AMD P-State Preferred Core handling for modern Ryzen systems. This is for leveraging ACPI CPPC data between CPU cores for improving task placement on AMD Ryzen systems for cores that can achieve higher frequencies and also helping in hybrid selection between say Zen 4 and Zen 4C cores. This AMD Preferred Core support has been in development since last year.
Performance gains on AMD 4th Gen EPYC
AMD FRU Memory Poison Manager merged along with other work as part of better supporting the AMD MI300 series.
AMD has continued upstreaming more RDNA3+ refresh and RDNA4 graphics hardware support into the AMDGPU driver.
#Intel
Intel Xeon Max gains in some AI workloads
Intel FRED was merged for Flexible Return and Event Delivery with future Intel CPUs to overhaul CPU ring transitions.
Reworked x86 topology code for better handling Intel Core hybrid CPUs.
Intel Fastboot support is now enabled across all supported graphics generations.
Intel Core Ultra “Meteor Lake” tuning that can yield nice performance improvements for those using new Intel laptops.
Continued work on the experimental Intel Xe DRM kernel graphics driver that Intel is aiming to get ready in time for Xe2 / Lunar Lake.
Video, Filesystem & Network
Support for larger frame-buffer console fonts with modern 4K+ displays.
Dropping the old NTFS driver.
Improved case-insensitive file/folder handling.
Performance optimizations for Btrfs.
More efficient discard and improved journal pipelining for Bcachefs.
FUSE passthrough mode finally made it to the mainline kernel.
More online repair improvements for XFS.
Much faster exFAT performance when engaging the “dirsync” mount option.
It refers to modern Intel CPUs where there are two types of cores - performance cores (P-cores) and efficient cores (E-cores). This is similar to ARM’s big.LITTLE architecture which we’ve seen in smartphones for many years already.
Yeah but this is Intel, they’re not capable of doing something the most efficient way possible. E cores use less power than P cores, but that doesn’t mean they’re very good at getting the job done using the least amount of power. Currently (using Intel’s management) 12th and 13th gen start a task on the P cores, and if it runs for longer than X time it gets shifted over to the E cores where it can churn away. Meteor lake has 3 stages of things since there’s P cores, E cores, and LP E cores. If I remember right Meteor lake starts a task on the LP E cores, then shifts it to P cores, then shifts it to the E cores if it’s taking too long. But Intel likes to blast the power away with turbo boost and runs the E cores way past their actual efficient zone unless you wrangle them back down. Sometimes it’s faster to blast the task away on the P cores then return to idle, other times letting it churn away on the E cores forever is the best way.
Also all those extra cores being active still uses power. So if you went from a 4 core CPU to a 4 + 4 cpu now you have all the same power draw as the old one, plus the extra 4 efficient cores sipping at even more power. I think that’s where 12th gen really suffers the most.
They’re not really made for power efficiency, but rather space efficiency. ~4 E-cores fit into the size of a P-core.
They’re there to boost multi-core performance without having a huge die-size or increasing latency in the P-cores when doing lightly threaded tasks, essentially.
Here’s the TL;DR from Phoronix:
#AMD
AMD P-State Preferred Core handling for modern Ryzen systems. This is for leveraging ACPI CPPC data between CPU cores for improving task placement on AMD Ryzen systems for cores that can achieve higher frequencies and also helping in hybrid selection between say Zen 4 and Zen 4C cores. This AMD Preferred Core support has been in development since last year.
Performance gains on AMD 4th Gen EPYC
AMD FRU Memory Poison Manager merged along with other work as part of better supporting the AMD MI300 series.
AMD has continued upstreaming more RDNA3+ refresh and RDNA4 graphics hardware support into the AMDGPU driver.
#Intel
Intel Xeon Max gains in some AI workloads
Intel FRED was merged for Flexible Return and Event Delivery with future Intel CPUs to overhaul CPU ring transitions.
Reworked x86 topology code for better handling Intel Core hybrid CPUs.
Intel Fastboot support is now enabled across all supported graphics generations.
Intel Core Ultra “Meteor Lake” tuning that can yield nice performance improvements for those using new Intel laptops.
Continued work on the experimental Intel Xe DRM kernel graphics driver that Intel is aiming to get ready in time for Xe2 / Lunar Lake.
Video, Filesystem & Network
Support for larger frame-buffer console fonts with modern 4K+ displays.
Dropping the old NTFS driver.
Improved case-insensitive file/folder handling.
Performance optimizations for Btrfs.
More efficient discard and improved journal pipelining for Bcachefs.
FUSE passthrough mode finally made it to the mainline kernel.
More online repair improvements for XFS.
Much faster exFAT performance when engaging the “dirsync” mount option.
Many networking improvements.
Full summary here: https://www.phoronix.com/review/linux-69-features/
I wonder what an Intel Core hybrid CPU is
It refers to modern Intel CPUs where there are two types of cores - performance cores (P-cores) and efficient cores (E-cores). This is similar to ARM’s big.LITTLE architecture which we’ve seen in smartphones for many years already.
See: https://www.intel.com/content/www/us/en/gaming/resources/how-hybrid-design-works.html
12th gen and up intel CPUs. All but a few have P and E cores.
Also like one weird i5 from about 10th gen.
I have a mobile 11th i7 one, sounds like no benefits 😞
Same boat. Good news is 11th gen generally gets better battery life than 12th+ because all those extra cores still eat power.
Bad news is I already get as bad as 30 minutes of battery life so IDK how 12th gen can be even worse.
My battery life is pretty good, the power profiles daemon is actually working very well.
I thought you saved power because the e-cores are more efficient for the same workload?
Yeah but this is Intel, they’re not capable of doing something the most efficient way possible. E cores use less power than P cores, but that doesn’t mean they’re very good at getting the job done using the least amount of power. Currently (using Intel’s management) 12th and 13th gen start a task on the P cores, and if it runs for longer than X time it gets shifted over to the E cores where it can churn away. Meteor lake has 3 stages of things since there’s P cores, E cores, and LP E cores. If I remember right Meteor lake starts a task on the LP E cores, then shifts it to P cores, then shifts it to the E cores if it’s taking too long. But Intel likes to blast the power away with turbo boost and runs the E cores way past their actual efficient zone unless you wrangle them back down. Sometimes it’s faster to blast the task away on the P cores then return to idle, other times letting it churn away on the E cores forever is the best way.
Also all those extra cores being active still uses power. So if you went from a 4 core CPU to a 4 + 4 cpu now you have all the same power draw as the old one, plus the extra 4 efficient cores sipping at even more power. I think that’s where 12th gen really suffers the most.
Spoilers: they are efficient in space, not really power
So for multicore workloads they are actually power efficient because you get more throughput (more performance iso power)
But for idle that has no benefit
They’re not really made for power efficiency, but rather space efficiency. ~4 E-cores fit into the size of a P-core.
They’re there to boost multi-core performance without having a huge die-size or increasing latency in the P-cores when doing lightly threaded tasks, essentially.