Imagine a world where if you wanted to occasionally run small business errands in your car, manufacturers could force you to buy their work-grade vehicles. You'd be pretty pissed off, right? Well, that's essentially what Nvidia did. GeForce users who wanted to pass their GPUs through to a Virtual Machine were suddenly greeted with the infamous Code 43 error when they tried to install drivers. And they've had to spend the last seven years dealing with janky workarounds, until today. Nvidia seems to have finally decided that it is time to make amends.
So let's learn a bit more and see if maybe, just maybe, Nvidia is turning over a new leaf. What I'll never turn over is my segues to sponsors like Honey. Honey is the free-to-use shopping tool that helps search for some of the best promo codes at lots of your favorite sites. Get it today. Join honey.com/ltt. (upbeat music) Code 43 has been a huge deal in certain circles for a long time, in no small part because of how ridiculous it is, because there's no technical reason why like most PCI Express devices, a GeForce card can't function just fine when passed through to a Virtual Machine, like in our 2 Gamers, 1 CPU project. And there's definitely no reason why a Quadro that uses the exact same silicon and often the same board layout would do it any better. And yet if the GeForce driver detected that it was running in a virtual environment, it would effectively say to you, what Linus Torvaldsfamously said to Nvidia. - So, Nvidia, you. - To understand why any of this matters though we need to go back to the beginning. Virtualization has been common on desktop computers since as far back as the nineties when Connectix VirtualPC allowed Mac users to use Windows on their PowerPC-based machines.
Now back then everything had to be emulated, so performance kind of sucked. But as technology improved and especially as hardware-assisted virtualization support became common, more and more of the host machines' resources became available with very little overhead to a guest operating system on a Virtual Machine or VM. That includes the ability to assign a device entirely to the guest using a feature called passthrough. Major players like VMware and Microsoft with Hyper-V have supported these features since very early on. And while hardware support for passthrough began in 2007 with AMD's K10 family of CPUs, in 2008 with Intel's Wolfdale, Chipset and BIOS support from motherboard manufacturers were sporadic on both sides. And it wasn't until AMD's900 series chipsets in 2012, and Intel's Haswell Refresh in 2014, that consumer hardware started really handling it. Well, AMD branded this feature AMD-Vi and Intel went with VT-d, but the technology is more commonly called input-output memory-management unit or IOMMU. And it is incredibly important for Virtual Machines. Not only does IOMMU allow pretty much any PCI device to be passed through to a VM, it can be used to sandbox faulty or even malicious hardware without worrying about compromisingthe rest of the system. This makes it useful, not only from a functionality standpoint, but also for security. Oh, and performance too. While devices like networkadapters and drive controllers could be emulated, passing through a dedicated card, could save that emulation overhead and allow a Virtual Machineto have IO performance that was nearly on parwith a bare metal one. Over time, however, this last point has become less and less relevant. Thanks to the broad adoptionof paravirtualization. The ability for very fast virtual devices to communicate directly with the host without the need to emulate or pass a full device through.
One network card for example,can service multiple VMs with very little overhead. However, one somewhat importantpiece of hardware still has no viable pair ofvirtualization support. The GPU. And one group that heavily relied on GPU passthrough back then,was the Linux community. It let them daily drive Linux and game in a virtual Windows environment with like bare metal performance. Perfect. That is until Nvidia's337.88 driver release in May, 2014, just daysbefore Intel announced the pass through friendly Haswell Refresh. This was one of Nvidia'sfirst major middle fingers to the Linux community, andit wouldn't be the last. But get subscribed becauseNvidia has been doing a lot of things latelyto try to make up for it and we've got some videosplanned to talk about them. The question though iswhy did they do this? Because whenever a newbie inevitably stumbled across Code 43, their outrage was swiftand it was consistent, because not only did itjust hand a win to AMD who decided not to laugh down Radeon, it was just totally ridiculous. I mean, it's not like passinga GPU through to a VM, unlock some super secretperformance enhancement that would directly eatinto Nvidia's Quadro sales, but apparently they thought differently, or at least thought the collateral damage was low enough not to matter. Lttstore.com by the way.
Now at the time, that bet probably looked pretty good because back, when this policy was enacted Ryzen, was but a twinkle in Lisa Su's eye and AMD was staring down bankruptcy. Quad-Core was as high as it went on the desktop and it would stay that way till 2017, so from Nvidia'sperspective, realistically, what gamer was gonna sacrifice precious course for this kind of use case when they had so few? Anyone passing through must be a professional user. But then everything changed. The overwhelming majority of new systems these days now have six cores or more and there are a growing number of people looking to legitimately take advantage of all that extra horsepower that goes woefully underutilized, and even the most multi-threaded games. Take someone like our camera guy, David. He wants his Android NAS togive him high-speed storage for personal video projects,act as a Plex media server, and serve as a capturePC for his streaming rig with Nvidia NVENC doing the encoding, using passthrough of course. Or take someone like Anthony who has somehow managed togame on Linux everyday at home since Proton came out, andyet still runs into games that straight up run only on Windows. But hold on a minute Linus, you might say, "Those are projects that youguys did before this change. "So what's the difference?" Okay, there were work arounds to Code 43. Some of them are even pretty simple but Nvidia's meddling causedno end of headaches for novices who were just as likely toabandon the idea entirely as to find a solution which sucks because it opens up a lot of possibilities and the hardware's freaking capable of it. Now though, it's as simpleas passing through your GPU and installing the driver as normal, just like you would do on AMD. So let's say you're aLinux user like Anthony. You can split your CPU andRAM up and then use an AMD GPU for its superior Linux experience, while using an Nvidia GPUfor, in my opinion still, superior Windows gamingexperience, completely hack free. Got Xbox Game Pass, I'm gonnaplay Microsoft store games? You can do that. Wanna play Epic Games Storeor EA Origin exclusives? You can do that too. Want real-time ray tracing and DLSS? Go for it. And this is really cool. With Looking Glass, you don't even need to changethe inputs on your display although it should be noted that at high resolutionsand refresh rates, you can run into PCI-Express bottlenecks. For our part though our Ryzen 5000 Series running PCI-Express Gen 4managed 1440P, 144 Hertz despite only running eightlanes to each of our GPU's. If you're okay switching inputs to play your Windows exclusives though, you don't need to worry about any of that because it's all native. There are some drawbacksto this approach however. While you do get the fullcapabilities of the GPU including features likeshadowplay and Nvidia Broadcast, you might still run into problems with anti-cheat software. Engines like BattlEye and Easy Anti-Cheat are able to detect when a user is running in a Virtual Machine. And the situation withcheating using Virtual Machines has unfortunately become bad enough that they've begun kickingplayers who are running them as a preventative measure. This is because anti-cheatrunning inside a VM has no idea what the hostoperating system is doing. And that host OS just happens to be able to manipulate the Virtual Machines memory in ways that areundetectable by anti-cheat hence the ban on VMs. They do say that users running in a VM will be banned outright, but there are reports ofexactly that happening. So just be careful if you decide to try playing Fortnite or Rainbow Six Siege. Are there ways around these limitations? Probably, but I'll leave that up to you to research on your own. For now, the bottom line isthat Nvidia is making changes to how they handle Linux and virtualization, and many people myself included could not be happier about it. I don't agree with everything they do, but damn it, you gotta give credit where credit is due.
Thanks for reading guys.
0 Comments