5800X3D CO and RAM - Thoughts?

[vroktylF]

Just trying to catch the low hanging fruit here. If any of these settings isnt worth the hassle or if some is obviously "wrong", please tell.

MSI B450 GAMING PRO CARBON AC - AGESA ComboAM4v2PI 1.2.0.Cc
5800X3D with Pearless Assassin 120
4x8GB SR Crucial Ballistix DDR4 3600 16-18-18-38
PowerColor RX 7800 XT Hellhound/Red Devil (@1100mV, core@stock, mem@2514, power +15%)

Attaches pictures of
XMP profile 1 stock -
XMP profile 2 stock - (This seems to change SoC and VDDP voltages for some reason)
XMP profile 1 changes - Settings with red numbers is changed, except for tRFC, when set to 560 system didn't boot and needed a cmos reset. Everything else is set to AUTO.
XMP profile 1 tuned - Reported timings from Zentimings
mem.png - Thaiphoon burner reported memstick specs.

Curve optimizer -30 (preferred cores -25)
C-states - Enabled
CPPC - Enabled
CPPC Preferred - Disabled
SVM mode - Disabled

It was a while I looked into this and it seems that later AGESA versions have higher core clocks voltages?

 

[dgianstefani]

It doesn't really matter how you tune it. Before the 9800X3D the X3D chips don't gain much if anything from tuning. Maybe 1-3 % in games at best.

At best you might reduce your power consumption slightly and improve your productivity app performance.

 

[The King]

X3D CPU you can just run XMP and call it a day but you can get some little extra performance with tuned timings.

I will post some quick screenshot of what I got with my C9BLM kits. These may work for you. the best performance for these sticks are with
GDM Disabled CL15. But be aware if you don't know how to OC RAM you can corrupt your windows installation. So please careful make a system restore point before attempting
to run GDM disabled.

C9BLM are basically the best Rev.E ICs you can get it can do well over 4000MT. However you will get WHEA errors in most cases above 1900/1967 FCLK.

If you want to learn how to OC RAM on the 5800X3D then start here.

5800X3D Owners

www.overclock.net

 

[Chrispy_]

Tune RAM for stability with any X3D chip, since RAM speed and latencies have minimal impact on them. DDR4-3600 and 1900 1800 FCLK is very very close to peak performance already.

Yes, there's a tiny bit of performance margin if you push for a memory OC, but honestly unless you're prepared to do all the stability testing that entails, just don't bother. Your time is so much better spent on looking for the optimum offsets for curve optimiser.

X3D memory tuning is definitely not low-hanging fruit. It's the withered last leaf at the very top of the tree and almost not worth reaching for.

edit - whoops, I can't even divide by 2 today :\

 

[Waldorf]

@vroktylF
unless something changed i didnt read about, C states off, and all cppc enabled.

should see if WTRL 16 gives better perf (shouldn't it be 4x WTRS ?)

 

[vroktylF]

Thank you for your answers!

Hmm, CPPC enabled, CPPC preferred cores disabled and C-states enabled is whats seems to be suggested here

What about the XMP profiles? Profile 1 and profile 2 differs. Hard to find any information about the difference but there is some about Profile 1 including basic timings and that the motherboard sets the rest. Profile 2 is supposed to be a full SPD profile including all the timings and nothing left for the board to decide.

I think i've always been running Profile 2 and have not noticed the difference in voltages.

Screenshot from hwinfo after Cinebench R23 been running for about 20 minutes. Its well over a year go since I tested this the last time and im pretty sure the core voltages and the power draw was lower. Any changes in AGESA lately? No problem holding 4.45ghz back then.

This is with CO -30 (-25 preferred cores) and stock XMP profile 1. Maybe its time for a repaste.

 

[freeagent]

Maybe its time for a repaste.

Might be, those temps are brutal for R23, I don't even see those for Linpack, unless it was summer.

Mine is air cooled, no real tune other than my curve and fclk/mclk. Stock power limits, pretty much stock. Dell territory.

 

[vroktylF]

Okey cpu and case fans all at 100% managed to get 15131 points. Case fans usually follows MB System sensor. Works good for gaming with <70C deg temps for both gpu and cpu and is reasonably quiet.
Maybe another fan for push-pull on the cpu would help all core workload.
Think thermal paste is OK after all.

also combined XMP profile 1 and 2. Think thats enough for now, lets see how the stability holds up over time.

 

[Chrispy_]

Dell territory.

I thought "Dell territory" was considerably under stock power limits, artificially crippled by Dell's zero-option BIOS so that they can get away with crappy cooling and the cheapest power delivery that functionally POSTs the CPU.

 

[freeagent]

I thought "Dell territory" was considerably under stock power limits, artificially crippled by Dell's zero-option BIOS so that they can get away with crappy cooling and the cheapest power delivery that functionally POSTs the CPU.

It feels like a Dell compared to my other CPU

But it plays games pretty decently..

 

[vroktylF]

Nothing stupid here, right?

 

[freeagent]

Not at all

 

[Chrispy_]

Looks good to me. 1866 is literally the best AMD-sanctioned 1:1 sweet spot, and CL16 is a decent latency for 3733MT/s.

The only shame is that a 5800X3D doesn't really need any memory tuning. Most tests have shown that the 5800X3D runs damn near the same whether you feed it caviar or tripe.

 

[vroktylF]

Looks good to me. 1866 is literally the best AMD-sanctioned 1:1 sweet spot, and CL16 is a decent latency for 3733MT/s.

The only shame is that a 5800X3D doesn't really need any memory tuning. Most tests have shown that the 5800X3D runs damn near the same whether you feed it caviar or tripe.

Do you mean that it will be any disadvantage of running higher clock than 1866?

 

[freeagent]

If your setup will do it, then yes 1900 is usually faster than 1866.

 

[Chrispy_]

Do you mean that it will be any disadvantage of running higher clock than 1866?

From Tom's:

"However, overclocking the 5800X3D's memory yielded an average performance increase of only about 1%, which isn't too meaningful. In addition to our own expansive testing and experiments, we've seen plenty of benchmarks from multiple sources that indicate that memory overclocking is a fruitless endeavor with the 5800X3D. That's a good thing because you can pair the chip with inexpensive memory and get nearly the absolute best performance available"​

You don't gain any performance by pushing your RAM with an x3D chip. Your kit's default 3600 CL16 already got you that 1% gain that Tom's talked about. Reddit posts of people benchmarking have shown that a 5800X3D doesn't really slow down much if you give it DDR4-2133, and you're losing almost nothing at all even if you give it a slow DDR4-3000 kit with slow timings.

There are only two things you get from pushing a memory overclock on an X3D chip:

1. You do get 20-30% higher idle power on a CPU that already has relatively high idle power consumption.
2. You do run a higher risk of memory errors/instability that will corrupt your OS install, so you also need to be very rigorous with your stability testing.

Personally, I'd have enabled the 3600 CL16 XMP and forgotten about it. The fact you have 3733 CL16 isn't a problem, as long as it's still 100% stable. Pushing higher only risks stability, it won't get you more performance.

 

[Veseleil]

There are already a lot of older threads that have answers to your questions including topics like:
Zen 3 RAM tuning,
Specific RAM modules tuning (Micron, Samsung),
5800X3D RAM tuning,
5800X3D CO tuning,
And overall Zen 3 everything tuning.

You would be amazed how much of a relevant info can come up from a forum search function.

Don't get me wrong, but we already spent a lot of our energy, shared a lot of detailed info of our testing, and mostly ended up with a quite decent amount of solutions, years ago.
RAM tweaking on a Zen can be a very tiresome and frustrating process, Ryzen IMC quality varies from chip to chip, and add to that even more variables, like different RAM kits, SR, DR modules, number, capacity of modules, different motherboard OC features and capabilities.
It might be better for you to do some research, as many people won't jump in to help with this, like it was the case couple of years ago. Not because we are lazy or assholes, but because it was a while already, and we might forgot a lot of details on the way, and don't have time nor patience to go through all of it once again.

 

[INSTG8R]

Looks good to me. 1866 is literally the best AMD-sanctioned 1:1 sweet spot, and CL16 is a decent latency for 3733MT/s.

The only shame is that a 5800X3D doesn't really need any memory tuning. Most tests have shown that the 5800X3D runs damn near the same whether you feed it caviar or tripe.

Yep I came from a 5600X with my 3600 RAM running at 3800/1800 IF and it did actually make a difference in Synth benches. Using the same settings on my 5800X3D amounted to nothing tangible to making the effort worthwhile without going down the "subtimings Rabbit Hole" that I didn't even bother on my 5600X so not even worth that with the 5800X3D

 

[Chrispy_]

Yeah, on Zen3 non-X3D, there's a decent risk-reward ratio because there's very real improvement from increasing your RAM and FCLK speeds, but the risk of RAM instability is quite high because it will easily corrupt your OS in the background. It's not like a CPU overclock that instantly crashes your PC and forces a reboot - unstable RAM will replace system files during updates and game save data with corrupt garbage that you may not even spot until your PC either refuses to boot or starts to exhibit ultra weird behaviour.

So for a regular Zen3 you want to get the RAM tuning as agressive as possible for maximum performance, and then dial it back a few notches to hopefully get 100% stability, then test the ever-living shit out of it to make sure that you're not going to screw up your OS or your data.

Since an X3D chip doesn't even need fast RAM, why even do all that work and take that risk for almost no benefit whatsoever?

 

[Veseleil]

Yeah, on Zen3 non-X3D, there's a decent risk-reward ratio because there's very real improvement from increasing your RAM and FCLK speeds, but the risk of RAM instability is quite high because it will easily corrupt your OS in the background. It's not like a CPU overclock that instantly crashes your PC and forces a reboot - unstable RAM will replace system files during updates and game save data with corrupt garbage that you may not even spot until your PC either refuses to boot or starts to exhibit ultra weird behaviour.

So for a regular Zen3 you want to get the RAM tuning as agressive as possible for maximum performance, and then dial it back a few notches to hopefully get 100% stability, then test the ever-living shit out of it to make sure that you're not going to screw up your OS or your data.

Since an X3D chip doesn't even need fast RAM, why even do all that work and take that risk for almost no benefit whatsoever?

Some of us just can't but try everything chasing perfection when our field of enthusiasm is involved. It doesn't matter if it's needed, relevant or not, it's the habit from the old times. Those 0.1% improvements makes our heart skipping beats. It's like gambling but without cons.

 

[vroktylF]

There are already a lot of older threads that have answers to your questions including topics like:
Zen 3 RAM tuning,
Specific RAM modules tuning (Micron, Samsung),
5800X3D RAM tuning,
5800X3D CO tuning,
And overall Zen 3 everything tuning.

You would be amazed how much of a relevant info can come up from a forum search function.

Don't get me wrong, but we already spent a lot of our energy, shared a lot of detailed info of our testing, and mostly ended up with a quite decent amount of solutions, years ago.
RAM tweaking on a Zen can be a very tiresome and frustrating process, Ryzen IMC quality varies from chip to chip, and add to that even more variables, like different RAM kits, SR, DR modules, number, capacity of modules, different motherboard OC features and capabilities.
It might be better for you to do some research, as many people won't jump in to help with this, like it was the case couple of years ago. Not because we are lazy or assholes, but because it was a while already, and we might forgot a lot of details on the way, and don't have time nor patience to go through all of it once again.

Yes I've spent hours, days and weeks searching and reading through several forums including techpowerup and overclockers.net since I bought my 5800X3D over a year ago.

If it was for the performance only it would have been smarter to just buy something faster instead

With the settings below I havent noticed anything while gaming or doing everyday tasks.

TestMem5 reports no errors after 3-5 cycles of 1usmus_v3.cfg and Extreme1@anta777.cfg

Highest 3DMark Timespy physics score improves with higher clock and reduced timings but I've noticed more inconsistent scores. Both for graphics and physics and in my opinion the variance is too big to be just run-to-run difference.

Only game i play is Escape from Tarkov and its kind of impossible to benchmark or draw a conclusion even on larger improvements because of how the game behaves

 

[Veseleil]

Seems that your IMC is better than mine. Or your motherboard handles RAM OC better. Or both.
I could run 3800 easily with 2 modules before, even on my r5 3600. But since I went with 4 modules, not anymore.
I've settled for these values a while ago:

 

[vroktylF]

Seems that your IMC is better than mine. Or your motherboard handles RAM OC better. Or both.
I could run 3800 easily with 2 modules before, even on my r5 3600. But since I went with 4 modules, not anymore.
I've settled for these values a while ago:

View attachment 386473

I did dial memclk down a bit to 1867. I may have been a coincidence but the benchmark scores is more consistent now. 3dmark shows no lower score with 1867 than 1900.

You have lower tRFC than me. I tried once with something around 300ns but the system would not boot. Maybe thats due to too low voltage and that ohmic settings could have stabilized it.
Won't dig into this anymore now.

 

[Veseleil]

I did dial memclk down a bit to 1867. I may have been a coincidence but the benchmark scores is more consistent now. 3dmark shows no lower score with 1867 than 1900.

You have lower tRFC than me. I tried once with something around 300ns but the system would not boot. Maybe thats due to too low voltage and that ohmic settings could have stabilized it.
Won't dig into this anymore now.

Believe me when I say I've tried a lot of values. All of the main, secondary and tertiary timings, and it made me sick. Because it's not just about question will it boot, could I screw up my OS, but stress testing with tm5 and other tools takes time, one changed value at a time, each damn time. I had enough for a decade at least.

I just hope the Zen 6 will have a much better IMC. When I eventually upgrade to the appropriate X3D counterpart, and I'm about to endure similar kind of suffering with RAM tweaking, it will be more rewarding than this was.