3800X build bad performance - what am I doing wrong?

[TheLostSwede]

Won’t have time until evening to test. 11am here. I will dl the other benches.

this is probably also an important info: this 2070s is already the second 2070s I have in the sysrem. The first one was too loud. With both cards I got almost the same results!

I doubt you have any busted hardware.
Something to try on the SSD, is to run a TRIM command, leave it for 10 minutes and then run a benchmark.
Also try setting CrystalDiskMark to 4K Q8T8 and you can compare with the numbers in the graphs below to see if you're with the ballpark of those numbers in terms of performance. Then you at least have a solid target in terms of performance, although your random read IOPS should be higher. I also included my results with the same settings, which you should beat if the SSD is working as it should.

 

[fwiler]

I'm that guy that said don't use hwinfo. But that was to measure temps and voltage at idle.
The op had high voltage and temps at idle.
hwinfo is too aggressive at idle and shoots the cores up each time it measures. So you don't get a realistic reading.
Others like Ryzen master do not.
If you are satisfied with your core idle temps and voltage then hwinfo is fine for measuring during all other testing.

 

[birdie]

You have really low write speed.

Additionally, your large changes from run to run are likely from background processes.

Advice like this is what makes people feel stupid and mad.

Only the Ryzen 9 3900 and better CPUs have full memory write speed, everything below including the 3800X feature halved memory write speed.

Cinebench Score Updates:

• 4891pts - only windows update - default nvidia - default chipset - cpu load 0% /// no PBO - no doccp
• 4928pts - only windows update - default nvidia - default chipset - cpu load 0% /// no PBO - doccp enabled
• 4910pts - only windows update - default nvidia - default chipset - cpu load 0% /// no PBO - manual 16-16-16-36 @ 1.35v
• 4942pts - only windows update - default nvidia - chipset updated - cpu load 0% /// no PBO - manual 16-16-16-36 @ 1.35v
• 4924pts - only windows update - nvidia driver updated - cpu load 0% /// no PBO - manual 16-16-16-36 @ 1.35v
• 4904pts - only windows update - nvme Samsung driver updated & AIDA64 installed - cpu load 0% /// no PBO - manual 16-16-16-36 @ 1.35v
• 4923pts - only windows update - BT+wifi driver update+samsung magician install - cpu load 0% /// no PBO - manual 16-16-16-36 @ 1.35v

when doccp is disabled voltage is on auto, could that be a reason? Gonna run now all benches with doccp on and off

Looks like a Windows reinstall has solved all your issues. Your scores are perfectly fine now.

to sum it up - please correct any rumbling down here if it is wrong, no idea how I'm still able to type:

• booting time issue seems resolved - (gonna closely watch it during my program installs. gonna keep fast boot disabled until all is installed
• temperatures of CPU is still not great but acceptable under load
• Samsung evo has too little IOPS - so far no solution
• GPU bench not sure cause I'm tired as f
• rams still need tweaking but it looks like that the latency is where it should be
• I still don't have PBO enabled and the common understanding in this thread is that it should be kept disabled??
• chuck norris doesn't have an ESC key on his computer
• I should not try to manually OC CPU cause it won't do much
• I should stick to 3600 so far and try to tighten timings

Don't bother with your SSD IOPS - they are perfectly fine for your use.

PBO will make your system run maybe 2% faster at the cost of a hugely inceased power consumption and even higher temps. You're already not content with your temps, why would you want to OC your system?

Tightening RAM timings without then running memtest86 for at least few hours may lead to instability, crashes and sometimes even data loss.

 

[TheLostSwede]

I'm that guy that said don't use hwinfo. But that was to measure temps and voltage at idle.
The op had high voltage and temps at idle.
hwinfo is too aggressive at idle and shoots the cores up each time it measures. So you don't get a realistic reading.
Others like Ryzen master do not.
If you are satisfied with your core idle temps and voltage then hwinfo is fine for measuring during all other testing.

You can actually set how often HWInfo should poll the system, so you can reduce that by yourself. But yes, the default setting is quite aggressive.
Change that 2000ms to something higher and it's not as aggressive.

 

[moproblems99]

like this is what makes people feel stupid and mad.

Sorry, no advice was given. Only observation.

 

[TheHunter]

Sry about the bad quality but here's the task manager.
The app was "Snagit" - a screenshot app, but it only was at the beginning, now at 0% where it is supposed to be.
I need the iCUE for my RGB profils unfortunately and in it I have my commander pro controlling the case fan speeds as well.

I really don't have another SSD lying around but I could just test the drive again when I have a fully clean system?
And yes, it is on AHCI in bios. I only have RAID as another option which would be clearly wrong for one drive

View attachment 154725

Eset is shit lately.. remove that and you will have resposive system from start..

Ive used eset for years until it really started to piss me off by startup last few months, 1-3min lag by boot.. stupid firewall service is slowing it down. Although i like its firewall rules setup..

Atm at kaspersky trial and it loads fast.




About idle temp., asus c states are usually set to auto/off by default and cpu can idle at higher temp., it wont use c states at all even at balanced..


About slow logo boot, yes there is something in bios conflicting. I guess like others said ram training failing..


Also in bios there is a setting by sata called agressive power saving , turn that off.

Idk about ryzen but maybe it also has this tinylake lake setting that can control c states via intel rst.


Have you tried older icue app.. i use that for my h115i pro rgb, seems less cluttered and buggy.

Edit;
I see you fixed your ram issues with new ram, good.

And dont monitor CB with background apps like hwinfo or aida, that will have score variations .. same with icue runing in the background.

Use them only for troubleshooting.

 

[ecopsorn]

@ecopsorn
sounds good so far.

-set your min cpu perf to 20-30%. will lower volt/temps, but not really impact fps.
-haven't tested heaven past 7xx series.
run valley or better superposition (both 1080p/extreme, change to custom if the stock setting wont match that).
did u install gfe/shadowplay? (usually eats a bit fps even if not recording (but present/buffering).

you mean this? from 99 to 20?

I'm downloading the benches now and report.
I did not install shadowplay

I doubt you have any busted hardware.
Something to try on the SSD, is to run a TRIM command, leave it for 10 minutes and then run a benchmark.
Also try setting CrystalDiskMark to 4K Q8T8 and you can compare with the numbers in the graphs below to see if you're with the ballpark of those numbers in terms of performance. Then you at least have a solid target in terms of performance, although your random read IOPS should be higher. I also included my results with the same settings, which you should beat if the SSD is working as it should.

will try that asap

Looks like a Windows reinstall has solved all your issues. Your scores are perfectly fine now.
PBO will make your system run maybe 2% faster at the cost of a hugely inceased power consumption and even higher temps. You're already not content with your temps, why would you want to OC your system?
Tightening RAM timings without then running memtest86 for at least few hours may lead to instability, crashes and sometimes even data loss.

don't feel like all is resolved yet but it was definitely a step forward. With the new install I've had it 2 times now that the windows button (start menu) stopped working. I also had 2-3 restarts where the computer wouldn't restart, all lights and fans were still on though - had to manually reset.

I let PBO disabled, that's fine

Once I've tested more on the gpu now and some disk benches, I might tune some timings again and let the memtest run overnight

Eset is shit lately.. remove that and you will have resposive system from start..

hmm, also have it running on my computers since years, but probably it is time for a switch. I don't have it installed yet.

@Fry178
this is the valley run

@TheLostSwede
latest run, haven't changed anything. somehow 4k is much much higher

and here is the Q8T8 bench - aren't those numbers suddenly very good now and if they are - how the hell can it vary that much from day to day on a clean system with nothing changed except a good night sleep in between?

 

[TheLostSwede]

@TheLostSwede
latest run, haven't changed anything. somehow 4k is much much higher
View attachment 155609

and here is the Q8T8 bench - aren't those numbers suddenly very good now and if they are - how the hell can it vary that much from day to day on a clean system with nothing changed except a good night sleep in between?
View attachment 155610

Looking good now. I guess there might've been some updates or something running in the background that might have caused the hiccup.
Nothing to worry about here now.

The random restarts and lockups are not good though. Not sure what's causing that, but most likely some setting or the other.

 

[ecopsorn]

Samsung magician thingy still reports that though - but like a lot of you mentioned, I should ditch that program anyway

 

[TheLostSwede]

Samsung magician thingy still reports that though - but like a lot of you mentioned, I should ditch that program anyway
View attachment 155611

No need to dump the magician, as it's used for updating the firmware on the SSD, which I presume you have done?
As you've seen, there are a lot of variables when you test and we don't know what settings Samsung use, unless there's somewhere you can find that.
Just don't leave it running in the background, as it's obviously eating some resources, which is pointless.

 

[ecopsorn]

No need to dump the magician, as it's used for updating the firmware on the SSD, which I presume you have done?
As you've seen, there are a lot of variables when you test and we don't know what settings Samsung use, unless there's somewhere you can find that.
Just don't leave it running in the background, as it's obviously eating some resources, which is pointless.

yes, all updated. than I cross the SSD performance of my list of things to fix

Regarding ram timings - most timings I try now give me the following error sequence: F9 -> d6 -> 02 | at 02 nothing happens anymore and I have to manually power off. After that windows boots, pressing DEL won't get me into bios, in windows JEDEC is applied when I check timings, after restart I can get back into Bios.
I'm gonna run stock timings now with membench overnight.

 

[moproblems99]

yes, all updated. than I cross the SSD performance of my list of things to fix

Regarding ram timings - most timings I try now give me the following error sequence: F9 -> d6 -> 02 | at 02 nothing happens anymore and I have to manually power off. After that windows boots, pressing DEL won't get me into bios, in windows JEDEC is applied when I check timings, after restart I can get back into Bios.
I'm gonna run stock timings now with membench overnight.

Things are looking up! Great progress.

 

[tabascosauz]

@ecopsorn valley scores are about right for a 2070S. You sure you weren't running heaven on crazy settings or had an update going in the background?

Random restarts, BSODs and lockups are symptoms of an unstable ram profile. Go back to 16-16-16-36, take your DRAM voltage off auto and set it manually at 1.35V, add more voltage if DRAM calc recommends it. Remember that the 3600 /14 profiles all require a bit more voltage, at least 1.37V I think.

Hence why I've been saying that while high end B-die has a lot of potential, DRAM overclocking is an exhausting business. You can stop at any point you wish, but you still need to make sure your current profile is stable to the best of your ability. That means MT86 + P95 large + games.

Standard testing suite is minimum 4-8 runs of memtest86 booting from a USB, 1hr of P95 Large FFTs, plus additional programs if you wish like a few runs of AIDA or membench. Then, it helps to play some more demanding games (eg. MW 2019/) that can heat up the RAM, although P95 Large will do the same thing.

With B-die, you can apply voltage a bit more liberally, but remember that the more Vdram you add, the more heat is produced. b-die is notoriously sensitive to heat, so when the sticks get hot to 50C, what is stable cold may not be stable anymore. When running P95 Large fpr example, use HWInfo to keep an eye on your DIMM temperatures.

 

[fwiler]

don't feel like all is resolved yet but it was definitely a step forward. With the new install I've had it 2 times now that the windows button (start menu) stopped working. I also had 2-3 restarts where the computer wouldn't restart, all lights and fans were still on though - had to manually reset.

Without keeping track of every change you are making, and you are making a lot, you'll never figure out why.
Make one change. Run stress test on cpu, mem, and gpu for a couple of hours.
If it works, then continue on. If not, and you lock up again, then you only have to make one change back and not wonder what went wrong.

Troubleshooting is methodical.

Although it probably won't help, as I don't believe this is a software issue, you should check your event viewer just in case it does give some clue.

 

[moproblems99]

Without keeping track of every change you are making, and you are making a lot, you'll never figure out why.
Make one change. Run stress test on cpu, mem, and gpu for a couple of hours.
If it works, then continue on. If not, and you lock up again, then you only have to make one change back and not wonder what went wrong.

Troubleshooting is methodical.

Although it probably won't help, as I don't believe this is a software issue, you should check your event viewer just in case it does give some clue.

Good advice. If I have a lot to do I will generally do a couple settings at a time to speed up the process.

 

[ecopsorn]

Without keeping track of every change you are making, and you are making a lot, you'll never figure out why.
Although it probably won't help, as I don't believe this is a software issue, you should check your event viewer just in case it does give some clue.

All the changes I did are documented in this thread after windows install. Which lot of changes do u mean? I’m basically only running benches and applied like 3 ram timing settings. I installed only the minimum so far - documented where The cb scores are.
Gonna check the event viewer later.

I’m just back on the stock ram settings now and let memtest86 do the testing
Gotta sleep

 

[Waldorf]

@birdie
Well, most buy something because of its (expected) performance,
and not so its just "good enough".
Should i be happy if my 500hp 2 door only gets 200hp and cant pass 80mph,
because i cant legally drive faster anyway and its still good enough to get groceries? No.
as long as i didn't get something for free, i personally would not be happy if perf isnt within 1-5% of what is advertised/expected..


@ecopsorn
For now, turn off updates for win and dont install stuff.
Lets get it stable, once that is the case, i recommend waiting for may update and clean install once the new build is out, to rule out possible issue from installing win while sys was not really stable.

large changes for ram i would quick test with memtest running test 5/6/7, which are usually most sensitive.
once that doesn't show errors, run full set about 5 times, after that running it more often will usually not making any diff.

prime is great to see how "unstable" your settings are.
the worse, the quicker it will throw errors.
faulty stuff shows within minutes, to tight settings usually 20 to 30 min,
anything past that put before 1h can most of the time fixed with upping voltage, so i would leave it at 1.35, lower timings so you don't get error under 30-40 min then add some v and see if it works.

running prime for 10-12h, better 24h is a good way to see that there arent any major things, but unfortunately not a guarantee.
there are ppl running prime for 20h no probs but rig still had issues.
so i tend to stop after running it for about 12h.


Edit:
Use windows security, or get bitdefender free.
Usually much less impact on perf/gaming, than many others.

 

[ecopsorn]

@ecopsorn valley scores are about right for a 2070S. You sure you weren't running heaven on crazy settings or had an update going in the background?

Random restarts, BSODs and lockups are symptoms of an unstable ram profile. Go back to 16-16-16-36, take your DRAM voltage off auto and set it manually at 1.35V, add more voltage if DRAM calc recommends it. Remember that the 3600 /14 profiles all require a bit more voltage, at least 1.37V I think.

Hence why I've been saying that while high end B-die has a lot of potential, DRAM overclocking is an exhausting business. You can stop at any point you wish, but you still need to make sure your current profile is stable to the best of your ability. That means MT86 + P95 large + games.

Standard testing suite is minimum 4-8 runs of memtest86 booting from a USB, 1hr of P95 Large FFTs, plus additional programs if you wish like a few runs of AIDA or membench. Then, it helps to play some more demanding games (eg. MW 2019/) that can heat up the RAM, although P95 Large will do the same thing.

With B-die, you can apply voltage a bit more liberally, but remember that the more Vdram you add, the more heat is produced. b-die is notoriously sensitive to heat, so when the sticks get hot to 50C, what is stable cold may not be stable anymore. When running P95 Large fpr example, use HWInfo to keep an eye on your DIMM temperatures.

good advice, will look into it. I took the day off today to do some more testing, the memtest86 showed 0 errors with these timings, but of course they are not tight at all.

The thing is, I did printout all possible scenarios from dram calc (A0, manual import thaiphoon), safe and fast, etc.) and none of these let me boot, just like with the old ram.
So either the new ram b-die is also sucky or I'm just expecting too much. I try to take it slow now and adjust 1 primary timing, try to boot, etc.

About the heaven score, here I posted the score with 1080p, I did always run it on my native resolution 1440.

Ram temps at idle after computer was running overnight to do memtest look like this:

@ecopsorn
For now, turn off updates for win and dont install stuff.
Lets get it stable, once that is the case, i recommend waiting for may update and clean install once the new build is out, to rule out possible issue from installing win while sys was not really stable.

large changes for ram i would quick test with memtest running test 5/6/7, which are usually most sensitive.
once that doesn't show errors, run full set about 5 times, after that running it more often will usually not making any diff.

prime is great to see how "unstable" your settings are.
the worse, the quicker it will throw errors.
faulty stuff shows within minutes, to tight settings usually 20 to 30 min,
anything past that put before 1h can most of the time fixed with upping voltage, so i would leave it at 1.35, lower timings so you don't get error under 30-40 min then add some v and see if it works.

running prime for 10-12h, better 24h is a good way to see that there arent any major things, but unfortunately not a guarantee.
there are ppl running prime for 20h no probs but rig still had issues.
so i tend to stop after running it for about 12h.

Edit:
Use windows security, or get bitdefender free.
Usually much less impact on perf/gaming, than many others.

also good advice, try to test as much as possible today. And about antivirus, I'm listening to the advice of you guys and ditch ESET. I think I'm just sticking with windows security for now.

 

[tabascosauz]

@ecopsorn that's not the XMP profile, and that's not DRAM calc's highly suspect A0 recommendation either. If you're having trouble with these profiles out of the calculator, start tweaking from XMP as a foundation.

First things first, MT86 doesn't prove anything on its own. Memtest86 is a baseline test you usually run when you have brand new sticks and you don't know if they have any serious errors or not. Once you've established once that your kit doesn't have major issues, its importance diminishes in the context of overclocking. It does NOT work the IMC or the RAM very hard, hence the cool temperatures you see. That's why it's imperative to move to demanding tests such as P95 Large FFTs and some extended period of intense gaming, to see how the RAM performs when it's hot. Like 45-50C. And while you're in the OS, you may as well throw some extra membench or memtest64.exe utilities in the mix.

Have you adjusted the DRAM voltage as needed, as I've mentioned before? You will not hit the two 3600 CL14 profiles without moving past at least 1.37V, guaranteed. You may need to go up to 1.4V. Shouldn't be a problem with airflow.

The issue without having DRAM voltage monitoring on your board is that we don't know what the voltage regulation on the DDR side is like. Is it automatically overvolted beyond whatever setting you enter (as Gigabyte does), or is the LLC weak and allows the Vdram to drop at times far below the voltage setting you've entered?

Also, whatever profile you're on, turn Gear Down Mode off. It's 0.5ns in latency, the equivalent of 1.5T command rate, and won't help you at these speeds.

 

[ecopsorn]

Hi @tabascosauz - yeah not seeing dram v in windows sucks.
I did went up with voltage up to 1.4 so far.
This is my latest booting config. @1.38V.
Very strangely on this one I had gear down power down on auto, when I tried to disable it, I got error F9 and couldnt boot?

I did went up with voltage up to 1.4 so far but cl14 won’t post yet.

as a general note for everyone - I could just accept the cl16 timing and be done with it, but somehow I want this challenge
Hell I’m even still considering changing the board just because lol.

and about xmp, ppl suggested just go manual instead of xmp as a base - gets me confused haha

 

[oxrufiioxo]

Xmp/docp is very dependent on your board on how well performs and there are 3600 cl14 and 3800 CL14 kits if you wanted to gurantee yourself those speeds.


I'm actually tempted to pick up the 3800 cl14 kit as my board is on the qvl

 

[tabascosauz]

@ecopsorn An enthusiastic attitude is great, but at least establish a baseline at 16-16-16-36 and make sure it's stable. Otherwise, you'll be going through the profiles and have nothing working to fall back on. XMP is literally just an Auto button. Sometimes it doesn't work so well for some people. It's just a freq + set of timings + 1.35V at a push of a button.

Never really seen tRP and tRCD lower than CL. I don't know what implications for stability there are in doing so. Usually you'd just go to CL15 at that point, but you'd need GDM off for odd number tCL to work. Error F9 is unstable RAM.

tRAS = tCL + tRCD(RD) + 2

The tRAS of 32 makes no sense, it doesn't follow the above rule and may actually cost you performance. Also, fix your tRRDS/tRRDL/tFAW/tWR. tWR is whatever works, doesn't need to correspond to the others.



Also, do any programs other than HWInfo report DRAM voltage? And not what you've put in the setting, the actual live reading of Vdram. I have a feeling that DRAM voltage is being set by the board lower than the value you enter, but until you can make sure of the actual DRAM voltage I wouldn't go acting on that suspicion.

 

[ecopsorn]

@tabascosauz
Oh yes I did have some screwed settings there.
I did read somewhere to focus on the main timings first and than worry about subtimings, thats why I never paid attention to tRRFS tRRDL and such.

about my screwed tras number - I didn’t know that formula u posted, thx a bunch. i was trying to change one value after the other and test, but now I know how to to it better

here's a ryzen master screenshot - seems all differently named so u gotta help me out. but could VDDCR SOC be the DRAM voltage? and CLDO VDDP soc, just the SOC Voltage?

Got gear down power down disabled now but figured out that computer wants 2T now instead of 1T, whats the reason for that again?

 

[tabascosauz]

@ecopsorn 2T is easier to run, more stable. 1T is faster, less stable at higher speeds/difficult timings. GDM is in the middle, like 1.5T of sorts.

Uhhh VDDIO is DRAM voltage. Worrying thing is mine just shows is as a static value, the current BIOS setting. Can't think of any reason it's 0. I've never seen it as 0 before. A quick search came up with nothing.

I know for a fact that Asus' ROG AM4 boards have DRAM voltage monitoring like literally every other board out there, so either no one with your board has ever done any proper RAM overclocking before, or something is going on here.

 

[ecopsorn]

Xmp/docp is very dependent on your board on how well performs and there are 3600 cl14 and 3800 CL14 kits if you wanted to gurantee yourself those speeds.


I'm actually tempted to pick up the 3800 cl14 kit as my board is on the qvl

thanks mate but I just got new kits. I’m sure if I’d get cl14 kits I’d want it to be lower as well lol