Linux 5.9 Brings Safeguard Following NVIDIA's Recent "GPL Condom" Incident

[Shevchen]

You are not aware then Nvidia also does have pretty good OC margins as well undervolting. Only undervolting monster was Vega, on RX5xx AMD can get much less from tweaking then nvidia and i heard 5xxx serie is same case.

2nd no OEM will go ever for hand written undervolting/underclocking attempts. Literally no one.

3rd. Comparing 1050TI to RX5700 is stupid, and i doubt it would be still more efficient. Now 5600 serie underclocked maybe can compete against it, but bring from Nvidia 1650 and 1660 serie.

To you first point: You can tune every card.
To the second point: EVGA does it with its Kingpin editions
To your thrid point: Why is it stupid? Because the 5700Xt has 8GB vs 4GB of the 1050Ti? Or because its more expensive? We are talking about the demand of a single user here. Well - I can take the right to create another set of checkboxes too then.

[jeisom]

I read this whole thread complaining mostly about Nvidia by amd fans/pro oss and the only bug complaints on Nvidia were about prime. Thing is the amd fans are are complaining about more bugs on their systems, enough to make me happy with buying Nvidia.

As is birdie complained about his navi board being buggy and then got a bunch of responses about pre-navi boards work just fine with the exception of 1 post which was using linux 5.8 of which hasn’t been out 2 weeks yet.

[Paul Frederick]

You want a nation, and you want that nation to be free but you want no constitution that protects that freedom. Thats BSD. Thats why BSD failed.
The GPL not just gives freedom, it protects it successfully.

All Nvidia would have to do is to mainline their kernel driver, they could still have their proprietary user-land driver with all the game optimizations and workarounds they want to protect. AMD does the same.

Just who do you think writes Nvidia's open source driver? You chowderheads have no idea what is going on. The code Nvidia does not open up they license from other parties. So it is not theirs to just give away. They are doing the absolute best that they can. Which in my opinion is pretty good. But the open source community is free to do better, if they can. So far I have not seen anything even remotely competitive to what Nvidia is doing. Which is why I've run their hardware on Linux since 1997. I ran their binary driver while it was in beta. It was the best thing going then too.

[Paul Frederick]

A driver that doesn't prevent tearing is high quality? LOL

Learn how to sync.

[entropy]

Nvidia was the first hardware manufacturer that supported Linux. Just in case you did not know.

Cool. That means they had the longest time to come up with a reasonable FOSS driver.

[oiaohm]

First, big minus to phoronix for not saying it was Facebook's patch, not Nvidia's patch.

Its not that straight forwards Nvidia provided the interfaces for other driver code to use. And their documentation for use of those functions from their driver does suggest wrapping with no warnings not to-do it with GPL code. So there is a more than two in this Tango.

Third, then tons of people literally lie, like for example Nvidia doesn't have proper VRR support on linux (they do, I use it, they even did release freesync support on linux before AMD),

https://forums.developer.nvidia.com/t/adaptive-sync-causes-the-screen-to-go-blank-when-the-refresh-rate-drops-below-a-certain-value/79697

Having support first does not mean having working support. The hard reality is issues with VRR and Nvidia cards with Linux that are not mirrored by the Windows driver from Nvidia is still turning up. AMD has their freesync behave the same between windows and Linux at this stage. So VRR Nvidia has more work to-do.

or that open sourcing driver is not that hard (WTF? you know that AMD open source stuff is literally written from scratch because they couldn't open source their existing stuff?) Nvidia would have to do the same.

Not 100 percent true. Nvidia cards are used in a lot of super computers. If Nvidia was to release the firmware with documentation required to make drivers the parties running super computers would most likely write the drivers for them. Mostly to be able to optimise the heck out of it.

Yes there is a open source Nvidia driver that mostly is useless because they cannot get rights to ship the required firmware to fire cards up properly.

Then I see something like oiaohm said that it is not problem to passivly cool AMD, oh it is because green team according to Phoronix benchamarks etc is simply more energy efficient, even 12nm Turing is more efficient then 7nm RDNA1. Passivly cooling is limited by how much heat you output and here AMD does have problem.

This is true but is also true that the Nvidia passive cooled cards in fact have a undersized cooler on them so when you start using them they thermal throttle . Yes AMD does generate more heat yes this you straight up using a case designed for passive cooling yes it larger heat sink than you can put on a card and it turns out you need a larger heatsink than you can put on a card to get the max out the Nvidia cards as well.

Basically piotr there is a interesting problem of weight. Does not matter if you have a lower thermal output Nvidia card its still putting out more heat than you can deal with using a passive heatsink directly connected to the card for full performance. Basically there is threshold.

Passively cooling is limited by 2 major factors. 1 the size of the heatsink you can connect 2 generated heat output. Yes factor 1 the size of the heatsink you can connect is most commonly overlooked. It does not matter if it a AMD card or a Nvidia card if you want it to run to full performance using passive cooling you will be using a custom case designed for passive cooling as that is the only way you get a big enough heatsink. Once you start having a heatsink as a complete side of case or larger you start dealing with heat-sinks that you physically cannot directly mount on motherboard because they are too heavy for motherboard.

Passively cooling GPU to run well turns out to be hard with physical limits. What you have todo to get a AMD and Nvidia cards to perform to max with passive cooling is basically the same thing. If you are not going to that you are better off for motherboard stress to go with fan based coolers or water based cooling.

Please note I am not saying that if AMD released a card that run cooler this would not be a good thing. If AMD cards would run cooler they would be able to run fans slower and make less noise with air cooled versions. If you are saying the thermal difference between AMD and Nvidia helps with passive cooling that is bogus because both put out more heat than you can deal with using a passive cooler mounted to card. Once you stop using card mounted passive cooling and move to case based passive cooling most are massive 50 watts here or there is not a major thing.

Basically the passive cooling arguement is wrong like it or not. Its also deceptive people buy cards with integrated passive cooling for more money fan based and would be better off to keep the money difference for their case and get a proper passive cooling case instead. A passive cooling case gives you a chance of having a properly sized passive cooler for the GPU.

About security/backdoors etc. of open source vs closed source. I know people like Gynveal who fuzzed a lot of open source stuff and closed source stuff (pdf readers, stuff like ffmpeg etc.) and it doesn't seem like open source projects are higher quality in that aspect.

https://news.synopsys.com/2015-07-29-Coverity-Scan-Open-Source-Report-Shows-Commercial-Code-Is-More-Compliant-to-Security-Standards-than-Open-Source-Code

Really the difference is there. But the issues are mixed. Defects per line of code open source is generally lower than closed source.

0.61 vs 0.71 defects per 1000 lines of code with variations between open and closed projects way larger than that. So yes it possible just by doing fuzzing to get numbers that look close. Problem with fuzzing when you don't have the source code you are depending on some luck that you will guess what will find the defect.

Parties like Coverity have advantage over those doing fuzz compares due to them in fact getting access to the closed source software source code. Lots of coverity access is so different commercial programs can be sold to the USA government. Yes there are other groups by other governments who get the same kinds of accesses and their number all align. Open source has minor advantage in defect numbers over all. Of course that does not mean you cannot have a open source project that is bad.

Gynveal is not really in a position to look from the source code side on closed source.

[Paul Frederick]

As if 2020 couldn't get any worse there's more nonsense now. Linux users don't have enough problems. Now devs have to code crap that benefits no one into the kernel. Appreciate it. I guess that 0.9% desktop market share is starting to get to all of your heads? Keep winning! Nothing spells success like alienating a hardware manufacturer that has been supporting Linux with the highest performing 3D acceleration available for 23 years now. Way to go. My hat's off to all of you.

Here's an idea. Instead of pissing on Nvidia why don't you just write an open source driver? Oh that's right because you can't! That fucking slideshow called Nouveau is the best you can do. Good going clown world inhabitants.

[oiaohm]

Not true. AMD threw out their efficiency gains on 7nm to (again) throw the cards into redline power levels well beyond their sweet spot. A couple of MHz less and a bit of undervolting results in significant less power draw. A passively cooled card isn't designed to match the performance of a custom watercooled overclocking monster, but to hit the point of maximum efficiency without sacrificing too much performance. A lot of passively cooled cards are underclocked to meet thermal specifications of the cooling block.

Something to consider here when you use passive cooling cases instead of coolers mounted to cards you don't undervolting or run with less MHz. You really run the cards ball to the wall. Just a case based passive heatsink can in fact be big enough. There are passively cooler cases where you can overvolt and overclock cards because you have a passive cooler with some head room. How much a passive cooler can take is really dependant on heat-sink surface area so size of cooler is critical.

Lot of custom passively cooled cases using lots of heat pipes are in fact designed to go head to heat with custom watercooled overclocking with a lot higher reliability.

[JustK]

[...]
As is birdie complained about his navi board being buggy and then got a bunch of responses about pre-navi boards work just fine with the exception of 1 post which was using linux 5.8 of which hasn’t been out 2 weeks yet.

vainfo
libva info: VA-API version 1.8.0
libva info: Trying to open /usr/lib64/va/drivers/radeonsi_drv_video.so
libva info: Found init function __vaDriverInit_1_8
libva info: va_openDriver() returns 0
vainfo: VA-API version: 1.8 (libva 2.8.0)
vainfo: Driver version: Mesa Gallium driver 20.0.8 for AMD Radeon RX 5700 (NAVI10, DRM 3.37.0, 5.7.11, LLVM 10.0.0)
vainfo: Supported profile and entrypoints
VAProfileMPEG2Simple : VAEntrypointVLD
VAProfileMPEG2Main : VAEntrypointVLD
VAProfileVC1Simple : VAEntrypointVLD
VAProfileVC1Main : VAEntrypointVLD
VAProfileVC1Advanced : VAEntrypointVLD
VAProfileH264ConstrainedBaseline: VAEntrypointVLD
VAProfileH264ConstrainedBaseline: VAEntrypointEncSlice
VAProfileH264Main : VAEntrypointVLD
VAProfileH264Main : VAEntrypointEncSlice
VAProfileH264High : VAEntrypointVLD
VAProfileH264High : VAEntrypointEncSlice
VAProfileHEVCMain : VAEntrypointVLD
VAProfileHEVCMain : VAEntrypointEncSlice
VAProfileHEVCMain10 : VAEntrypointVLD
VAProfileJPEGBaseline : VAEntrypointVLD
VAProfileVP9Profile0 : VAEntrypointVLD
VAProfileVP9Profile2 : VAEntrypointVLD
VAProfileNone : VAEntrypointVideoProc

Note: Video decoding is done in mesa not in the kernel.

[cyrix]

I don't know what's wrong with all the guys criticizing Nvidia.

Nobody's forcing you to buy Nvidia. You can have a 100% open-source Radeon system, albeit broken and unstable. Just take a look at https://gitlab.freedesktop.org/drm/amd/-/issues. But people often needs to get their work done and they can't afford their DE crashing twice a day. That's what Nvidia drivers are for. They're closed but you get no issues. And this has been like that for almost 20 years. Remember what was Linux ATI/AMD like back in the 2000s, up to until a couple of years ago? That was the real middle finger here.

So, thank you, Nvidia.