Ryzen Owners Zen Garden

[mplayerMuPDF]

Reinstalling Windows shouldn't help, it's either a firmware/hardware problem. There's not too many things you can try

• Chipset drivers? Latest is from August '22
• Playing with VSOC/VDDG_IOD to see if anything helps
• Keeping/getting rid of DAC drivers (probably only works for crappy Focusrite)
• Cstates off maybe? Generally not a great idea for idle power and sometimes performance

As you said yourself, it's really board dependent. AMD has tried to solve this problem like 3 distinct times through AGESA. But if your board just has it then sometimes your only solution is go to a different board. My B550M TUF Wifi couldn't get away from the USB issues but my Impact and all my ITX boards have never had any problems.

Though, on a lot of AM4 block diagrams I've seen, the onboard audio is usually connected to the CPU not the PCH. Not sure what's going on there with your board, the audio popping issues usually are from external audio devices and part of the USB problems on AM4

So if you get a different board you might get away from the audio issues but get the USB issues instead or vice versa... Man, I like my 2600 lite/Zen+ and Ryzen in general (until the Pluton garbage/6000 series, at least) but this stuff is just absolutely unacceptable to me. It's probably partly due to how I use my PC/what I value but to me this stuff is a big deal, much more so than getting 10-20% more performance, 2 more cores or whatever.

Last i checked, i was able to get just under or above 60ns
I need to figure out whats messing with me, before i slap in the x3D (just arrived)

Anyone got recommended benchmarks and such to do before/after with? R23 obviously isnt the best choice
View attachment 265085


SoC voltage to 1.15V, PCI-E 3.0 - dont use PCI-E risers or USB extensions for high power devices
The top four USB ports are direct from CPU and usually better, but instability from connected devices feeds back into it
Most onboard audio is PCI-E connected, errors on the bus (WHEA included) can cause everything from crackles and issues, to bus resets (the infamous USB dropouts)
Since i solved the memory issues (faulty stick) on my 2700x system, i've not had any issues in months

So some USB ports are connected directly to the CPU? I just checked to see if this is why I haven't had any more USB issues with my keyboard and mouse at least in a while (since the last AGESA update at least but probably longer) but it turns out my mouse is actually in the Startech 4*USB 2.0 header adapter, so definitely not connected directly to the CPU (or is it?) I was constantly switching stuff around when I was experiencing those issues. Now it is just the USB WiFi every so often and IIRC simply replugging into the same port fixes it.

 

[tabascosauz]

So if you get a different board you might get away from the audio issues but get the USB issues instead or vice versa... Man, I like my 2600 lite/Zen+ and Ryzen in general (until the Pluton garbage/6000 series, at least) but this stuff is just absolutely unacceptable to me. It's probably partly due to how I use my PC/what I value but to me this stuff is a big deal, much more so than getting 10-20% more performance, 2 more cores or whatever.


So some USB ports are connected directly to the CPU? I just checked to see if this is why I haven't had any more USB issues with my keyboard and mouse at least in a while (since the last AGESA update at least but probably longer) but it turns out my mouse is actually in the Startech 4*USB 2.0 header adapter, so definitely not connected directly to the CPU (or is it?) I was constantly switching stuff around when I was experiencing those issues. Now it is just the USB WiFi every so often and IIRC simply replugging into the same port fixes it.

Each board has a different distribution of which ports are CPU, which ports are PCH, and which ports are third party controller (if applicable). You have to look at your board specs/manual/block diagram to find out. This isn't an AMD exclusive thing.

Ryzen was a lot simpler in all respects back in the 2000 days.

I won't get into whether these issues are acceptable or whether Intel is better. But to be entirely fair, on both Intel and AMD platforms there are so many issues that come not out of platform firmware but out of vendor- and board-specific implementations, that in the case of a lot of issues, the only solution is to go to a different board. I don't view that as an AMD exclusive.

But I can't deny that dealing with a number of those problems wasn't one of the factors pushing me towards switching to Comet Lake, at the time.

 

[Mussels]

@Mussels did you set yourself up for benching? ie. no background programs, disable internet + reboot once, disable search service if necessary, no HWInfo or other monitoring for absolute best latency, etc.

Or just straight up safemode works too. Give your timings a quick check too.

Copy looks a little low (closer to 3200CL16 bandwidth). L3 latency is really high, seems like you didn't get rid of background stuff. First run usually suffers too

You can double click on any one box to just test that field (ie. DRAM latency). Saves time and allows you to bench back to back to see if some background process is interfering

Expect a 3-4ns penalty from the 5800X3D on both DRAM and L3 latencies.

I mean, it's DRAM performance so you wouldn't be using R23 either. Maybe Linpack, Geekbench3 memory score or the Membench inside DRAM Calc? Although, none of them will tell you anything meaningful without a normal baseline

No, it's my usual setup - but quit running programs
I peaked at 59ns with everything quit, usually low 60's

That was at 3900, i dropped to 3800 a few weeks back diagnosing a crash (nvidia driver in the end, and never set it back)
Ram speed and IF are 1:1

Now it's 70-75 with everything quit and i'm not sure why, but a lot has changed - drivers, windows to the new build, etc.

This was retests

@Mussels did you set yourself up for benching? ie. no background programs, disable internet + reboot once, disable search service if necessary, no HWInfo or other monitoring for absolute best latency, etc.

Or just straight up safemode works too. Give your timings a quick check too.

Copy looks a little low (closer to 3200CL16 bandwidth). L3 latency is really high, seems like you didn't get rid of background stuff. First run usually suffers too

You can double click on any one box to just test that field (ie. DRAM latency). Saves time and allows you to bench back to back to see if some background process is interfering

Expect a 3-4ns penalty from the 5800X3D on both DRAM and L3 latencies.

I mean, it's DRAM performance so you wouldn't be using R23 either. Maybe Linpack, Geekbench3 memory score or the Membench inside DRAM Calc? Although, none of them will tell you anything meaningful without a normal baseline

I figured out the AIDA scores - it's because i'm running an all core OC, and not PBO. that 500Mhz drop in cache speeds has a hit.


I was just after something to compare with that's easy to run, in about 30 minutes i'll drain the loop

So if you get a different board you might get away from the audio issues but get the USB issues instead or vice versa... Man, I like my 2600 lite/Zen+ and Ryzen in general (until the Pluton garbage/6000 series, at least) but this stuff is just absolutely unacceptable to me. It's probably partly due to how I use my PC/what I value but to me this stuff is a big deal, much more so than getting 10-20% more performance, 2 more cores or whatever.


So some USB ports are connected directly to the CPU? I just checked to see if this is why I haven't had any more USB issues with my keyboard and mouse at least in a while (since the last AGESA update at least but probably longer) but it turns out my mouse is actually in the Startech 4*USB 2.0 header adapter, so definitely not connected directly to the CPU (or is it?) I was constantly switching stuff around when I was experiencing those issues. Now it is just the USB WiFi every so often and IIRC simply replugging into the same port fixes it.

Yes, four USB 3.0 5Gb ports are direct to the CPU. These are the most reliable and have the best power delivery (best one to use my Rift S on, regardless of which ryzen CPU was in these boards)

As said tho, they can suffer disconnects if attached devices are erroring. PCI-E risers and unstable memory were common causes, it's why some people have never ending issues but others are problem free - because it's usually caused by an external piece of hardware

I'll do it the lazy, barely repeateable way:


DRG, 4K ultra, DX12 all settings ultra, DLSS auto
Starting point, turn left to focus on the display screen
GPU unlocked to full wattage

Since i play this a lot, let's see how that changes with the new CPU

 

[mplayerMuPDF]

Each board has a different distribution of which ports are CPU, which ports are PCH, and which ports are third party controller (if applicable). You have to look at your board specs/manual/block diagram to find out. This isn't an AMD exclusive thing.

Ryzen was a lot simpler in all respects back in the 2000 days.

I won't get into whether these issues are acceptable or whether Intel is better. But to be entirely fair, on both Intel and AMD platforms there are so many issues that come not out of platform firmware but out of vendor- and board-specific implementations, that in the case of a lot of issues, the only solution is to go to a different board. I don't view that as an AMD exclusive.

But I can't deny that dealing with a number of those problems wasn't one of the factors pushing me towards switching to Comet Lake, at the time.

I mean, I guess it is theoretically possible, but it seems very unlikely to me that this Biostar X470GTQ just happens to coincidentally independently have USB issues when B450/X470 were in general plagued by this issue. I certainly don't mean to suggest that Intel never has had or has issues but I still find these issues unacceptable when buying a high-end chipset such as X470. Now, maybe this is normal and what the DIY PC world is like in general. If that is the case then next time I will buy an industrial motherboard or thin client with Ryzen Embedded SoC instead or something similar. If that makes me weird, then so be it. I prefer having the basics work reliably over having 12 cores, PCIe 5.0 or some other gizmo. When I was using a Kaveri-based EliteBook as a desktop replacement (connected to my monitor and keyboard etc), I never once had any issue with IO AFAIR.

 

[Count von Schwalbe]

Now, maybe this is normal and what the DIY PC world is like in general.

Fixed it for ya. Got an Intel based Dell and it has a lot of random tiny issues. USB dropping out, weird iGPU artifacts in MS Office, Chrome crashing, random freezes, etc. It's a computer thing.

I will buy an industrial motherboard or thin client with Ryzen Embedded SoC

AMD 4800S desktop kit? AKA the PS5 without iGPU?

 

[mplayerMuPDF]

Yes, four USB 3.0 5Gb ports are direct to the CPU. These are the most reliable and have the best power delivery (best one to use my Rift S on, regardless of which ryzen CPU was in these boards)

As said tho, they can suffer disconnects if attached devices are erroring. PCI-E risers and unstable memory were common causes, it's why some people have never ending issues but others are problem free - because it's usually caused by an external piece of hardware

Hmm, seems like my mouse has been connected to the USB 2.0 header and my G84-5200 keyboard is to a 3.0 port (don't know if it is CPU direct). I don't remember to what port my old CM Storm QuickFire XT was last connected. Now, that I think of it I have had this mouse from the beginning with this desktop and I have had my QuickFire XT and USB WiFi (well, technically it is another one but I have had issues with both on this computer) longer. I wonder if this cheap mouse may have been responsible for my USB issues. On the other hand, I have, certainly for a long time, only had issues where (often on boot or after resuming from sleep) some device does not work until I replug it (sometimes in another port) but after that I stop having issues, at least until the next reboot, so no truly random disconnects or anything like that. And when I experience it I always get the same or very similar dmesg output on Linux (I have posted it before in another thread) indicating that a USB port is not working correctly. I don't have any kind of PCIe riser and I have always had this Crucial RAM installed which should be good quality memory. I have never OCed anything on this system either, actually I discovered recently that I don't even have XMP enabled so my RAM is running underclocked technically @ 2667 MHz (it is rated for 3200 MHz).

Fixed it for ya. Got an Intel based Dell and it has a lot of random tiny issues. USB dropping out, weird iGPU artifacts in MS Office, Chrome crashing, random freezes, etc. It's a computer thing.

AMD 4800S desktop kit? AKA the PS5 without iGPU?

Asrock Industrial, for example, sells Ryzen Embedded (V and R series) Mini ITX motherboards. There are HP thin clients (e.g. T640/T740) with V and R series SoCs (succeeding the old cat and construction core based G series SoCs) that you can buy on eBay as well.

 

[tabascosauz]

I mean, I guess it is theoretically possible, but it seems very unlikely to me that this Biostar X470GTQ just happens to coincidentally independently have USB issues when B450/X470 were in general plagued by this issue. I certainly don't mean to suggest that Intel never has had or has issues but I still find these issues unacceptable when buying a high-end chipset such as X470. Now, maybe this is normal and what the DIY PC world is like in general. If that is the case then next time I will buy an industrial motherboard or thin client with Ryzen Embedded SoC instead or something similar. If that makes me weird, then so be it. I prefer having the basics work reliably over having 12 cores, PCIe 5.0 or some other gizmo. When I was using a Kaveri-based EliteBook as a desktop replacement (connected to my monitor and keyboard etc), I never once had any issue with IO AFAIR.

I didn't say your board is the only one affected. I said that all these "fixes", short of using a different board, don't have a great track record of working. AMD only got around to seriously fixing the USB issues on AGESA 1200-1202, which was well after the launch of Ryzen 5000 and B550, and clearly even that didn't do the trick for everyone.

Every device is only as good as the firmware the hardware/board vendor is willing to write for it. If Asus/Gigabyte/MSI/ASRock are already struggling to cope with AGESA and setting an incredibly low bar for reliability, for minor vendors like Biostar and Colorful the bar is probably just lying on the ground.

Intel's advantage is that they don't have a product model that requires continuous support for new features in board firmware over time. They aim to have firmware ready at launch for the feature set advertised at launch. What you see is what you'll ever get, for the most part. Their idea of big firmware controversy is Rocket Lake launching without final microcode, patched in a week's time. Your Kaveri laptop had the same philosophy, as do most Intel mobility platforms (except where mitigations for major vulnerabilities like Plundervolt are concerned), which is why you can have some legitimate expectations as to stability.

AM4 isn't the same. To support 4+ generations of wildly different CPUs, AGESA has had to change significantly. Go find your PBO settings in the launch BIOS for your X470 board, then go see if it's in the same place in a recent AGESA 1206/1207 BIOS. Curve Optimizer being introduced late for Ryzen 5000. Split VDDGs that don't exist in monolithic parts. ReBAR appearing on X370 boards.

Regardless of whether you view it as a good thing or a bad thing, if you're expecting top notch consistency from AM4, you gotta manage those expectations.

 

[Mussels]

4.55GHz vs 4.6GHz, i can see this being so similar

Totally not optimised or anything yet, i loaded the same BIOS profile as well - so it's got some mixed settings i need to check/reset too

Hmm, seems like my mouse has been connected to the USB 2.0 header and my G84-5200 keyboard is to a 3.0 port (don't know if it is CPU direct). I don't remember to what port my old CM Storm QuickFire XT was last connected. Now, that I think of it I have had this mouse from the beginning with this desktop and I have had my QuickFire XT and USB WiFi (well, technically it is another one but I have had issues with both on this computer) longer. I wonder if this cheap mouse may have been responsible for my USB issues. On the other hand, I have, certainly for a long time, only had issues where (often on boot or after resuming from sleep) some device does not work until I replug it (sometimes in another port) but after that I stop having issues, at least until the next reboot, so no truly random disconnects or anything like that. And when I experience it I always get the same or very similar dmesg output on Linux (I have posted it before in another thread) indicating that a USB port is not working correctly. I don't have any kind of PCIe riser and I have always had this Crucial RAM installed which should be good quality memory. I have never OCed anything on this system either, actually I discovered recently that I don't even have XMP enabled so my RAM is running underclocked technically @ 2667 MHz (it is rated for 3200 MHz).


Asrock Industrial, for example, sells Ryzen Embedded (V and R series) Mini ITX motherboards. There are HP thin clients (e.g. T640/T740) with V and R series SoCs (succeeding the old cat and construction core based G series SoCs) that you can buy on eBay as well.

sleep mode related issues are almost always RAM if it wont wake, and SoC/chipset voltages with USB issues. Voltage is drooping when it wakes up, not enough to reset the specific device and it 'hangs'

You can easily try an externally powered USB hub, 2.0 or 3.0 - it'll help prevent a lot of issues since the external power will keep the devices 'awake' the entire time

 

[tabascosauz]

4.55GHz vs 4.6GHz, i can see this being so similar

Totally not optimised or anything yet, i loaded the same BIOS profile as well - so it's got some mixed settings i need to check/reset too

Always should clear CMOS when swapping CPUs, X3D's Tweaker screen is very different

Write looking a bit wack but things look fine. CPU-Z ST a little low but carryover dirty settings from 5800X and being on a AGESA 1207 BIOS can easily account for that.

AIDA is usually very optimistic on its displayed speed, you'll find usually there's a bit of a wall at 4.45 due to how the boost works on this CPU. 4.55 is still Fmax, but think of it like the i7-8086K problem of reaching 5.0 only on strictly 1 core - only running purely a 1T load with nothing else going on will you ever see 4.55. Different from most other Zen 3 behaviour. Sometimes that last 100MHz really feels a bit like TVB for Intel

 

[Mussels]

So far:
Many of the modded settings do nothing, or change what windows things is the base clock but get over-ridden, as expected.

-30 in the BIOS
PBO set to motherboard
Nothing else can really be changed:
Not a clean OS, just quit my running programs:

Most things quit but hardly a benchmark environment:
TIM may need time to settle in or even need reapplication, this cooler sadly isn't the best for AM4 despite it being made specifically for them (fucking EK)

About 2.7%, 350 points higher than w1zzard and i've had this installed and running for about 20 minutes of playtime
AMD Ryzen 7 5800X3D Review - The Magic of 3D V-Cache - Rendering | TechPowerUp

 

[tabascosauz]

@Mussels all core Fmax is basically 4450. If you're pushing more than 110W for that then you're wasting that power, set lower limits. I'm at 15100, just under 100 watts, around 4.4GHz all core and 75-80C. To get meaningfully farther than 15000 you'll have to dig into BCLK because of the 4450 limit.

I wouldn't blame EK so quickly, it's just a hot chip. Yes it clocks low, yes Vcore is low, but it's just a hot chip. The point is to get optimal performance without having to worry about RAM/Fabric or clocks. You won't be anywhere near that in games.

You can also play with additional positive/negative vcore offset if it helps you out

 

[Mussels]

DRG, 4K ultra, DX12 all settings ultra, DLSS auto
Starting point, turn left to focus on the display screen
GPU unlocked to full wattage
View attachment 265100

Since i play this a lot, let's see how that changes with the new CPU
View attachment 265101

And with the new CPU: Pretty much the same GPU limited or even game limited, one assumes

And then my undervolted GPU settings:
oh no, 99% is at 226FPS for 140W less -.-

 

[Ruru]

Let's try with SOC 1.15V

Get yourself external audio and free yourself from onboard sound forever....

I carry over my Fiio K5 Pro and Micca Origain (or x amp/rec) to every system, same excellent audio every rig it goes into.

I said that the crackling happens with DAC as well..

 

[Mussels]

Ehehehehehee.

Sadly, WHEA's appeared over time. Still fun to even boot this high.

Let's try with SOC 1.15V



I said that the crackling happens with DAC as well..

Oh, looking up WHEA info on x3D and found this

"I was able to fix my USB Dropouts and Sound Issues by increasing the vddg iod voltage"

 

[Ruru]

Oh, looking up WHEA info on x3D and found this

"I was able to fix my USB Dropouts and Sound Issues by increasing the vddg iod voltage"

I put both VDDG to 1.05V. If it doesn't help, I'll still increase those a little bit.

 

[Mussels]

I put both VDDG to 1.05V. If it doesn't help, I'll still increase those a little bit.

vddg needs to be 500mv below your SoC voltage

So if SoC is 1.15v, those shouldbe 1.10v

Mine auto raises but only to a certain limit, so 1.15v SoC leaves them at 1.0477v


These are my Auto values, SoC was set to 1.175v for testing and the others are auto
This is for 2000MHz IF testing, so these are on the high side

 

[Ruru]

vddg needs to be 500mv below your SoC voltage

So if SoC is 1.15v, those shouldbe 1.10v

Mine auto raises but only to a certain limit, so 1.15v SoC leaves them at 1.0477v

Good to know, I'll boot immediately and fix that.

 

[Mussels]

Good progress here

1. FPS is higher and smoother, 99% has less dips as things happen on screen, overall performance is fairly similar.
2. IMC is definitely better.

3933 rock stable, no WHEA after 2 hours of DRG and an hour of R23

If i cant get 4000, this will absolutely be acceptable. All i had to do was put TRFC back on auto, and i can tune that in again once i know the maximum.

I gotta sleep, but heres the cleaned up (as best i can by end tasking everything) aida64 results

 

[Ruru]

Now sounds like that the crackling is gone, thanks to fine-tuning the SOC voltages. I still need yet to try with the onboard sound later, but with DAC, seems to be fine.

edit: Here's mine with just PBO on and RAM at stock XMP settings (as I was troubleshooting the sound issue)

 

[Dr. Dro]

Spent the day optimizing my rig a little bit. Downgraded BIOS back to one with AGESA 1.2.0.3c to work around the nasty EDC bug. Win11 seems to consistently score lower in Cinebench, though. I've never quite figured out why, though I suspect it has to do with so many background tasks.

Timings:

 

[Toothless]

Can't hit 3600 on the memory since I'm running four dual-rank dimms.

 

[Dr. Dro]

Can't hit 3600 on the memory since I'm running four dual-rank dimms.

So am I, 3600 is doable with my Dominator Platinum kit at 1.375 V. They are some weird downbin of Samsung B-die, at least according to the research I did and Thaiphoon data, but amazing sticks nonetheless. I didn't push them too hard, figured, the performance gains aren't worth the headache of getting it stable, or the extra voltage/heat and strain it brings.

You may have better luck getting them to boot at 3600 by adjusting your tRFC values and increasing DRAM training voltage if possible. The formula for the tRFC adjustment should be:

tRFC2 = tRFC / 1.34
tRFC4 = tRFC2 / 1.625

I've read that the tRFC2 and tRFC4 values don't matter on AM4 platform, I don't know though, I just tweaked them according to that formula anyway, and it worked fine for me. This chart is a quick cheat sheet for converting tRFC clocks into real time latency:

My CB scores are on the same ballpark of yours here. My high score was about 29,3k, though, with a barebones installation of Windows 10.

 

[tabascosauz]

If i cant get 4000, this will absolutely be acceptable. All i had to do was put TRFC back on auto, and i can tune that in again once i know the maximum.

I gotta sleep, but heres the cleaned up (as best i can by end tasking everything) aida64 results

They do say that memory and Fabric hardly matter much for X3D..........but almost 3-digit tRC and 4-digit tRFC are really stretching that statement a bit even for MJR

You should improve quite a bit if you shave off that tRFC and tRC a bit (try somewhere in the 60s for tRC?)

So am I, 3600 is doable with my Dominator Platinum kit at 1.375 V. They are some weird downbin of Samsung B-die, at least according to the research I did and Thaiphoon data, but amazing sticks nonetheless. I didn't push them too hard, figured, the performance gains aren't worth the headache of getting it stable, or the extra voltage/heat and strain it brings.

You may have better luck getting them to boot at 3600 by adjusting your tRFC values and increasing DRAM training voltage if possible. The formula for the tRFC adjustment should be:

tRFC2 = tRFC / 1.34
tRFC4 = tRFC2 / 1.625

I've read that the tRFC2 and tRFC4 values don't matter on AM4 platform, I don't know though, I just tweaked them according to that formula anyway, and it worked fine for me. This chart is a quick cheat sheet for converting tRFC clocks into real time latency:

My CB scores are on the same ballpark of yours here. My high score was about 29,3k, though, with a barebones installation of Windows 10.

The idea for tRFC2 and tRFC4 goes like this:

tRFC2 and tRFC4 are timings that activate when the operating temperature of DRAM hits 85 °C. Below these temperatures, these timings don't do anything.

Whether this is true, can't really say, but I do know:

• OCers don't care about tRFC2 and tRFC4
• Every board vendor sets them differently. Some have tRFC2 high, some roughly follow that formula (no idea where that even came from), some set all 3 the exact same. It doesn't make a difference.
• Appropriately, they don't matter to performance

 

[Dr. Dro]

The idea for tRFC2 and tRFC4 goes like this:

tRFC2 and tRFC4 are timings that activate when the operating temperature of DRAM hits 85 °C. Below these temperatures, these timings don't do anything.

Whether this is true, can't really say, but I do know:

• OCers don't care about tRFC2 and tRFC4
• Every board vendor sets them differently. Some have tRFC2 high, some roughly follow that formula (no idea where that even came from), some set all 3 the exact same. It doesn't make a difference.
• Appropriately, they don't matter to performance

Nice, just learned something new

With the conservative voltage profile, hefty heatsink and open frame case, my RAM won't be really hitting anywhere close to 85, but that is one thing I hope I don't have to test

 

[Mussels]

Can't hit 3600 on the memory since I'm running four dual-rank dimms.

Have you tried more SoC voltage until it all catches on fire?

So am I, 3600 is doable with my Dominator Platinum kit at 1.375 V. They are some weird downbin of Samsung B-die, at least according to the research I did and Thaiphoon data, but amazing sticks nonetheless. I didn't push them too hard, figured, the performance gains aren't worth the headache of getting it stable, or the extra voltage/heat and strain it brings.

You may have better luck getting them to boot at 3600 by adjusting your tRFC values and increasing DRAM training voltage if possible. The formula for the tRFC adjustment should be:

tRFC2 = tRFC / 1.34
tRFC4 = tRFC2 / 1.625

I've read that the tRFC2 and tRFC4 values don't matter on AM4 platform, I don't know though, I just tweaked them according to that formula anyway, and it worked fine for me. This chart is a quick cheat sheet for converting tRFC clocks into real time latency:

View attachment 265222

My CB scores are on the same ballpark of yours here. My high score was about 29,3k, though, with a barebones installation of Windows 10.

That cheat sheet is based on samsung memory iirc - as an example, my 32GB hynix sticks will not go below 704 TRFC at 3600, and they're over 1000 for 3800/3933

Nice, just learned something new

With the conservative voltage profile, hefty heatsink and open frame case, my RAM won't be really hitting anywhere close to 85, but that is one thing I hope I don't have to test

I mean... i've had several systems go unstable if the ram went bast 50C, and one dodgy kit at 40C - so yeah they should never hit that