Is it worth it staying at DDR4-3733 OC + how to check for micro stutters? (AMD 5900x)

[rickje139]

I currently have my ram overclocked to 3733 14-15-15-15 and was wondering if there is some tool i can use to check if its fully stable or even worth it using this in the long term.

I had to add +0.03V to the dimm and +0.06 to the SOC to make this run without errors according to my tests.

 

Tests i have used:

DRAM calculator membench (15 minutes)

Prime 95 (5 minutes)

AIDA64 (2 minutes)

 

I dont know any tool that can check for micro stutters or whether its even worth it staying at this frequency.

 

Do you guys have any recommendations? Should i stay at 3733 or go back to 3600 with lower latencies instead?

 

How can i find out whether my system is unstable or has micro stutters or big framerate drops?

 

 

I could run the 3600 OC at 1.420V stable, so i can go back to there if the 1.48V is too dangerous.

[Fasauceome]

If your performance is satisfactory and your games don't crash, you have achieved a stable overclock in my opinion. 

 

If you're going for the numbers alone then run benchmark suites such as unigine superposition to see if you're breaking any records.

[Hairless Monkey Boy]

15 minutes ago, rickje139 said:

Should i stay at 3733 or go back to 3600 with lower latencies instead?

What were your latencies at 3600? You should always prioritize latency over bandwidth. You can use AIDA64 to test both.

 

Many people don't realize that the timings are relative to the frequency. Forgive me if you know this, but from a latency perspective 3200Mhz CL16 is equivalent to 3600Mhz CL18.

15 minutes ago, rickje139 said:

I could run the 3600 OC at 1.420V stable, so i can go back to there if the 1.48V is too dangerous.

1.48V is not dangerous. 1.5V is about as high as you should go for daily, though. It's the highest XMP voltage. I daily 1.5V.

[rickje139]

3 minutes ago, HairlessMonkeyBoy said:

What were your latencies at 3600? You should always prioritize latency over bandwidth. You can use AIDA64 to test both.

 

Many people don't realize that the timings are relative to the frequency. Forgive me if you know this, but from a latency perspective 3200Mhz CL16 is equivalent to 3600Mhz CL18.

1.48V is not dangerous. 1.5V is about as high as you should go for daily, though. It's the highest XMP voltage. I daily 1.5V.

I had a random latency of 61.5 - 62.7 at 3600 in aida64 and a random latency of 60.5 - 61.5 at 3733,

however though, i have really loose timings now at 3733 for quick testing so i can increase them even further if i'd like.

 

Is there something i should check for because i have a feeling i am experiencing micro stutters ingame or random framerate drops.

[Hairless Monkey Boy]

4 minutes ago, rickje139 said:

I had a random latency of 61.5 - 62.7 at 3600 in aida64 and a random latency of 60.5 - 61.5 at 3733,

however though, i have really loose timings now at 3733 for quick testing so i can increase them even further if i'd like.

 

Is there something i should check for because i have a feeling i am experiencing micro stutters ingame or random framerate drops.

Good. 3733 it is then.

 

For testing, make a memtest86 bootable USB and let that run through 4 passes of every test or more. That will take quite a long time, but won't find everything because it isn't quite hard enough on the IMC. For that, if you really want to be sure you are stable, you can use a custom Prime95 run testing FFT size 512-4096 on all of your threads, 15 minutes per FFT size, and 75% of your RAM in "memory to use". Or, just keep an eye on the WHEA count in HWInfo64 or in the Event Viewer. I find Cinebench likes to crash with unstable RAM, so you can try looping that, too.

 

If you aren't crashing in games, and you don't have any WHEA events, then you are basically good to go. I've never heard of RAM errors causing stuttering, at least not without other obvious issues in addition.

[rickje139]

2 minutes ago, HairlessMonkeyBoy said:

Good. 3733 it is then.

 

For testing, make a memtest86 bootable USB and let that run through 4 passes of every test or more. That will take quite a long time, but won't find everything because it isn't quite hard enough on the IMC. For that, if you really want to be sure you are stable, you can use a custom Prime95 run testing FFT size 512-4096 on all of your threads, 15 minutes per FFT size, and 75% of your RAM in "memory to use". Or, just keep an eye on the WHEA count in HWInfo64 or in the Event Viewer. I find Cinebench likes to crash with unstable RAM, so you can try looping that, too.

 

If you aren't crashing in games, and you don't have any WHEA events, then you are basically good to go. I've never heard of RAM errors causing stuttering, at least not without other obvious issues in addition.

I wasn't thinking of ram errors causing fps drops or stuttering, i had heard somewhere that going above 3600 requires a ton of research and causes micro stutters, this was especially visible for 3800 i had heard.

[rickje139]

25 minutes ago, HairlessMonkeyBoy said:

What were your latencies at 3600? You should always prioritize latency over bandwidth. You can use AIDA64 to test both.

 

Many people don't realize that the timings are relative to the frequency. Forgive me if you know this, but from a latency perspective 3200Mhz CL16 is equivalent to 3600Mhz CL18.

1.48V is not dangerous. 1.5V is about as high as you should go for daily, though. It's the highest XMP voltage. I daily 1.5V.

This is what the DRAM calculator recommended me.

 

It seems to be stable as of now, do you recommend me to change any of the timings a bit further down for extra performance?

 

I had quickly set these timings so i dont know which ones will cause trouble.

[Hairless Monkey Boy]

2 minutes ago, rickje139 said:

I wasn't thinking of ram errors causing fps drops or stuttering, i had heard somewhere that going above 3600 requires a ton of research and causes micro stutters, this was especially visible for 3800 i had heard.

I see. I'm not familiar with that phenomena. It's sounds like it may have come from someone who has no idea what they are talking about .

 

As long as your MCLK, FCLK, and UCLK are synchronized, which yours are, then you should be fine. But if you are noticing stuttering above 3600Mhz, and it isn't placebo, then you can try increasing SOC voltage up to at most 1.2V, or lowering your RAM back to 3600Mhz.

 

In the end it makes little difference to performance, but I'm sure you know that already.

[rickje139]

1 minute ago, HairlessMonkeyBoy said:

I see. I'm not familiar with that phenomena. It's sounds like it may have come from someone who has no idea what they are talking about .

 

As long as your MCLK, FCLK, and UCLK are synchronized, which yours are, then you should be fine. But if you are noticing stuttering above 3600Mhz, and it isn't placebo, then you can try increasing SOC voltage up to at most 1.2V, or lowering your RAM back to 3600Mhz.

 

In the end it makes little difference to performance, but I'm sure you know that already.

I am always worried about voltages, what is the max recommended voltage for SOC for daily running? I am currently running it at 1.09.

 

 

 

 

[Hairless Monkey Boy]

1 minute ago, rickje139 said:

I am always worried about voltages, what is the max recommended voltage for SOC for daily running? I am currently running it at 1.09.

 

 

 

 

1.2V.

 

You get less worried about it as you build experience. 

 

16 minutes ago, rickje139 said:

do you recommend me to change any of the timings a bit further down for extra performance?

I'm not familiar enough with the Ryzen IMC to make specific recommendations. I'm much more familiar with the behavior of the Intel IMC.

 

But with that caveat out of the way....

 

After the Primary timings, Tref (the time interval between refreshes) and Trfc (the amount of time it takes to refresh) make the biggest difference to performance. You can't change Tref with Ryzen afaik. So, with my B-die kit, which seems to be a lesser bin than yours, I'm able to set Trfc to 267 at 3600Mhz, and 296 at 4000Mhz. A little math and a conservative rounding up and at 3733Mhz, it should be able to do 280.

 

But I don't know if the Ryzen IMC will affect this.

[rickje139]

11 minutes ago, HairlessMonkeyBoy said:

1.2V.

 

You get less worried about it as you build experience. 

 

I'm not familiar enough with the Ryzen IMC to make specific recommendations. I'm much more familiar with the behavior of the Intel IMC.

 

But with that caveat out of the way....

 

After the Primary timings, Tref (the time interval between refreshes) and Trfc (the amount of time it takes to refresh) make the biggest difference to performance. You can't change Tref with Ryzen afaik. So, with my B-die kit, which seems to be a lesser bin than yours, I'm able to set Trfc to 267 at 3600Mhz, and 296 at 4000Mhz. A little math and a conservative rounding up and at 3733Mhz, it should be able to do 280.

 

But I don't know if the Ryzen IMC will affect this.

Also i just noticed something.

At 3600, the custom latency was 2.1.

On 3733, the custom latency is 3.1, but the random latency is way lower.

 

I have no idea what the custom latency is for.

[rickje139]

45 minutes ago, HairlessMonkeyBoy said:

1.2V.

 

You get less worried about it as you build experience. 

 

I'm not familiar enough with the Ryzen IMC to make specific recommendations. I'm much more familiar with the behavior of the Intel IMC.

 

But with that caveat out of the way....

 

After the Primary timings, Tref (the time interval between refreshes) and Trfc (the amount of time it takes to refresh) make the biggest difference to performance. You can't change Tref with Ryzen afaik. So, with my B-die kit, which seems to be a lesser bin than yours, I'm able to set Trfc to 267 at 3600Mhz, and 296 at 4000Mhz. A little math and a conservative rounding up and at 3733Mhz, it should be able to do 280.

 

But I don't know if the Ryzen IMC will affect this.

Do you know what i can do to fix broken programs?

Nahamic seems to have broken for me and steam forgot about some games being installed (probably due to the ram being unstable while it was running in the background.

 

I had heard of a command that fixes windows after messing with ram timings, but i forgot what it was.

 

I ended up with these timings:

i still need to stress test for a whole day straight though before i actually know its 100% stable, but for now it seems to be holding well and finished an entire prime 95 test.

[freeagent]

Check out TestMem 5 and run the Anta777 Extreme config. Default is 3 cycles you can change  the cfg file to set additional parameters.

[Columbo]

Whenever you run into micro stuttering during gaming. That is unstable ram. I have experienced it myself. I think it has more to do with the memory controller than the memory. I have asked many Ryzen OC'ers what memory speeds they run their memory at. For some of the (different forum) experts, it's a closely guarded secret. They do not like to admit fancy ram or motherboards have micro stutters or fails. I assumed that Zen 3 CPU's and X570 motherboards all ran @ either 3733 or 3800mhz or more. 

 

I have a Zen 2 CPU. My memory will boot @ 3800mhz and run but usually results in micro stuttering or phantom crashes within 30 minutes. 3733mhz never crashes but the micro stuttering is a serious problem. The memory controller on the Zen3 chips are the same as the Zen 2 chips. It's not that the ram cannot run perfectly stable at those speeds. It's the memory controller that cannot handle speeds of 3733mhz or greater with 100% stability. 

 

 

[freeagent]

3 hours ago, Columbo said:

Whenever you run into micro stuttering during gaming. That is unstable ram. I have experienced it myself. I think it has more to do with the memory controller than the memory. I have asked many Ryzen OC'ers what memory speeds they run their memory at. For some of the (different forum) experts, it's a closely guarded secret. They do not like to admit fancy ram or motherboards have micro stutters or fails. I assumed that Zen 3 CPU's and X570 motherboards all ran @ either 3733 or 3800mhz or more. 

 

I have a Zen 2 CPU. My memory will boot @ 3800mhz and run but usually results in micro stuttering or phantom crashes within 30 minutes. 3733mhz never crashes but the micro stuttering is a serious problem. The memory controller on the Zen3 chips are the same as the Zen 2 chips. It's not that the ram cannot run perfectly stable at those speeds. It's the memory controller that cannot handle speeds of 3733mhz or greater with 100% stability. 

 

 

I can run my ram @ 4000 1:1 though.. and a lot of other guys can too.. My SLI setups used to stutter at times. It would be just easier to say if your system isn't stable there is a good chance she will stutta like a muthaf.. you get it. Your system doesn't even have to be overclocked to be unstable.. If there is a compatibility issue somewhere between the uefi and ram, or where ever.. it will show up in your game and daily computing life. Your pc goes from being a calculator to a random number generator.

[Columbo]

1 hour ago, freeagent said:

I can run my ram @ 4000 1:1 though.. and a lot of other guys can too.. My SLI setups used to stutter at times. It would be just easier to say if your system isn't stable there is a good chance she will stutta like a muthaf.. you get it. Your system doesn't even have to be overclocked to be unstable.. If there is a compatibility issue somewhere between the uefi and ram, or where ever.. it will show up in your game and daily computing life. Your pc goes from being a calculator to a random number generator.

I call it the coffins corner of memory. Over time the instability is not like a blue screen. The ghost in the machine effect. I assumed that all the X570 board users could do 4000mhz on the Zen3. It turns out they can but stability issues happen over time without hard crashes. 

[rickje139]

6 hours ago, Columbo said:

Whenever you run into micro stuttering during gaming. That is unstable ram. I have experienced it myself. I think it has more to do with the memory controller than the memory. I have asked many Ryzen OC'ers what memory speeds they run their memory at. For some of the (different forum) experts, it's a closely guarded secret. They do not like to admit fancy ram or motherboards have micro stutters or fails. I assumed that Zen 3 CPU's and X570 motherboards all ran @ either 3733 or 3800mhz or more. 

 

I have a Zen 2 CPU. My memory will boot @ 3800mhz and run but usually results in micro stuttering or phantom crashes within 30 minutes. 3733mhz never crashes but the micro stuttering is a serious problem. The memory controller on the Zen3 chips are the same as the Zen 2 chips. It's not that the ram cannot run perfectly stable at those speeds. It's the memory controller that cannot handle speeds of 3733mhz or greater with 100% stability. 

 

 

Is there a way i can see the 1% lows or check for micro stutters somewhere?

I dont really know any game or tool that can show me micro stutters.

[freeagent]

2 hours ago, Columbo said:

I call it the coffins corner of memory. Over time the instability is not like a blue screen. The ghost in the machine effect. 

OH yeah, I get what your saying now. I have  an old system that is like that but is not in use. Probably just the CMOS battery

 

Edit:

 

New X570 S boards should be dropping now, they will have improved mem topology for better clocks. Should be able to do 5k DDR now if your sticks are rippers and your CPU can do it..