• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Linux Kernel 6.13 brings an option for AMD 3D Cache Processors

Joined
Jul 31, 2024
Messages
599 (3.42/day)
Note: This is not a support request topic. This is just an opinion and just for information.

This really shows how far behind we are in regards of optimisations. I see a lot of changes for the scheduler and such.



source: linux kernel / or:

CONFIG_AMD_3D_VCACHE: AMD 3D V-Cache Performance Optimizer Driver​

General informations​

The Linux kernel configuration item CONFIG_AMD_3D_VCACHE:

Help text​

The driver provides a sysfs interface, enabling the setting of a bias that alters CPU core reordering. This bias prefers cores with higher frequencies or larger L3 caches on processors supporting AMD 3D V-Cache technology.

If you choose to compile this driver as a module the module will be called amd_3d_vcache.

Hardware​

LKDDb​

Raw data from LKDDb:

Sources​

This page is automaticly generated with free (libre, open) software lkddb(see lkddb-sources).

The data is retrived from:

 
Joined
Jan 3, 2021
Messages
3,726 (2.52/day)
Location
Slovenia
Processor i5-6600K
Motherboard Asus Z170A
Cooling some cheap Cooler Master Hyper 103 or similar
Memory 16GB DDR4-2400
Video Card(s) IGP
Storage Samsung 850 EVO 250GB
Display(s) 2x Oldell 24" 1920x1200
Case Bitfenix Nova white windowless non-mesh
Audio Device(s) E-mu 1212m PCI
Power Supply Seasonic G-360
Mouse Logitech Marble trackball, never had a mouse
Keyboard Key Tronic KT2000, no Win key because 1994
Software Oldwin
By "we", you mean Linux users compared to Windows users?
 

bug

Joined
May 22, 2015
Messages
13,960 (3.95/day)
Processor Intel i5-12600k
Motherboard Asus H670 TUF
Cooling Arctic Freezer 34
Memory 2x16GB DDR4 3600 G.Skill Ripjaws V
Video Card(s) EVGA GTX 1060 SC
Storage 500GB Samsung 970 EVO, 500GB Samsung 850 EVO, 1TB Crucial MX300 and 2TB Crucial MX500
Display(s) Dell U3219Q + HP ZR24w
Case Raijintek Thetis
Audio Device(s) Audioquest Dragonfly Red :D
Power Supply Seasonic 620W M12
Mouse Logitech G502 Proteus Core
Keyboard G.Skill KM780R
Software Arch Linux + Win10
"Far behind" means nothing in the context. It's heterogeneous computing after all. Some parameters will favor some workloads, other parameters will favor others. There will never be a day when you will wake up and say "ok, we're done, now the driver does the perfect job". There is no perfect job in this context.
 
Joined
Jul 31, 2024
Messages
599 (3.42/day)
#2 We in the context of someone who has that processor, cares for optimising the operating system and uses a recent "custom" made linux kernel.

I saw a lot of optimisations in past weeks which looked like aimed for those intel P and E core processors.
There were also optimisations or changes to the scheduler a lot. In terms of security, but also a lot in terms of changed or newer kernel options.

edit: I just realised - the most affected cpus are most likely the ryzen 7950X3d or ryzen 7900X3d.
The ryzen 5800X3d, 7800X3d, 7600X3d, 5700X3d, 5600X3d should not be affected.


--

#3 there lies a lot of potential in regards of the compiler and in the form of the kernel.

e.g. less seconds to compile a code

I think it matters a lot where the code is executed on e.g. ryzen 7950X3d or ryzen 7900X3d


--

typical amd - no specs
when you look up an intel processor you see inital the launch date for any processor.

It seems the launch date was the 20th april 2022 for the ryzen 5800X3d. (~ a bit less than 3 years)

IT looks like March 2023 was the release date for the 7950X3d cpus. (~2 years)
 
Last edited:
Joined
May 10, 2023
Messages
525 (0.84/day)
Location
Brazil
Processor 5950x
Motherboard B550 ProArt
Cooling Fuma 2
Memory 4x32GB 3200MHz Corsair LPX
Video Card(s) 2x RTX 3090
Display(s) LG 42" C2 4k OLED
Power Supply XPG Core Reactor 850W
Software I use Arch btw
I consider the Ryzen 5800X3d one of the first processors where this kernel option should be applied to. That is very long on the market.
This would be irrelevant to the 5800x3D, since there's no bias to set given that all cores are uniform with the V-cache.
This is relevant to stuff like 7950x3D, where you have CCDs with/without the V-cache, and may want to manually set a bias in an easier manner.
FWIW, there were similar patches for stuff like P/E-cores, for both intel and AMD.
#3 there lies a lot of potential in regards of the compiler and in the form of the kernel.

e.g. less seconds to compile a code
Not really? This specific example you gave is a embarrassingly parallel workload, there's no need for a bias to certain core types when you're going to use all of those anyway.

I think it matters a lot where the code is executed on e.g. ryzen 7950X3d or ryzen 7900X3d
I don't think it's that relevant in this case, the previous scheduler iterations were already good enough when deciding where to allocate stuff. It already had "knowledge" of different frequencies and capabilities in heterogeneous CPU configs. Friendly reminder that the kernel is pretty used to dealing with multiple different clusters of CPUs in the ARM world already.
 
Top