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

[nguyen]

So for the less informed of us, can you tell me why the board affects memory when the memory controller is in the CPU? What is contained in the board besides the slot and traces that would affect memory compatibility?

I have never bought qvl in my life. Also, never had a problem. And no, not being able to boot my Neo over 3200 isn't a problem because that is all AMD guarantees.

I never bothered with QVL either when I bought the 8700K, they just work at XMP, overclock beyond that is luck with Motherboard, Ram and CPU IMC (Furthermore I don't gain any mhz using 2 DIMM instead of 4 as my Z370 board use T-Topology design); but for Ryzen which is extremely finicky with RAM, I tend to look for Samsung B-die, which is not noted anywhere but the motherboard manufacturer QVL or the RAM advertisements.
Sure you could have a CPU with a dud IMC, but in that case looking the the QVL would tell you there is something wrong with your CPU no ?

Just look at the entire QVL, they are comprehensive, the Asus X570-E Gaming even work with select 4x16GB kit...

 

[moproblems99]

personally have a 2 x 16GB Trident Z RGB with DJR. The XMP was originally 3600/17-19-19, a little bit looser than the common 3600/16-19-19 Tridents. I run 3600/16-19-20 at the XMP 1.35V, it runs fine. I'd expect the 3600/16-19-19 to be possibly have a bit more OC potential.

I can't get my Neos less than 16-19-x-x either. When I am less lazy I am going to try my Samsung A-Die a that can do 15-15-15-something according to ThaiBurn.

Sure you could have a CPU with a dud IMC, but in that case looking the the QVL would tell you there is something wrong with your CPU no ?

Except the same ram in the same board works at 3800 with a different cpu so can't be the board, no?

 

[ecopsorn]

Try the NVMe driver here, it might help improve the IOPS performance.

Samsung 970 EVO Plus PCIe 3.0 SSD | Samsung Semiconductor Global

Samsung 970 EVO Plus maximizes the potential of NVMe™ bandwidth for unbeatable computing. In capacities up to 2TB, with reliability of up to 1,200 TBW.

www.samsung.com

I did install the driver and can confirm that I didn't have it installed before. Results though are similar, even worse for write lol

Getting the stuff done now as promised, report back soon

 

[nguyen]

I can't get my Neos less than 16-19-x-x either. When I am less lazy I am going to try my Samsung A-Die a that can do 15-15-15-something according to ThaiBurn.

Except the same ram in the same board works at 3800 with a different cpu so can't be the board, no?

So let me get this straight,
just tell me which one of these has a higher chance of happening:
1. Ram maker change ICs
2. CPU with Dud IMC or a dud motherboard
3. Memory support not listed in QVL or listed as not working but they work in real life
4. Memory support work or not work as intended in the QVL

I did install the driver and can confirm that I didn't have it installed before. Results though are similar, even worse for write lol
View attachment 154634

Getting the stuff done now as promised, report back soon

Sorry to ask now but have you install the AMD chipset driver ?

https://www.amd.com/en/support/chipsets/amd-socket-am4/x570

 

[tabascosauz]

I can't get my Neos less than 16-19-x-x either. When I am less lazy I am going to try my Samsung A-Die a that can do 15-15-15-something according to ThaiBurn.

Except the same ram in the same board works at 3800 with a different cpu so can't be the board, no?

Yeah, it's just something of a hard wall for Hynix. It's like how only Samsung can do unusually low tRFC. Despite its general reputation as a trash tier overclocker, my 4Gb D-die would probably be faster than my DJR if it it could be OCed to 3600. I just can't be bothered because there's not an IC that's more loathed than geriatric 4Gb D-die.

To be honest, during the period of time when I was running 16-18-18, benchmark differences were pretty minimal at 3600. Performance differences probably nonexistent between 3600 16-18 and 16-19. The gap might widen a little if it was 16-16-16 instead, or at 3800.

Also, I'm not sure what this other guy and his QVL crusade is all about. Is this some sort of personal vendetta against G.skill and their use of CJR/DJR?

 

[moproblems99]

Yeah, it's just something of a hard wall for Hynix. It's like how only Samsung can do unusually low tRFC. Despite its general reputation as a trash tier overclocker, my 4Gb D-die would probably be faster than my DJR if it it could be OCed to 3600. I just can't be bothered because there's not an IC that's more loathed than geriatric 4Gb D-die.

To be honest, during the period of time when I was running 16-18-18, benchmark differences were pretty minimal at 3600. Performance differences probably nonexistent between 3600 16-18 and 16-19. The gap might widen a little if it was 16-16-16 instead, or at 3800.

Also, I'm not sure what this other guy and his QVL crusade is all about. Is this some sort of personal vendetta against G.skill and their use of CJR/DJR?

Yeah, pretty much why I am not worried about the specs. Anything 3600-c16 something is perfectly fine.

So let me get this straight,
just tell me which one of these has a higher chance of happening:
1. Ram maker change ICs
2. CPU with Dud IMC or a dud motherboard
3. Memory support not listed in QVL or listed as not working but they work in real life
4. Memory support work or not work as intended in the QVL

#3.

Edit: do any board manufacturers actually list modules that don't work? And are they any modules that are proven to not work?

I did install the driver and can confirm that I didn't have it installed before. Results though are similar, even worse for write lol
View attachment 154634

Getting the stuff done now as promised, report back soon

That is really odd. I suspect there is a bios setting causing some issue like not being in ahci mode (but that hasn't been a thing in ages).

 

[nguyen]

#3.

Edit: do any board manufacturers actually list modules that don't work? And are they any modules that are proven to not work?

Just for once look over at the QVL ? some of the 4x8GB kit are listed as not working with 4 DIMM installed, would you buy them knowing Asus tested and they don't work with 4 DIMM installed ?
Or whatever, just buy what you want and if that don't work, swap out for a new kit anyways....that is if I live in the States, other place they don't give you exchange without valid reasons, and buying 4x8GB kits that were tested not to work on your motherboard is not one of them....

 

[ecopsorn]

yeah I surely don't wanna waste money but I'm also not willing to accept a 2800$ build to run like crap.
I've downloaded the latest DRAM calc 1.7.1 and this what it spits out for 3200 with imported taiphoon readout
View attachment 154586

not much difference to the old version and I'm 100% sure it won't POST like this.

I can confirm @TheLostSwede that these settings won't let me boot, error F9 on the mobo. I have tried with DOCCP, Auto and Manual setting. I have tried entering only the basics (first 5) and I have tried entering it all. I'm done trying anything for this piece of crap ram anymore.

 

[TheLostSwede]

Edit: do any board manufacturers actually list modules that don't work? And are they any modules that are proven to not work?

Not as such, they simply list the highest achievable speed. So a 3600MHz kit might be listed as only doing 2400MHz on whatever their specific test setup is.

Just for once look over at the QVL ? some of the 4x8GB kit are listed as not working with 4 DIMM installed, would you buy them knowing Asus tested and they don't work with 4 DIMM installed ?
Or whatever, just buy what you want and if that don't work, swap out for a new kit anyways....that is if I live in the States, other place they don't give you exchange without valid reasons, and buying 4x8GB kits that were tested not to work on your motherboard is not one of them....

Or you know, it's like I said, they only tested two modules out of that kit, as it was almost lunch time...

I can confirm @TheLostSwede that these settings won't let me boot, error F9 on the mobo. I have tried with DOCCP, Auto and Manual setting. I have tried entering only the basics (first 5) and I have tried entering it all. I'm done trying anything for this piece of crap ram anymore.

But did you set the tCL to 16, rather than 14? Normally it's either the tCL, or one of the next three sub timings, as well as tRFC that prevents boot. Also, in my experience, Ryzen 3000 series doesn't really like odd tCL timings, but that's obviously not the issue here.
If you're at 1.35V, then it really ought to work, but alas here we are...

 

[moproblems99]

@ecopsorn , this is quite and interesting problem. I need to recheck the thread but perhaps post the UEFI pic of storage controllers.

Just for once look over at the QVL ?

I looked at the image you posted, doesn't show not working. So let's look at that pic you posted. There are no 2 x 16 GSkill kits. Does that mean that board doesn't work with 2 x 16 GSkill kits? I can only by 2x8 or 4x8 if I want GSkill? Also, my last post. Don't want to derail the ops thread, who actually needs help.

 

[nguyen]

I can confirm @TheLostSwede that these settings won't let me boot, error F9 on the mobo. I have tried with DOCCP, Auto and Manual setting. I have tried entering only the basics (first 5) and I have tried entering it all. I'm done trying anything for this piece of crap ram anymore.

Not sure what you put in that Calculator give you a cas latency of 14
Try these settings

And lastly change the Command Rate to 2T in the bios,
After making sure that your RAM work stable, maybe follow the guide here regarding your slow booting

[SOLVED] - boot slow time solution on ryzen 3700x and x570

hello I have just build new pc I took a long time when i start windows and i dont know why It takes me around 15-20 sec to see the first bios picture AND 35-40 TO SEE MY DESKTOP is there setting should i do in bios my pc: ryzen 3700x asus x570 tuf gaming plus wifi Asus ROG Strix GeForce RTX™...

forums.tomshardware.com

I looked at the image you posted, doesn't show not working. So let's look at that pic you posted. There are no 2 x 16 GSkill kits. Does that mean that board doesn't work with 2 x 16 GSkill kits? I can only by 2x8 or 4x8 if I want GSkill? Also, my last post. Don't want to derail the ops thread, who actually needs help.

So you buy a 4x8GB kit and ASUS tested the same kit and found their board won't boot with 4 slots occupied, do you consider your 4x8GB kit to be working properly ?
Can anyone in this forum read a manual for once before they start advising people ?

 

[ecopsorn]

Your memory speed does not explain your extremely low CineBench 20 results.

1. Please try to load BIOS defaults, save them and do not change anything afterwards. Do not even enable XMPP/DCCP or anything - just pure defaults. You may disable additional controllers (e.g. Firewire) in BIOS which you're sure you'll never use.
2. Please then try to boot Windows in safe mode to eliminate any backgrounds tasks - make sure task manager reports 0% CPU load after you booted this way.
3. Then try running CM20 again and report your results.

Once you've done the above please load an XMPP profile (enable DCCP) and reboot into safe mode once again. Run CB20 and report your results again.

From your HWiNFO screen your system looks 100% fine - the CPU is using up to 120W of power which is line with a slightly overclocked Ryzen 7 3800X.

Your CPU temperatures look a little bit too high to my liking but maybe they are OK. I have a pretty cheap air cooling and in BIOS my 3700X CPU is 10C colder than yours.

@birdie , OMG I think we are getting somewhere.
I did exactly what you asked and these were the results! I did check the CPU and it was as at 0% load at idle.
And now it comes!!! When I start windows in normal boot and open the Performance tab, my CPU is at 25% IDLE (swapping between 16-29% - holy moly guacamole!

1.) Reset Bios settings by loading optimized defaults

than I went back and enablec DOCCP:

so the question to all of you now, what does that tell me?

• are my ram still shit and should be replaced?
• obviously I need a fresh windows install and will plan ahead what I will install including drivers, etc.
• it still doesn't explain the extremely slow boot time- Boot time is still 58sec until loading screen with the default or straight DOCCP bios settings

Sorry to ask now but have you install the AMD chipset driver ?

https://www.amd.com/en/support/chipsets/amd-socket-am4/x570

yes, got that one installed

But did you set the tCL to 16, rather than 14? Normally it's either the tCL, or one of the next three sub timings, as well as tRFC that prevents boot. Also, in my experience, Ryzen 3000 series doesn't really like odd tCL timings, but that's obviously not the issue here.
If you're at 1.35V, then it really ought to work, but alas here we are...

I tried it all man, really, not this time around but I had so many timings tested and even pretty low ones that I could actually boot with, but latency and scores were sucky still.

@ecopsorn , this is quite and interesting problem. I need to recheck the thread but perhaps post the UEFI pic of storage controllers.

will do, coming back in a sec.

Not sure what you put in that Calculator give you a cas latency of 14
Try these settings
And lastly change the Command Rate to 2T in the bios

tried it all already without real success

What the hell happened to the thread overnight?

@ecopsorn the Royal is a more high end kit because the timings pretty much guarantee it to be B-die. 16-16-16 flat, Rev.E and DJR can't do that. With B-die like that, if you know what you're doing, chances are its OC capabilities will leave your CPU in the dust. Read: your CPU's memory controller will limit you to 3600/3733/3800, not the kit. G.skill usually reserves its best ICs for Royal that has high freq tight timings.

Your Vengeance kit has B-die too, but one is literally bottom of the scrap heap and the other is a stone's throw from the crown jewel.

Hence why it's so expensive. But if you want great performance that'll work, and maybe optimize a bit and OC to 3733, the first Neo kit should work very well for you. Timings look like it shoud be CJR or DJR. The Royal is good, but it's got the price to back that up. The other Neo kit is priced out of its own league; Neos can come with B-die, but Neo is a new line for Ryzen while quite a number of Royal SKUs are G.skill's absolute cream of the crop for Intel.

I personally have a 2 x 16GB Trident Z RGB with DJR. The XMP was originally 3600/17-19-19, a little bit looser than the common 3600/16-19-19 Tridents. I run 3600/16-19-20 at the XMP 1.35V, it runs fine. I'd expect the 3600/16-19-19 to be possibly have a bit more OC potential.

@tabascosauz
thanks man, yeah I didn't expect this thread to get so heated .

I haven't decided yet between the 2x Neo or the 4x Royal. Besides speed, the bling of the royal would suit my build because I have a Guardians of the Galaxy theme. I could use some infinity stone ram lol

 

[Makaveli]

Your Score seems about right now.

Also windows 10 has a built in screen capture software either the snipping tool or sketch and snip on the newer builds you don't need to take screenshots with a cell phone.

 

[ecopsorn]

@moproblems99 - don’t know what I should screenshot from the bios in terms of storage. Seems all good, here’s a pic

 

[tabascosauz]

@ecopsorn Have your CPU temps changed after restoring defaults?

I personally like the brushed look of the RGB, Z and Neo better, but the Royals are gorgeous (apart from the font they chose for the heatspreader). And being 4 DIMMs, it visually fills up the banks in a way that 2 DIMM kits can't. I'm sure you are aware of the advantages and disadvantages of 2 vs 4 DIMMs, now, so I won't bore you on that.

It depends on what direction you want to take now. Are you content with the RAM performance as is? It doesn't appear like there will be any magic solutions for this Vengeance kit. As long as it's stable and you don't care about the minor gap in real world performance, I don't see too much problem with keeping it.

And if you do want another kit, your choices are laid out in front of you.

4900 is acceptable. I would personally want a little higher from a 3800X, but if game performance is where it should be, then no issues there. I'm also around the 4900 mark with my 3700X bone stock, with optimized PBO settings it goes up to 4980 but in terms of power and heat, it just isn't worth it for me to run it like that.

More important is that your CB single thread score should be above 510.

 

[ecopsorn]

@tabascosauz - well I’m not sure, I guess I only know more when I reinstalled windows with the guide provided here to hopefully get proper boot times. I’m still wondering why there is such a insane difference in score between safe mode and normal.

Unfortunately the temps haven’t changed with default bios settings

so my main issues are still

1. Temps (if that’s an issue)
2. Boot time
3. Cinebench score in normal boot (hopefully fixed with re-install win
4. GPU underperforming as well
5. m2 issue

i doubt that only new ram would fix it all but I gotta start somewhere

 

[tabascosauz]

@tabascosauz - well I’m not sure, I guess I only know more when I reinstalled windows with the guide provided here to hopefully get proper boot times. I’m still wondering why there is such a insane difference in score between safe mode and normal.

Unfortunately the temps haven’t changed with default bios settings

so my main issues are still

1. Temps (if that’s an issue)
2. Boot time
3. Cinebench score in normal boot (hopefully fixed with re-install win
4. GPU underperforming as well
5. m2 issue

i doubt that only new ram would fix it all but I gotta start somewhere

You benched in safe boot?

Safe boot yields consistently higher scores in all the benches that I've found to work in safe boot. My CPU-Z ST scores are close to what a 3900X/3950X can do. I'm starting to think that it's changed nothing for you with loading BIOS defaults.

New RAM should change boot time, and obviously your memory performance and latency. But a clean install should come first. I have no idea what sort of stuff you have on there.

What power plan are you using? Are you using the Ryzen power plans included in the drivers?

 

[ecopsorn]

You benched in safe boot?

Safe boot yields consistently higher scores in all the benches that I've found to work in safe boot. My CPU-Z ST scores are close to what a 3900X/3950X can do. I'm starting to think that it's changed nothing for you with loading BIOS defaults.

New RAM should change boot time, and obviously your memory performance and latency. But a clean install should come first. I have no idea what sort of stuff you have on there.

What power plan are you using? Are you using the Ryzen power plans included in the drivers?

yes in safe boot cause @birdie suggested it.

using the ryzen balanced power plan as suggested everywhere.

for windows it is pretty clean i just installed a month ago, so really no idea what’s the issue there.

 

[MxPhenom 216]

@birdie , OMG I think we are getting somewhere.
I did exactly what you asked and these were the results! I did check the CPU and it was as at 0% load at idle.
And now it comes!!! When I start windows in normal boot and open the Performance tab, my CPU is at 25% IDLE (swapping between 16-29% - holy moly guacamole!

1.) Reset Bios settings by loading optimized defaults
View attachment 154656

than I went back and enablec DOCCP:
View attachment 154657

so the question to all of you now, what does that tell me?

• are my ram still shit and should be replaced?
• obviously I need a fresh windows install and will plan ahead what I will install including drivers, etc.
• it still doesn't explain the extremely slow boot time- Boot time is still 58sec until loading screen with the default or straight DOCCP bios settings

yes, got that one installed


I tried it all man, really, not this time around but I had so many timings tested and even pretty low ones that I could actually boot with, but latency and scores were sucky still.


will do, coming back in a sec.


tried it all already without real success


@tabascosauz
thanks man, yeah I didn't expect this thread to get so heated .

I haven't decided yet between the 2x Neo or the 4x Royal. Besides speed, the bling of the royal would suit my build because I have a Guardians of the Galaxy theme. I could use some infinity stone ram lol
View attachment 154658
View attachment 154659

When your system is booting, does it look like its rebooting multiple times? Or does it just sit at the splash screen for a while until it finally loads windows?

 

[ecopsorn]

When your system is booting, does it look like its rebooting multiple times? Or does it just sit at the splash screen for a while until it finally loads windows?

It just boots slow, rog strix logo on screen for like 45sec than 12sec black screen and than i see the login screen.

when i try timings that won’t work, system dies try to boot 3x and than it ends up at error 22, F9 or so and if I am lucky I end up back in bios. At 22 I need to CMOS reset

 

[birdie]

@birdie , OMG I think we are getting somewhere.
I did exactly what you asked and these were the results! I did check the CPU and it was as at 0% load at idle.
And now it comes!!! When I start windows in normal boot and open the Performance tab, my CPU is at 25% IDLE (swapping between 16-29% - holy moly guacamole!

It tells you that there are a lot of BS'ers in this thread who

1) Believe that SSD speed may affect CB scores and affects Windows boot time by more than 0.25 seconds
2) Believe that drivers for an SSD decide everything
2) Believe that memory can run so bad it results in > 20% of performance being lost in a CPU bound test. CB barely touches RAM at all, it's not 7-Zip or WinRAR - the only tests which are very sensitive to RAM speeds.

What does it tell you?

Well, my honest advice will be

1) Run the Windows boot media tool and prepare a USB flash disk for installation.
2) Install Windows from scratch (on a new fresh device, or simply wipe/reformat everything), and then do not install any software or drivers afterwards. Maybe let Windows install all the updates and reboot several times just to be sure you're done with updates. The Windows Updates service can be a dick to your benchmarking sessions.
3) Run CB20 and remember the scores.
4) Install e.g. AMD chipset drivers, rerun CB20 and check that scores haven't changed
5) Install GPU drivers and check that scores haven't changed

If you're here now enjoying your full performance, it means you had something running in your previous W10 install which negatively affected your scores. Do not install everything at once. Do not play with Windows power schemes. I will go as far as to claim that AMD chipset drivers are not necessary to get your Ryzen system running almost perfectly. Intel INF drivers are often equally useless.

 

[moproblems99]

so the question to all of you now, what does that tell me?

• are my ram still shit and should be replaced?
• obviously I need a fresh windows install and will plan ahead what I will install including drivers, etc.
• it still doesn't explain the extremely slow boot time- Boot time is still 58sec until loading screen with the default or straight DOCCP bios settings

No, not necessarily. The RAM seems fine for what it is. When 25% of your CPU is used, what is shown as using it?

 

[xman2007]

It tells you that there are a lot of BS'ers in this thread who

1) Believe that SSD speed may affect CB scores and affects Windows boot time by more than 0.25 seconds
2) Believe that drivers for an SSD decide everything
2) Believe that memory can run so bad it results in > 20% of performance being lost in a CPU bound test. CB barely touches RAM at all, it's not 7-Zip or WinRAR - the only tests which are very sensitive to RAM speeds.

What does it tell you?

Well, my honest advice will be

1) Run the Windows boot media tool and prepare a USB flash disk for installation.
2) Install Windows from scratch (on a new fresh device, or simply wipe/reformat everything), and then do not install any software or drivers afterwards. Maybe let Windows install all the updates and reboot several times just to be sure you're done with updates. The Windows Updates service can be a dick to your benchmarking sessions.
3) Run CB20 and remember the scores.
4) Install e.g. AMD chipset drivers, rerun CB20 and check that scores haven't changed
5) Install GPU drivers and check that scores haven't changed

If you're here now enjoying your full performance, it means you had something running in your previous W10 install which negatively affected your scores. Do not install everything at once. Do not play with Windows power schemes. I will go as far as to claim that AMD chipset drivers are not necessary to get your Ryzen system running almost perfectly. Intel INF drivers are often equally useless.

I've been following the thread and I dont believe that anyone suggested the r20 scores were affected by the SSD speed, but that is a seperate issue he has, so it was suggested to install the samsung drivers to see if it fixed that particular problem, neither was it pointed out that would "decide everything" as you have put it. Also IF plays a big part in CPU performance with Zen 2 and having 1600 IF compared to 1800/1900 can and will have a negative effect on CPU bound benches.

Agree with a clean Windows install though I believe the OP has done this previoulsy though I could be mistaken and is a good call otherwise, also he may just have to live with the fact his RAM and specific config wont run at anything above 3200 which I think he has and has decided to buy a more compatible kit that he can at least run at 3600 to get the best out of both the RAM and 3800x which is a good call imo

 

[birdie]

"IF playing a big part" is another egregious overstatement but then this whole thread has been mostly a laughing stock so far. From the benchmarks I've seen so far it bestows or undermines benchmark scores by a miniscule amount. The first piece of advice I gave immediately resulted in the improvements the OP was looking for. Not fiddling with memory timings, QVL certifications, memory topology, BIOS settings, drivers, power schemes and other similar nonsense. Have fun discussing new ingenious ways of fixing the OP issues.

@ecopsorn

Please reapply a thermal paste if you were the one who built your rig. If someone else did it, please ask them if you're allowed to do that and if it doesn't void your warranty.

I would not recommend OC'ing a Ryzen 3000 desktop CPU or even enabling PBO for two reasons: performance increases will be minimum while power consumption and temps will jump through the roof.

 

[xman2007]

"IF playing a big part" is another egregious overstatement but then this whole thread has been mostly a laughing stock so far. From the benchmarks I've seen so far it bestows or undermines benchmark scores by a miniscule amount. The first piece of advice I gave immediately resulted in the improvements the OP was looking for. Not fiddling with memory timings, QVL certifications, memory topology, BIOS settings, drivers, power schemes and other similar nonsense. Have fun discussing new ingenious ways of fixing the OP issues.

@ecopsorn

Please reapply a thermal paste if you were the one who built your rig. If someone else did it, please ask them if you're allowed to do that and if it doesn't void your warranty.

I would not recommend OC'ing a Ryzen 3000 desktop CPU or even enabling PBO for two reasons: performance increases will be minimum while power consumption and temps will jump through the roof.

Not fiddling with Bios settings or windows settings? you told him to reset to default and boot up in safe mode or does that not count, you jumped in and thought you'd fixed 4 pages with one suggestion that only work/showed an improvement in 1 area when he is in safe mode, maybe he can run his PC in safe mode for MAX performance 24/7 yay... well done, you won the internet tonight, take a bow...