Linux Kernel 6.14 Is a Big Leap Forward In Performance, Windows Compatibility (zdnet.com) 33
An anonymous reader quotes a report from ZDNet, written by Steven Vaughan-Nichols: Despite the minor delay, Linux 6.14 arrives packed with cutting-edge features and improvements to power upcoming Linux distributions, such as the forthcoming Ubuntu 25.04 and Fedora 42. The big news for desktop users is the improved NTSYNC driver, especially those who like to play Windows games or run Windows programs on Linux. This driver is designed to emulate Windows NT synchronization primitives. What that feature means for you and me is that it will significantly improve the performance of Windows programs running on Wine and Steam Play. [...] Gamers always want the best possible graphics performance, so they'll also be happy to see that Linux now supports recently launched AMD RDNA 4 graphics cards. This approach includes support for the AMD Radeon RX 9070 XT and RX 9070 graphics cards. Combine this support with the recently improved open-source RADV driver and AMD gamers should see the best speed yet on their gaming rigs.
Of course, the release is not just for gamers. Linux 6.14 also includes several AMD and Intel processor enhancements. These boosts focus on power management, thermal control, and compute performance optimizations. These updates are expected to improve overall system efficiency and performance. This release also comes with the AMDXDNA driver, which provides official support for AMD's neural processing units based on the XDNA architecture. This integration enables efficient execution of AI workloads, such as convolutional neural networks and large language models, directly on supported AMD hardware. While Rust has faced some difficulties in recent months in Linux, more Rust programming language abstractions have been integrated into the kernel, laying the groundwork for future drivers written in Rust. [...] Besides drivers, Miguel Ojeda, Rust for Linux's lead developer, said recently that the introduction of the macro for smart pointers with Rust 1.84: derive(CoercePointee) is an "important milestone on the way to building a kernel that only uses stable Rust functions." This approach will also make integrating C and Rust code easier. We're getting much closer to Rust being grafted into Linux's tree.
In addition, Linux 6.14 supports Qualcomm's latest Snapdragon 8 Elite mobile processor, enhancing performance and stability for devices powered by this chipset. That support means you can expect to see much faster Android-based smartphones later this year. This release includes a patch for the so-called GhostWrite vulnerability, which can be used to root some RISC-V processors. This fix will block such attacks. Additionally, Linux 6.14 includes improvements for the copy-on-write Btrfs file system/logical volume manager. These primarily read-balancing methods offer flexibility for different RAID hardware configurations and workloads. Additionally, support for uncached buffered I/O optimizes memory usage on systems with fast storage devices. Linux 6.14 is available for download here.
Of course, the release is not just for gamers. Linux 6.14 also includes several AMD and Intel processor enhancements. These boosts focus on power management, thermal control, and compute performance optimizations. These updates are expected to improve overall system efficiency and performance. This release also comes with the AMDXDNA driver, which provides official support for AMD's neural processing units based on the XDNA architecture. This integration enables efficient execution of AI workloads, such as convolutional neural networks and large language models, directly on supported AMD hardware. While Rust has faced some difficulties in recent months in Linux, more Rust programming language abstractions have been integrated into the kernel, laying the groundwork for future drivers written in Rust. [...] Besides drivers, Miguel Ojeda, Rust for Linux's lead developer, said recently that the introduction of the macro for smart pointers with Rust 1.84: derive(CoercePointee) is an "important milestone on the way to building a kernel that only uses stable Rust functions." This approach will also make integrating C and Rust code easier. We're getting much closer to Rust being grafted into Linux's tree.
In addition, Linux 6.14 supports Qualcomm's latest Snapdragon 8 Elite mobile processor, enhancing performance and stability for devices powered by this chipset. That support means you can expect to see much faster Android-based smartphones later this year. This release includes a patch for the so-called GhostWrite vulnerability, which can be used to root some RISC-V processors. This fix will block such attacks. Additionally, Linux 6.14 includes improvements for the copy-on-write Btrfs file system/logical volume manager. These primarily read-balancing methods offer flexibility for different RAID hardware configurations and workloads. Additionally, support for uncached buffered I/O optimizes memory usage on systems with fast storage devices. Linux 6.14 is available for download here.
AMD XDNA driver for... what exactly? (Score:2)
Can anyone name an example?
Re: (Score:2)
I suppose it's useful for things like local image recognition and AI sorts of things. Very similar to capabilities Apple offers on their M series chips.
Re: (Score:2)
Re: (Score:2)
Tensor flow is one widely used library and I believe it can use amdxdna.
Re: (Score:2)
I use the built in intel gpu to do image recognition on security cameras, using Frigate.
It would work faster and I could use more accurate models if I had a Google Coral usb stick
Re:AMD XDNA driver for... what exactly? (Score:5, Interesting)
It's a tiling matmul ASIC, like others of this type.
On my Mac, I think it's really only used for Apple Intelligence shit, and neat shit like coming up with smart recommendations and classifications for my music and pictures. Nothing super fancy.
I can run Stable Diffusion on mine. Doesn't run quicker than the GPU, but it does use a small fraction of the power, but it requires the models to be in a really weird structure.
On phones, Android and Apple, they're typically used to create good HDR photos, classify images, suggest music, etc.
On Windows... I'm not sure. CoPilot?
Re:AMD XDNA driver for... what exactly? (Score:5, Interesting)
I do image processing for microscopy, spectral classification etc. Those NPUs are the reason why a large GPU cluster is not necessary for about 50% of the tasks I run. YMMV.
Only if it's a bottleneck (Score:3)
from: https://lwn.net/Articles/96188... [lwn.net]
The gain in performance varies wildly depending on the application in question and the user's hardware. For some games NT synchronization is not a bottleneck and no change can be observed, but for others frame rate improvements of 50 to 150 percent are not atypical.
So this could be everything or absolutely nothing, depending on the game and your hardware.
Re: (Score:2)
In benchmarks I've seen, most games get some improvement in maximum FPS, but it is usually just a couple of percent. Meanwhile, some games get massive gains. However, this is also a huge step upwards in compatibility and it may make some titles work which previously did not. Also, what's arguably more important than max frame rate (except of course for games which were formerly unplayable) is minimum rate. This will help improve that, again, for some titles.
TL;DR: fsync was already a big step upwards in per
Btrfs raid1 improvements (Score:5, Informative)
I'm excited to hear about the Btrfs raid1 improvements. I was dismayed to learn that raid1 read operations are not shared among the two discs. Each process is given a disk to read from, so a copy operation would only read from one disk. Now with this experimental option enabled, contiguous data will be read in chunks from all disks in a round robin fashion. More info here:
https://www.phoronix.com/news/... [phoronix.com]
Re: (Score:2)
Re: (Score:2)
Is there a serious advantage to using a file system's RAID implementation rather than just using 'md' combined with logical volumes?
I'm always mistrustful of a file system that implements RAID, they're trying to add too many features to a single package and thus massively increasing complexity and the risk of bugs. md, by comparison, is small and simple enough for devs to have a clear idea of what's going on and what the implications of each change are.
I'm not remotely surprised btrfs has been having proble
Re: (Score:2)
md RAID, various hwraids etc -- anything that doesn't talk to the filesystem -- can't recover from any errors other than 1. read errors that have been detected by the drive itself, 2. a complete failure of a drive with it never coming back. SSDs often return garbage instead of an error; especially a failure or a bug in the FTL will give you an unrelated sector. Cable failures happen, etc. If a drive comes back (after a transient error, reboot, or reseating), the RAID layer has no way to know which drive
Re: (Score:2)
I'm not sure how true that is. RAID implementations generally do some level of verification outside of what the drive reports, indeed the matrixed data used by RAIDs 5 and 6 requires implementations verify that data so a corrupt SSD that thinks its sending genuine data isn't going to be different from 'md's point of view to one reporting a bad sector.
Beyond that why can't a file system use its own checking in conjunction with a lower RAID implementation layer? If 'md' sees an error it attempts to fix it, an
Re: (Score:2)
why can't a file system use its own checking in conjunction with a lower RAID implementation layer? If 'md' sees an error it attempts to fix it, and if it can't it'll send it up the chain.
Does it, though? Does md ask the filesystem "this block is either 0xff or 0xef, please tell me which is correct"? Some internet searches lead me to believe that this scenario is unrecoverable, probably because md was designed before checksumming filesystems were popular.
Re: (Score:2)
Re: (Score:2)
I don't have any exotic system, it's an i7 7700k and a GTX3050. Installing the current ubuntu desktop can't be done with the standard settings.
It tries to load the graphical install screen and gets stuck.
Have to select safe mode to do the install. It's a pretty shit user experience when I started off with the idea of seeing if Linux was capable of running Windows games yet, now that Windows 10 security updates are almost over. The installer doesn't even have a functional graphics driver.
Re: (Score:1)
Re: (Score:2)
Re: (Score:2)
No good having a great operating system when so many people simply cannot install it.
Linux is not one operating system, and if you're confused about that, it explains how you could write this without irony.
It's sure not as easy as the fans make it out to be.
If you're talking about Arch or Devuan or any other non-mainstream Linux, that's true. But if you're suggesting that noobs run one of those, you're a dumbass.
For non-technical people, it's still almost impossible to install Linux easily.
Attach Ubuntu or preferably Mint install media, reboot to it, and hit next a bunch of times. OOH HOW HARD. What I don't get is why a non-technical person like you is here on Slashdot. What are you getting out of it?
A nontechnical friend or family member (Score:2)
For non-technical people, it's still almost impossible to install Linux easily.
Attach Ubuntu or preferably Mint install media
One obstacle for a nontechnical friend or family member is that making boot media isn't as easy as dragging and dropping files to a USB flash drive. You need to use a specific tool like Rufus to make the drive bootable. "What's this GPT thing? A chatbot?"
reboot to it
This involves looking up how to change boot order on a particular mainboard and how to enable a Secure Boot certificate for non-Windows operating systems.
What I don't get is why a non-technical person like you is here on Slashdot. What are you getting out of it?
My guess is someone might be seeking the right words for a technical user to help a nontechnical friend
Re:A nontechnical friend or family member (Score:4, Insightful)
One obstacle for a nontechnical friend or family member is that making boot media isn't as easy as dragging and dropping files to a USB flash drive. You need to use a specific tool like Rufus to make the drive bootable.
Oh yes, just look at how simple it is to make windows install media [microsoft.com], just run diskpart so you can set the partition active so you can boot from it! Here's a nickel, buy a real argument.
Re: (Score:2)
You've conveniently skipped the points made about boot priority and secure boot.
Re: A nontechnical friend or family member (Score:2)
You cannot boot Windows install media without changing boot order or using the boot menu either, and Windows is much more likely to need a reinstall because Windows is trash.
Re: (Score:2)
The vast majority of people buy computers with Windows preloaded.
Motherboards all come preloaded with secure boot signing keys for Microsoft.
Re: (Score:2)
Attach Ubuntu or preferably Mint install media, reboot to it, and hit next a bunch of times. OOH HOW HARD.
I did that a few weeks ago. Attach media, reboot, click a few times, left with a purple screen and a black mouse cursor and nothing else.
Re: (Score:2)
Ubuntu has an excellent installer as do most of its offspring.
I'd agree many GNU/Linux distributions have terrible ones (Debian needs to backport Ubuntu's....) but it's the case that good installers do exist today, no worse than installing Mac OS on a Mac with a fresh hard disk despite the comparative ease Apple's engineers have given the limited hardware range implied by that.
Re: (Score:2)
Ubuntu installer didn't work without using Safe Mode for me.
Misleading headline - there is no big perf leap (Score:3)
Steam's Proton can already use esync and fsync to improve the performance of Windows games. What ntsync gives you is a more accurate implementation that works in corner cases where esync and fsync don't (for some games, you actually have to disable esync/fsync to get them to work) - the perf difference between ntsync and esync/fsync is minimal. What the misleading headline is basing the "perf leap" on is a game using ntsync vs. a game using nothing, which isn't the norm for the vast majority of games run via Proton.
Re: (Score:3)
NTSYNC is not new - it's been available for over a year now.
The problem with fsync lead to esync (enhanced fsync), but it still had problems. Basically Windows had kernel level synchronization primitives that are at best simulated by Linux in userspace. The key problem is that it's kernel level and global, so both are implemented using daemons to maintain the global list, and that any locking request now has to do multiple round trips through userspace-kernel-userspace and back again. This is quite heavy fo
ntsync (Score:2)
Last time I read about ntsync, it was basically only useful in a handful of badly-programmed cases and no different to a number of alternatives that are already used for such.
The only thing it did was being a little "closer" to Windows-style sync primitives, not actually that much better. Sure, one or two games got a huge performance boost (for the sync function) but otherwise it didn't really make that much difference.
Is that still all hyperbole or does it actually do something different now?
Would NTSYNC affect Raspberry Pi or Steamdeck? (Score:2)
This might be a dumb question, but my limited experience with Linux gaming includes my Steamdeck and trying to use Steam Remote Play on an RPi 4 running Raspbian.
Assuming (I know I know) that either or both of these platforms move to the new kernal would they likely benefit? Trying to stream games to the RPi so far has been worthless, and the Deck could always use more performance improvements.