GUIDE: AMD AM5 System Optimization

[R3AP3RK1N6]

Hello dear members of the chronic tenosynovitis group and those whose membership is still pending!

I've been hanging around PC Games Hardware Extreme, but due to recent events, I want to offer you a little treat just before the end of 2024.

____________________________________________
"Please glide elegantly into the new year!"
- Wishing you a smooth transition!
____________________________________________

Over the past 21 days, on more than 20 pages, I've put together a guide to give everyone the opportunity to perform a procedure that makes the CPU run cooler, consume less power, yet generate more performance and maintain higher clock speeds.

_________
Which AMD CPUs does this guide apply to, you ask?
_________

Well, this procedure applies to every processor in the AM5 platform. It doesn't matter which chipset you are using.

____________________________________________________________________________________________
How does it go again? ^^ "If something sounds too good to be true, it usually is a lie!"


Well, in this case, and one reason why I put so much work into the guide, I've included cross-references. These lead to a small database containing screenshots from which all the guide's data can be derived. I've created bar charts that visually reflect the performance advantage directly, along with sources to explain terms and provide hyperlinks to all the programs used.

This way, anyone interested in the guide can verify the information if they wish. I did this in the interest of transparency.

Well ... I'm still undecided whether it was due to Office bugs or if I've just gotten out of practice over the years; anyway, I encountered some unpleasant complications during the creation process, and then there were test results with my OC profile that forced me into error analysis because the results simply "didn't make any damn sense!"

For this reason, at a certain point, the data resulting from the BIOS profile "Overclocked" is excluded from my guide. Currently, I have no more motivation to delve deeper into this topic.
Nevertheless, within the database, there are still screenshots of benchmark results.

I feel really burnt out after all this work...

But it's okay!
The core focus is on optimizing the platform, not overclocking.
That's a whole different topic!

If the feedback is positive, I might ... maybe ... revisit this topic.

Attached to this topic, you will find the guide in "German" and "English" - yes, I even went through the trouble of translating everything.^^

With kind regards, until further notice
Stefan André aka R3AP3RK1N6

 

[dcf-joe]

Thank you for posting this. I have been using Intel for many years, and am going to make the switch to the 9800x3d here very soon, so I am thankful for this learning resource.

 

[R3AP3RK1N6]

Thank you for posting this. I have been using Intel for many years, and am going to make the switch to the 9800x3d here very soon, so I am thankful for this learning resource.

Hello, nice to read that.

Currently, it's under further investigation, but I try to increase the performance per core.

Serious Enthusiast (and Overclocker by their job) do advice to do some deeper stability-tests:
Y-Cruncher = 12-48hrs
OCCT = 12-48hrs
CoreCycler = 24-48hrs /per Core
Prime95 = 24-48hrs
MemTest86 = 24hrs
Aida64 = 24-48hrs
Cinebench R23 as well as 2024 is not that good for long Stress-Test-Loops, but will assist to confirm a good system stability.

On every change of BIOS-Settings, you should re-start tests.
Whithin the first 1-2hrs you will reach errors and you will have to do adjustments. Just make one step after another, to be able to understand what causes the problem, to keep troubeshooting more easy.

I'm diving deeper into performance uplifts and as a base I'm using the method discribed in my work.
3 Cores were struggeling and needed further adjustments.
So - it's not a bad way, but for such specific workloads by these test-methods, it's not 100% stable.

Do not be afraid about that tests, it might look extreme, but you will reach longer test-runs the more stable your system is getting.

The next AM5-Docu that I will release is more detailed - more technical - more specific.

Enthusiasts and Overclockers call it "He started to learn walking and wrote down his first steps..." but 3 of 16 physical cores unstable on a system which had been running for a year on my personal workloads, is "okay" for "muggles like me", but way way way too less for serious professionals.

So - next work is focused onto professional-insights of this platform, to even make the professionals happy.

 

[A Computer Guy]

If I remember correctly one part of testing that is needed when using Curve Optimizer is testing light workloads to ensure cores don't crash when boosting. I did a quick read though and didn't happen to see anything like that mentioned in the PDF. Anyway thanks for posting this looking forward to reading more in depth later.

 

[phanbuey]

This is probably one of the more impactful options, it disables VBS as a direct effect. Wont work for people who need to run VMs though.

 

[Zack25]

Thanks for the guide.
Btw do you have a guide or help-thread for noobs around mainly focus on stability and good temps?

Also can you advice me with this one:

Spoiler: HELP me with this 7950X and RAM dilemma

I've bought a "cheap" combo on discount:

• 7950x + Asus TUF x670E-plus wifi (i know not the best x670 mobo around)

Sadly, where i live dont ve many DDR5 2x24GB kits around to pick, being these ones only available:

• Patriot Viper Elite 48GB 6000MT/CL42
• Adata XPG Lancer ROG Intel Certified 48GB 7200MT/CL34 (Patriot page saying this kit profile only compatible with Intel Z790, but its Datasheet also says that might work on "latest AMD systems")

Also found out from reviews and comments that this CPU is quite sensitive to CAS latency, so the Adata XPG kit feels best option to pick but isnt listed on any AM5 board QVL due its advertised "Intel XMP profile". Do you think will work out or even POST to bios on this mobo/cpu? my plan was underclock it even to 6000mts and start from there if POSTs.

Thanks in advance for any help!

 

[R3AP3RK1N6]

View attachment 378070

This is probably one of the more impactful options, it disables VBS as a direct effect. Wont work for people who need to run VMs though.

Hey, greetings!

Not "only" if you want to run Virtual-Maschines on your system, it is also disabling some security options of the Windows Defender.
"Core-Isolation" is working in a "virtual surrounding", so if SVM-Mode is disabled, it will impact the Defender as well, but it will save some ressources as well as decreasing power-consumption.

 

[R3AP3RK1N6]

Thanks for the guide.
Btw do you have a guide or help-thread for noobs around mainly focus on stability and good temps?

Also can you advice me with this one:

Spoiler: HELP me with this 7950X and RAM dilemma

I've bought a "cheap" combo on discount:

• 7950x + Asus TUF x670E-plus wifi (i know not the best x670 mobo around)

Sadly, where i live dont ve many DDR5 2x24GB kits around to pick, being these ones only available:

• Patriot Viper Elite 48GB 6000MT/CL42
• Adata XPG Lancer ROG Intel Certified 48GB 7200MT/CL34 (Patriot page saying this kit profile only compatible with Intel Z790, but its Datasheet also says that might work on "latest AMD systems")

Also found out from reviews and comments that this CPU is quite sensitive to CAS latency, so the Adata XPG kit feels best option to pick but isnt listed on any AM5 board QVL due its advertised "Intel XMP profile". Do you think will work out or even POST to bios on this mobo/cpu? my plan was underclock it even to 6000mts and start from there if POSTs.

Thanks in advance for any help!

Hey there!^^
For some boards and manufacturers the maximum supported RAM-Speed had climbed up to 6.800MT/s.
Your KIT isn't listed, because of 7.200MT/s - is the ITEM No. this: "AX5U7200C3424G-DCLARROG"?
Even if "BIOS-Updates" should increase compatibility - in the future - it's mostly because of pre-defined values for different RAM-Kits, who are tested to "should" improve compatibility and stability, but at the end, the user have to test the system by himself.

You know, the day that I bought my RAM-Kit it wasn't supported as well, but I gave it a try.^^
It was a risk that I was willing to take.
I knew that AMD has D.O.C.P. compatibility-mode for XMP-RAM and I knew that only Golden-Samples would support 6.400MT/s out of the box.
- I know, that there are communities on Discord, who are full of professionals who tweaked the RAM-CL-Timings onto their limits, but I'm not any of those professionals.

I saw a how-2 manual which was all about RAM-Tweaking and it was created by 10 users together.
Real Hardware-Freaks you could say and well ... in comparison to that, I would rate my manual as garbage as well.^^

It could work on your system, but to make sure it will work, you should search for one of the communities who work together, to get best compatibility and performance out of DDR5-RAM-Kits. I have no idea or advice for you rather than that, I'm so sorry.
I wouldn't ask ASUS for it anyway ... they don't know it, because the question that you have is for advanced users and experts, something the official support isn't able to solve.

I hope I was able to help a ... bit.

 

[Shou Miko]

Hey there!^^
For some boards and manufacturers the maximum supported RAM-Speed had climbed up to 6.800MT/s.
Your KIT isn't listed, because of 7.200MT/s - is the ITEM No. this: "AX5U7200C3424G-DCLARROG"?
Even if "BIOS-Updates" should increase compatibility - in the future - it's mostly because of pre-defined values for different RAM-Kits, who are tested to "should" improve compatibility and stability, but at the end, the user have to test the system by himself.

You know, the day that I bought my RAM-Kit it wasn't supported as well, but I gave it a try.^^
It was a risk that I was willing to take.
I knew that AMD has D.O.C.P. compatibility-mode for XMP-RAM and I knew that only Golden-Samples would support 6.400MT/s out of the box.
- I know, that there are communities on Discord, who are full of professionals who tweaked the RAM-CL-Timings onto their limits, but I'm not any of those professionals.

I saw a how-2 manual which was all about RAM-Tweaking and it was created by 10 users together.
Real Hardware-Freaks you could say and well ... in comparison to that, I would rate my manual as garbage as well.^^

It could work on your system, but to make sure it will work, you should search for one of the communities who work together, to get best compatibility and performance out of DDR5-RAM-Kits. I have no idea or advice for you rather than that, I'm so sorry.
I wouldn't ask ASUS for it anyway ... they don't know it, because the question that you have is for advanced users and experts, something the official support isn't able to solve.

I hope I was able to help a ... bit.

Ain't it still important to run 1:1 with the AMD Infinity fab on the Ryzen cpu'en so you don't get a penalty?

I know 6000MHz is the sweet spot if you want to be adventure you can try 6200MHz a lot of cpu's do this still running 1:1.

 

[R3AP3RK1N6]

Ain't it still important to run 1:1 with the AMD Infinity fab on the Ryzen cpu'en so you don't get a penalty?

I know 6000MHz is the sweet spot if you want to be adventure you can try 6200MHz a lot of cpu's do this still running 1:1.

Hello there,

You mean MEM-CLK=UCLK ration right?
For me personal it feels more better to have a ration of 1:1 too, but I wasn't able to find any penalties when I use 1:2 ration.
I don't know what use-cases might trigger diffrences.

It seems like stressing the memory-controller more to use a ration of 1:1 and it increases power-consumption as well as produced heat.
These values do produce electro magnetic fields too, so it's also possible to create interferences.

Everything on the AM5 CPU responses sensitive on interferences produced by next-by cores. The density is very high on these CPUs, so the memory-controller is nothing excluded of it or seperated or even isolated from the rest. I wish it would be some sort of isolation against magnetic fields existant, but it's not.

I have only the 7950X3D, so I'm not able to share experiences with the newest series, but I'm looking forward for the 9950X3D.
I hope it's overclocking potential is combined with less sensivity on interferences.

I'm looking forward for the next-gen DDR5 modules, because of an additional controller on the modules, which will help on data-transfers - it will increase compatibility and support of higher clocks, but for sure it will increase latency.

There're some overclockers who do adjust the impendance of the RAM-to-Memory-Controller pipeline/interface, to be able to run higher clocks.
It depends ... my RAM has Hynix-Modules, so a lot of tweaking is depending on what kind of modules your RAM has.

In every situation when there's a new bios released and there's something reported like " increased memory frequencies and compatibility" I do ask myself: "If the RAM is directly coupled to the CPU and the controller is on the CPU, how should a bios-update solve compatibility-problems, if there's no device between the RAM and the CPU?! Shouldn't the CPU-Drivers solve some incompatibility??" - I do think that there might adjustments below the BIOS-UI that we can't see, who might improve stability and compatibility, but tests resulting mostly in system-crashes when I try to run highly tweaked and/or RAM-OC-Settings, so I stoped trying it.

I don't have enough time to run enough tests, to make sure the system runs clean, until there's a new bios-update released.
It's the same problem with PBO-Settings ... it runs with the last bios-version, doesn't mean, that it will run with the next version.
It's always resulting into restarts of the work and that's exhausting when you do have a job and a reallife beside of the wish of using the performance for its best use-case "gaming".^^

Peace.

 

[Halo3Addict]

Great write-up, and very interesting to read! I'm still on AM4 but hoping to upgrade in the the near future