[SOLVED] Ryzen 7 3800x and Gigabyte B450 AORUS ELITE V2, working 24/7 on Linux

[keramxd]

Hello,
My PC is using to compressing files 24/7. I would like to do overclocking of 3800x. I'm using Ubuntu and have good 2 fans CPU cooler.
What settings do you recommend me to try considering that PC works all the time?

 

[sonofjesse]

that 3800x should be doing pretty good as is

If you want to overclock be sure your PC has lots of airflow (not just on the CPU)

Try small increments, most of these chips are pretty "hott" from the factory and don't have crazy headroom.

You could always sell the 3800x and get a 3900x used........if your dying to compress files faster........

Just go on youtube and watch a bunch of videos, you will see where to start

 

[drea.drechsler]

Hello,
My PC is using to compressing files 24/7. I would like to do overclocking of 3800x. I'm using Ubuntu and have good 2 fans CPU cooler.
What settings do you recommend me to try considering that PC works all the time?

Just enable PBO in BIOS and undervolt. Some simple PBO settings that may help: set PPT to 333W and both EDC and TDC to 230A.

Undervolt very slightly using negative offset only. I'd start at -0.0125V, or one notch. That lets the CPU run a bit cooler which helps it hold a higher clock for long periods. If you go too far it will lose performance or, worse, go unstable. Undervolt settings vary greatly depending on motherboard as well as CPU so some experimenting is needed to find the optimum UV.

Also set the following to ENABLED: AMD Cool n Quiet, Advanced C States, Processor CPPC and CPPC Preferred Cores.

PBO makes it run hotter but the boost algorithm will protect the processor by pulling back clocks when it gets too hot. So probably most important thing to maintain high boost clocks with PBO enabled is great cooling. Since it's running 24/7 you might even open the case and locate a room fan to blow cold air at the CPU cooler intake.

You want temp to stay under 80C if possible to get best boost performance; but temps into the mid-upper 80C range is OK for long term. Do not let it go above 90C long term. 95C is Tjmax.

And do not forget memory so make sure you've tweaked up memory clocks too. Once finished, especially with memory tweaking, be sure to run a stability test to know it's not turning out corrupt compressed files.

 

[keramxd]

Just enable PBO in BIOS and undervolt. Some simple PBO settings that may help: set PPT to 333W and both EDC and TDC to 230A.

Undervolt very slightly using negative offset only. I'd start at -0.0125V, or one notch. That helps the CPU stay a bit cooler which help it hold a higher clock for long periods. If you go too far it will lose performance or, worse, go unstable. Undervolt settings vary greatly depending on motherboard as well as CPU so some experimenting is needed.

Also set the following to ENABLED: AMD Cool n Quiet, Advanced C States, Processor CPPC and CPPC Preferred Cores.

PBO makes it run hotter but the boost algorithm will protect the processor by pulling back clocks when it gets too hot. So probably most important thing to maintain high boost clocks with PBO enabled is great cooling. Since it's running 24/7 you might even open the case and locate a room fan to blow cold air at the CPU cooler intake.

And do not forget memory so make sure you've tweaked up memory clocks too. Once finished, especially with memory tweaking, be sure to run a stability test to know it's not turning out corrupt compressed files.

Thank you so much!
Should I set up constant ghz for exmaple 4.2ghz or don't touch it?
Any advices regarding RAM overclocking? 😀

 

[drea.drechsler]

Thank you so much!
Should I set up constant ghz for exmaple 4.2ghz or don't touch it?
Any advices regarding RAM overclocking? 😀

I would not use a fixed, manual, overclock on any Ryzen 2nd or 3rd gen CPU. Some people do and it can be effective but it takes a lot of tweaking to find a safe voltage and frequency that won't result in CPU degradation. That's especially important for a system that's running 24/7 in heavy all-core tasking. PBO leaves the boost algorithm running and that protects the processor. And besides, a system set up right with PBO will probably have the same performance that fixed manual overclock can achieve in all-core and much better in light threaded workloads.

If you're willing to push the CPU harder with PBO then move the PBO Scalar setting up to 5X. That basically relaxes the criteria the algorithm uses to protect the CPU. I wouldn't go above 5X, though, unless temp's stay in the low- mid 80's.

For RAM overclocking beyond what your XMP settings might get look for a utility called Ryzen DRAM Calculator. It helps calculate optimum timings for higher RAM clocks to keep it stable, but it's a bit complicated to use it correctly. I'd not worry, though, about getting clocks above 3600 since you usually have to de-sync the IF to go higher.

 

[keramxd]

I would not use a fixed, manual, overclock on any Ryzen 2nd or 3rd gen CPU. Some people do and it can be effective but it takes a lot of tweaking to find a safe voltage and frequency that won't result in CPU degradation. That's especially important for a system that's running 24/7 in heavy all-core tasking. PBO leaves the boost algorithm running and that protects the processor. And besides, a system set up right with PBO will probably have the same performance that fixed manual overclock can achieve in all-core and much better in light threaded workloads.

If you're willing to push the CPU harder with PBO then move the PBO Scalar setting up to 5X. That basically relaxes the criteria the algorithm uses to protect the CPU. I wouldn't go above 5X, though, unless temp's stay in the low- mid 80's.

For RAM overclocking beyond what your XMP settings might get look for a utility called Ryzen DRAM Calculator. It helps calculate optimum timings for higher RAM clocks to keep it stable, but it's a bit complicated to use it correctly. I'd not worry, though, about getting clocks above 3600 since you usually have to de-sync the IF to go higher.

Hello, I did everything and see a little improvement. I used PBO and set PPT to 333W and both EDC and TDC to 230A. I did not change voltage and have set up Scalar 5X

After 12 hours of working on all cores it shows:
CPU Usage - 35-60%
Temp - 65 °

cpu MHz         : 2537.951
cpu MHz         : 3900.000
cpu MHz         : 4437.716
cpu MHz         : 4435.262
cpu MHz         : 3900.000
cpu MHz         : 2800.000
cpu MHz         : 3900.000
cpu MHz         : 3900.000
cpu MHz         : 2800.000
cpu MHz         : 4437.235
cpu MHz         : 3900.000
cpu MHz         : 2200.000
cpu MHz         : 2200.000
cpu MHz         : 2800.000
cpu MHz         : 2800.000
cpu MHz         : 3900.000

Can I still do something to compressing faster?

 

[drea.drechsler]

Hello, I did everything and see a little improvement. I used PBO and set PPT to 333W and both EDC and TDC to 230A. I did not change voltage and have set up Scalar 5X

After 12 hours of working on all cores it shows:
CPU Usage - 35-60%
Temp - 65 °

cpu MHz         : 2537.951
cpu MHz         : 3900.000
...

Can I still do something to compressing faster?

I'm not sure what you're using that gives you that report but it's apparently average clock speed by thread. That's a bad way to assess performance of a modern processor, especially a multi-threaded processor. They're too dynamic with rapid raising and lowering of core clocks as the code allows in an effort to stay cool and clock speed of a virtual thread is really meaningless since there's only one core for both of it's threads.

It might also be a better measure of the compression code efficiency than processor performance as it suggests to me it's only using a few cores heavily (3 cores at over 4.4Ghz, which is pretty good for sustained performance) and the rest much more lightly.

You have to use a benchmark and measure performance directly. I'd run a standard compression job and measure the elapsed time to complete it. Just make sure it runs fairly long, about 10 min's should be enough. I use a Handbrake video to assess my processor's performance.

Probably the only way you'll get to compressing faster is do as @sonofjesse suggests and step up to 3900 or 3950 (since you can more easily get one of those than 5900/5950X's) which puts more cores on the job. But if you do, keep an eye on the motherboard VRM temperature. And even then the code may not scale well to more cores, as that report suggests.

 

[keramxd]

I'm not sure what you're using that gives you that report but it's apparently average clock speed by thread. That's a bad way to assess performance of a modern processor, especially a multi-threaded processor. They're too dynamic with rapid raising and lowering of core clocks as the code allows in an effort to stay cool and clock speed of a virtual thread is really meaningless since there's only one core for both of it's threads. It might be a better measure of the compression efficiency than processor performance.

You have to use a benchmark and measure performance directly. I'd run a standard compression job and measure the elapsed time to complete it. Just make sure it runs fairly long, about 10 min's should be enough. I use a Handbrake video to assess my processor's performance.

Probably the only way you'll get to compressing faster is do as @sonofjesse suggests and step up to 3900 or 3950 (since you can more easily get one of those than 5900/5950X's) which puts more cores on the job. But if you do, keep an eye on the motherboard VRM temperature.

I've build my machine based on this: https://docs.google.com/document/d/18a5_MO88hv_DL_644OGFoCiKnpR9udl-PAWKlyIqqQU/edit#

That guy uses: Intel Core i7-10700 and generate about 2.76 TiB/day when I on the same parts with only different CPU and mobo get 2.20 TiB/day. Is this possible that 10700 gives 20% better result than 3800x?

 

[drea.drechsler]

I've build my machine based on this: https://docs.google.com/document/d/18a5_MO88hv_DL_644OGFoCiKnpR9udl-PAWKlyIqqQU/edit#

That guy uses: Intel Core i7-10700 and generate about 2.76 TiB/day when I on the same parts with only different CPU and mobo get 2.20 TiB/day. Is this possible that 10700 gives 20% better result than 3800x?

I'd think yes if the code is optimized for Intel...and especially yes if it's optimized for Intel to the exclusion of AMD. And looking at those thread clock results it may be optimized for light threaded operation and not scaling well to 8 cores/16 threads. It looks to be optimized for a 4 core/4 thread CPU, so favors Intel's IPC advantage.

 

[keramxd]

I'd think yes if the code is optimized for Intel...and especially yes if it's optimized for Intel to the exclusion of AMD. And looking at those thread clock results it may be optimized for light threaded operation and not scaling well to 8 cores/16 threads which looks to be the case. It looks to be optimized for a 4 core/4 thread CPU.

Is there something else that I can do to improve performance besides PBO to get closer to 10700 results ? 😀

 

[drea.drechsler]

Is there something else that I can do to improve performance besides PBO to get closer to 10700 results ? 😀

Get a 5800X...maybe even a 5600X if I'm right about it being optimized for only 4 cores.

You're getting over 4.4Ghz sustained for 3 cores, and 3.9Ghz for a few others. Let's say you could get a fixed overclock stable at about 4.3Ghz all cores, it still won't help those threads that only ran at 2.2-2.8Ghz in your report (not even the 3.9Ghz, most likely) because they weren't doing anything; that's why they are running that slow. It will only slow down those cores that are running all-out at 4.4Ghz+ and doing all the work.

Another way is find who coded your compression software and see if there's a way to get it to run right on an 8 core/16 thread AMD CPU.

 

[Karadjgne]

Guru3d.com and get ClockTuner2 and Dram Calculator. Also dl Typhoon burner.

First use TB to read the specs on the ram. After opening DC, there's an easy export that'll plug all the values in for you. That'll give the best settings for the ram speeds. There's even a video tutorial on certain setups, advised to watch it. Once manually plugged into bios, I'd advise saving as a User profile so if you ever need to reset bios, just load the profile to return ram to tailored settings.

ClockTuner2 is all about lowering voltages to maximize boosts. Ryzens boost according to voltage, temp and loads. If you max any of those 3, the Ryzen will start cutting back on MHz on a core by core basis. So if you can minimize those, you get higher stable boosts. It does have some default settings, but those are adjustable, you can set VID and Hz and let the program run, it'll test stability per CCX to get the best your cpu can do.

Even has 3 different settings based on workload that can automatically run, mine is set stock for light loads, high for medium loads and slightly lower for full core use. So I get 4.2-4.4GHz across any used core, and temps don't exceed 62°C @ 1.215v vs stock 1.475v all core and 85°C at 3.9GHz..

Written by 1usmus (Linus, Igor and multiple other Ryzen pro overclockers)

Ryzens perform best when left to perform as a Ryzen. A standard Intel 'push it hard at this setting' OC by maximizing clocks then dropping voltages will work just fine, as long as you are prepared for the constant high heat as a result.

Cinebench R20 got my 3700x at stock @ 3723. A full 4.4GHz OC got 5101 @ 85°C. CT2+DC got 5000 @ 62°C. To me, the 1300 point bonus performance from CT2+DC far outweighs the extra 100 points and 20°C gain from the OC.

Oh, and because it's a Ryzen, the manual OC actually had 30 point lower single core score vrs the CT2+DC score, which can be important in gaming as single core performance is often the basis for fps.

 

[popatim]

it doesn't look like your compression utility is set to only use 3 core, or maybe that's all it can use. I would expect all the cores to be hit hard.
Which one are you using?

 

[drea.drechsler]

Is there something else that I can do to improve performance besides PBO to get closer to 10700 results ? 😀

One other thing I forgot about....

You could try per-CCX overclocking. A 3800X is built on one CCD chiplet with two four-core CCX's. Many motherboard BIOS' allow to overclock individual CCX (or CCD's for 3900X and 3950X's). So the thing to do is determine which of your 8 cores are the strongest and in which CCX most of them are located. Then overclock that CCX alone, you'll have to get it stable above 4.25Ghz (since it's already doing that); I'd try for 4.5Ghz at least.

The idea is your compression utility isn't doing significant processing work on the other threads, probably just housekeeping or something similar, which isn't slowing down the primary threads. So clock the other CCX at something quite low, maybe 3.9Ghz or even 3.4Ghz. That will keep the entire CPU a bit cooler as only the 4 cores running the primary compression threads are going to be working hard, at a high clock. The hope is that using only the strongest cores you may achieve a 4.5-4.6Ghz overclock in a reasonable voltage and remain cool and stable for 24/7 compression.

You'll have to make sure the compression threads run on the high performing cores...and the "housekeeping" threads on the low-performing ones... of course. I'm not familiar at all with Linux but I do believe it has thread affinity tools that should allow you to do that. Or maybe the Linux scheduler is smart enough to do that on it's own; I'd bet Windows' is not!