CUDA.
Well DP is available on the original Keplar based Titan/Titan Black and most recently the bad boy that is Titan V, but all others fall short in that regard.

This is part of the balancing act of offering professional grade hardware to the average consumer, without damaging their high-end Quadro and Tesla brands.

But for sure CUDA is massive, Nvidia's big boss man has said they are a software company after all.
Posted on Reply
#10
evernessince
NdMk2o1o said:
Translated: We don't want you using consumer grade GPU's that are just as good as 5/10x more expensive workstation cards at certain workloads.

Time to use modded/3rd party drivers as it seems it's a software agreement that's being breached and not a hardware one, basically you can't use standard consumer Geforce drivers in an enterprise environment
FYI using reversed engineered drivers for commercial purposes is also illegal. It circumventing their EULA and the reproduction of prioprietery code without permission, the penalties of which would be much stiffer.

It's fine for end users because no one is making money off the modified drivers. In this commercial case though, it would be solely to the determent of Nvidia and they have every legal right to prevent modification of their driver code for commercial purposes. After all, it is their IP.

If you don't like it, don't buy Nvidia.
Posted on Reply
#11
OSdevr
evernessince said:
FYI using reversed engineered drivers for commercial purposes is also illegal. It circumventing their EULA and the reproduction of prioprietery code without permission, the penalties of which would be much stiffer.

It's fine for end users because no one is making money off the modified drivers. In this commercial case though, it would be solely to the determent of Nvidia and they have every legal right to prevent modification of their driver code for commercial purposes. After all, it is their IP.

If you don't like it, don't buy Nvidia.
This depends on how the reverse engineering is done, even if specifically prohibited in the EULA. Compaq reverse engineered IBM's BIOS at the dawn of the PC era, even though it was open source. They were legally required to.
Posted on Reply
#12
Solaris17
Creator Solaris Utility DVD
I guess for linux mining clusters this is a big deal, but you havent been able to install the drivers for any nvidia card or AMD for that matter on a Windows server OS without work arounds for sometime. (non-workstation that is)
Posted on Reply
#13
newtekie1
Semi-Retired Folder
This definitely sucks for people using these cards in data centers. But I'm more interested what a server in a data center needs with 4 Titan X cards, or any Titan X card for that matter. Other than mining.

The only thing I can think of would be something like Leap Computing Cloud gaming thing, but that seems like a pretty niche product that never really caught on. But maybe it did in other countries like Japan?
Posted on Reply
#14
OSdevr
newtekie1 said:
This definitely sucks for people using these cards in data centers. But I'm more interested what a server in a data center needs with 4 Titan X cards, or any Titan X card for that matter. Other than mining.

The only thing I can think of would be something like Leap Computing Cloud gaming thing, but that seems like a pretty niche product that never really caught on. But maybe it did in other countries like Japan?
Well other than support what advantages do Quadro and Tesla have over a Titan? Titan has it's DP capabilities intact while GeForce doesn't.
Posted on Reply
#15
evernessince
OSdevr said:
This depends on how the reverse engineering is done, even if specifically prohibited in the EULA. Compaq reverse engineered IBM's BIOS at the dawn of the PC era, even though it was open source. They were legally required to.
FYI Reverse Engineering is nothing like modifying

"The Compaq was the first sewing machine-sized portable computer that was essentially 100% PC-compatible. The company could not copy the BIOS directly as a result of the court decision in Apple v. Franklin, but it could reverse-engineer the IBM BIOS and then write its own BIOS using clean room design."

https://en.wikipedia.org/wiki/IBM_PC_compatible

Compaq essentially took apart their code to study it and then made their own BIOS from what they learned. Completely different than simply modifying Nvidia's drivers to bypass restrictions. That difference being, Compaq's BIOS was completely written by them while simply modifying Nvidia drivers means that 99.9% of the drivers were still done by Nvidia. Unless you think the comnunity can do a complete re-write of Nvidia's drivers, any company doing as you advise will be sued in a heartbeat.
Posted on Reply
#16
OSdevr
evernessince said:
Unless you think the comnunity can do a complete re-write of Nvidia's drivers, any company doing as you advise will be sued in a heartbeat.
I was simply pointing out that not all RE is illegal. I didn't mean to imply anything. BTW the nouveau team is doing exactly this (and it is indeed a massive effort). I nitpick :)
Posted on Reply
#17
mcraygsx
Does this mean my Server 2016 Standard will not working with my 1080 Ti any longer?
Posted on Reply
#18
newtekie1
Semi-Retired Folder
OSdevr said:
Well other than support what advantages do Quadro and Tesla have over a Titan? Titan has it's DP capabilities intact while GeForce doesn't.
This is really only true of Titan V and the first Titan. All the other Titans in between have had their DP capabilities locked.

mcraygsx said:
Does this mean my Server 2016 Standard will not working with my 1080 Ti any longer?
Is your Server 2016 machine in a datacenter?
Posted on Reply
#19
Fourstaff
Will datacentres now go to the red camp now?
Posted on Reply
#20
newtekie1
Semi-Retired Folder
Fourstaff said:
Will datacentres now go to the red camp now?
The question would be what would they switch to? If they are running a bunch of Titan X and Titan Xp cards, because they want the performance and don't want to pay the price premium, then AMD doesn't have much to offer as an alternative. Yeah, they could switch to a Vega64 or even Vega Frontier Edition, but they'd also be losing a good 20-30% in performance. I mean, basically nVidia knows these companies have no real alternative, and are milking it while they can.

So these companies will likely have to switch to Quadro cards. Or I wonder if the soft-mod to the Quadro drivers to allow you to install them on Geforce cards still work. I remember all you had to do was unpack the quadro driver, and modify an ini file by adding your Geforce GPU ID, and then the Quadro driver would install on Geforce cards. Obviously it wouldn't add any Quadro functionality, but it might get around the licencing issue. Of course, modifying the driver might bring up other licensing issues.
Posted on Reply
#21
eidairaman1
The Exiled Airman
newtekie1 said:
The question would be what would they switch to? If they are running a bunch of Titan X and Titan Xp cards, because they want the performance and don't want to pay the price premium, then AMD doesn't have much to offer as an alternative. Yeah, they could switch to a Vega64 or even Vega Frontier Edition, but they'd also be loosing a good 20-30% in performance. I mean, basically nVidia knows these companies have no real alternative, and are milking it while they can.
Are they getting loose or are they losing?
Posted on Reply
#23
londiste
newtekie1 said:
The question would be what would they switch to? If they are running a bunch of Titan X and Titan Xp cards, because they want the performance and don't want to pay the price premium, then AMD doesn't have much to offer as an alternative. Yeah, they could switch to a Vega64 or even Vega Frontier Edition, but they'd also be loosing a good 20-30% in performance. I mean, basically nVidia knows these companies have no real alternative, and are milking it while they can.

So these companies will likely have to switch to Quadro cards. Or I wonder if the soft-mod to the Quadro drivers to allow you to install them on Geforce cards still work. I remember all you had to do was unpack the quadro driver, and modify an ini file by adding your Geforce GPU ID, and then the Quadro driver would install on Geforce cards. Obviously it wouldn't add any Quadro functionality, but it might get around the licencing issue. Of course, modifying the driver might bring up other licensing issues.
Quadros or Teslas, depending on what the use for cards is.
Soft-mod Geforce/Titan cards into Quadros/Teslas can still be done but would be a similar breach of terms.

remixedcat said:
Wtf nvidia?!?!?!
These kinds of limitations are not limited to Nvidia. AMD will do exactly the same the moment someone tries to fill a datacenter with Vegas instead of WX9100s, MI25s or SSGs.

OSdevr said:
Well other than support what advantages do Quadro and Tesla have over a Titan? Titan has it's DP capabilities intact while GeForce doesn't.
Drivers, software and support. There are practically no hardware advantages for Quadros and Teslas. The exception has been P100 with all the FP64 and FP16 performance but V100 now has a "consumer" version in Titan V so that is not the case any more either.
Posted on Reply
#24
john_
It doesn't matter. We love Nvidia, we promote Nvidia, we buy Nvidia.
Posted on Reply
#25
Bayonet
For anyone wondering, Geforce cards are very cost effective when used for single precision workloads such as deep learning. Sakura Internet's offering of Titan X (Pascal) GPUs is about 4x as cost effective as their best Tesla alternative. Source: https://www.sakura.ad.jp/koukaryoku/specification/
Posted on Reply
Add your own comment