SoylentNews is people
SoylentNews
from the 32-bit-should-be-enough-for-anyone dept.
This week, NVIDIA has announced that they are ending mainstream graphics driver support for Fermi-based GeForce GPUs. Effective as of this month (i.e. immediately), all Fermi products are being moved to legacy support status, meaning they will no longer receive Game Ready driver enhancements, performance optimizations, and bugfixes. Instead, they will only receive critical bugfixes through the end of the legacy support phase in January 2019.
While the announcement mentions 'Fermi series GeForce GPUs,' the actual support plan specifies that mainstream driver support is limited to Kepler, Maxwell, and Pascal GPUs. So presumably all Fermi products are affected.
In the same vein, also effective this month is NVIDIA dropping mainstream driver support for 32-bit operating systems, as announced in December 2017. Like Fermi, 32-bit operating systems will still receive critical security updates through January 2019. This update also encompasses GeForce Experience, which will no longer receive software updates for Windows 32-bit operating systems.
Previously:
Nvidia to Stop Writing Drivers for 32-Bit Systems (Eventually)
(Score: 1, Informative) by Anonymous Coward on Monday April 09 2018, @12:14PM (14 children)
It just means their code is trash; they haven't used the proper abstractions to ensure (or at least to ease) portability, and thus it's simpler for them just to throw away support for 32-bit machines altogether.
(Score: 0, Troll) by Anonymous Coward on Monday April 09 2018, @12:18PM
Because 32 bit machines don't have Intel management engines, ya big dummy.
(Score: 5, Interesting) by ledow on Monday April 09 2018, @12:53PM (7 children)
It's jumping-through-hoops.
The graphics cards now have more memory than a 32-bit bus can directlyaddress (4Gb), and then you're trying to map that so that the graphics drivers can shuttle data between 32-bit applications and the graphics card, and that just means more and more and more layers and limitations on size as you go forward.
Games are stopping support for 32-bit anyway (isn't GTA V 64-bit only?). Operating systems are all 64-bit and have been for years. There's little left of 32-bit legacy now, except hangers-on and OS that are nearing end of support (which each have a 64-bit version anyway). Gamers eager to use the latest nVidia cards to their full extent, especially, are going to struggle on 32-bit only systems, or 64-bit hardware running a 32-bit OS.
The question is not "why" but "why not"? Few people will be affected that care. Most people only use the card and OS that came with the machine, they won't be affected. Those people who want latest-and-greatest will need 64-bit anyway (and good luck getting anything else installed with a new machine). All the hardware has been 64-bit capable for years. All the server OS are 64-bit even, nobody is up-in-arms about it.
Basically, if it affects you, you're going to hit other problems first and can carry on using the older drivers anyway. It's not like your support is stripped out and nothing works. You just won't be able to run latest drivers (but the ones you have will do anything you can already do). But if you buy a new card, you're going to need a 64-bit OS, which isn't unreasonable some... what... 6 years after things like Server 2012 demanded a 64-bit OS to install, 10+ years after all the chips were 64-bit capable, and 10+ years since all the OS's supported 64-bit in some fashion (I deliberately don't count XP 64 as it was a heap of junk).
And when 64-bit is tied in with the availability of new instructions, the ability to map a full 32-bit of address space to each process for virtualisation, the hardware bus underlying width, etc. and you want to have things like 8, 16, 24, 32Gb of Video RAM alone... why would you want to keep supporting a 32-bit shim for a 2018 driver that only adds features / card support if your card is very recent anyway (and everything else is bug-fix).
(Score: 0) by Anonymous Coward on Monday April 09 2018, @01:16PM (2 children)
If the customer wants to deal with those constraints, then that's the customer's business.
The customer is always right. Provide a software interface for your device, and then fuck off.
(Score: 0) by Anonymous Coward on Monday April 09 2018, @02:25PM (1 child)
(Score: 0) by Anonymous Coward on Monday April 09 2018, @02:58PM
Surely that brings us back to what the OP was saying: This just shows that Nvidia has poorly-designed software driving these things.
The whole point of software is that it is soft; it's supposed to be cheap to customize.
(Score: 2, Interesting) by Anonymous Coward on Monday April 09 2018, @01:19PM
The only systems supporting *DIRECT* addressing of all memory on a card are Intel X79 or newer motherboard chipsets and AMD Ryzen hardware. Everything older only supports 32 bit BARs and in most cases has a far smaller PCI apterture available that in most BIOS/Operating systems. For everything else, paging is used and has been since the original Parallel PCI was defined back in the 1990s.
In most cases the video cards themselves have some form of GART or IOMMU that handles system bus to card memory address mappings and the drivers in turn have to manage either lists or formula that define where data is being placed into memory on the graphics card and how it relates to records kept by the driver in main system memory.
Having said that: Basically all post-Kepler Nvidia hardware makes sense to focus on 64 bit drivers, since they are not included in any systems designed for less than 4 gigabytes of RAM. While I myself am a 32 bit holdout as far as userspace goes, its time is long since passed outside of netbooks and a few other items which are still being kept artificially limited memory-wise due to cost/marketing considerations rather than technical capabilities. Within a year Intel is dropping legacy bio and emulation support, at which point the IBM PC platform as we knew it will be dead.
I can only hope at that time we will see an upstart buck UEFI and help bring us into a more enlightened time with the wane of WinTel's dominance. I don't expect it to happen, but one can always dream.
(Score: 2) by SomeGuy on Monday April 09 2018, @02:17PM (2 children)
Keep in mind that there is a 32-bit Windows 10, and that some people do in fact use it. While I'm usually the last to know, I haven't heard anything about 32-bit Windows 10 going away.
Then again people using 32-bit Windows 10 probably aren't using silly video cards that need eight nuclear power plants to run them.
(Score: 2) by vux984 on Monday April 09 2018, @03:58PM
"Then again people using 32-bit Windows 10 probably aren't using silly video cards that need eight nuclear power plants to run them."
Right. 32bit windows 10 exists mostly for some little tablet devices, low spec hardware, embedded, and and backwards compatibility. None of those entail sticking in a next generation nvidia gpu into it.
(Score: 2) by toddestan on Wednesday April 11 2018, @03:28AM
Actually, you have that backwards. The people using 32-bit Windows 10 are doing it because they need to use some silly video card to run eight nuclear power plants. Said silly video card is some horrible propriety thing that cost well into 5 figures and only has drivers that were written for Windows 2000.
I wish I was kidding.
(Score: -1, Flamebait) by Anonymous Coward on Monday April 09 2018, @12:59PM
Porting game specific hacks to two architectures is probably to time consuming for their curry interns.
(Score: 2) by mobydisk on Monday April 09 2018, @03:01PM (3 children)
The article is about video drivers. The drivers *ARE* the abstractions. They are often hand-coded assembly or C code. I'm curious who still has a 32-bit x86 CPU these days. I find most 32-bit Windows machines are people running 32-OSs on 64-bit machines and don't even know it. Or corporate deployments where they haven't updated their images for a long time. (And they are probably using intel GPUs)
(Score: 0) by Anonymous Coward on Monday April 09 2018, @03:17PM (2 children)
A 32-bit computer is perfectly adequate for the vast majority of computing tasks.
In the "old" days, when you wanted to perform some kind of special-purpose computation, like rending a polygon for a video game, you'd extend your 32-bit machine with a special-purpose computing device called a "GPU". Very many millions of these extensible, 32-bit machines were manufactured and exist in this world today.
(Score: 4, Interesting) by bob_super on Monday April 09 2018, @05:27PM (1 child)
These devices do not require driver updates to optimize performance in applications that use more memory than they can support.
These devices haven't been sold for quite a while, so, as GP points out, very few are still used for anything but menial tasks, requiring no new drivers.
Exceptions do exist, but catering for revenue-free exceptions is a good way to run run a general-purpose company into the ground.
Driver designers and testers don't work for peanuts.
(Score: 0) by Anonymous Coward on Monday April 09 2018, @06:33PM
That's it. That's all you need to do.
This bitching is nonsensical.
(Score: 0) by Anonymous Coward on Monday April 09 2018, @01:32PM (4 children)
A month ago
To add insult to injury, the nvid card won't work at all with nouveau... I guess it's HAMMERTIME!
(Score: 2, Informative) by higuita on Monday April 09 2018, @02:17PM (2 children)
This is always a problem with closed source drivers...buy a amd card next time, they are now using open source drivers that do work and they life is way bigger
(Score: 1, Interesting) by Anonymous Coward on Monday April 09 2018, @02:54PM (1 child)
The "open source" driver for all modern AMD cards depends on proprietary software -- most functions do not work at all without it.
(Score: 1, Informative) by Anonymous Coward on Tuesday April 10 2018, @07:49AM
Are you talking about firmware? Because then it's no different than the microcode in your CPU, or any other hardware for that matter.
The thing is, the firmware only depends on the hardware, where as the driver works with a specific OS-version. A company no longer making new firmware for a product only means no more bug fixes, things that used to work won't suddenly stop working. Where as when they stop updating drivers for new OS versions, the next OS update can render the hardware unusable.
That's why lots of people had to buy new hardware when Windows 10 was released - no Windows 10 compatible drivers - or install Linux, because open source drivers means that hardware support won't be discontinued over night.
(Score: 1, Interesting) by Anonymous Coward on Monday April 09 2018, @02:34PM
I got a couple of 'junk' Tesla cards (GT218) and tested them recently with Mesa 18.x Other than the cards themselves running hot, they seem to work fine with most OpenGL apps that the R600 driver works with, but have none of the Computer support.
So at least as far as linux is concerned nouveau is usable on GT, GK, and probably GF (I didn't have a card to test.) up to OpenGL 3.3 or 4.x depending on the card, but if you need CUDA or OCL GPGPU support, you are screwed if the official Nvidia drivers have dropped support.
Given that even on AMD hardware the Mesa drivers only support OpenCL 1.1 ATM you can't run either Virtual Currency Mining, or the majority of OpenCL based applications with the Open Source drivers STILL, I am not sure what the future holds. Given the push to Vulkan we may just see all this OGL/OCL support thrown to the wayside, because Vulkan is easier to maintain and all that hardware is obsolete, so why bother anyway.
I am not looking forward to having to upgrade to Vulkan hardware given how both the Nvidia and AMD GPUs have become locked down since Vega and Maxwell V2...
(Score: 0) by Anonymous Coward on Monday April 09 2018, @03:52PM (1 child)
I dropped all plans to buy any more Nvidia kit years ago. The company sucks, and their support for FOSS sucks. The nouveau driver sucks.
Like 90% of PC users: I don't play games, and I have no need of 3D. I have no plans to get suckered into mining crapto-currencies.
Lets hope Nvidia drowns in their own press releases and takes Oracle and MS down with them.
(Score: 0) by Anonymous Coward on Monday April 09 2018, @04:05PM
... then pisses all over free-and-open-source Money.
(Score: 2) by KritonK on Friday April 13 2018, @11:17AM
Oddly enough, four days after the announcement, nVidia released a newer [geforce.com], beta version of their driver, which seems to still support Fermi GPUs. It is only available for 64-bit OSes, though. Perhaps they did this because this new driver contains both support for newer cards and a critical update for Linux (older versions will not compile under Kernel 4.16). Since this version had already been in the pipeline, and both the current and the legacy drivers were due for a refresh anyway, they saw little point in not releasing what was already available.