RAM Overclocking

[Waifu_Lewds]

Hello all,

Long time listener, first time caller. I recently have been messing around with overclocking due to graphics card shortage and I find it to be a fun hobby. I've had good luck with both my processor and gpu overclocks, but RAM was a different story.

 

Went from XMP 3200 to 3266 with no problem, but crashed the system at 3333. I reset to 3266 and lowered the timings successfully for a bench. I went up to 3333 again but with lowered timings and crashed again. I set everything back to XMP, benched successfully,  then moved back up to 3266 and then 3333. I thought I fried the motherboard. After hours of troubleshooting I found my bios was corrupted and everything was normal after a flash.

 

I wanted to learn how to overclock RAM after seeing abnormally high latency during benchmark. My question is two parts:

1. How to overclock past 3266 without crashing the system (I imagine this is a voltage issue).

2. How to reduce RAM latency (I'm at 61.0-62.0ns).

 

Ive watched probably about 6 hours of videos and spent the same amount of time readings forums with little success.

 

Specs

MOTHERBOARD: MSI B450 Tomahawk Max

CPU: Ryzen 7 5800 X @4.6ghz all core AMD autoclocked voltage

RAM: Corsair Vengeance LPX 3200 CL16 (4x8gb)

GPU: Asus Rog Strix 5600 XT

PS: Corsair 750W gold

 

Please feel free to refer me to threads with similar specs. 

 

Edit: I use my computer for gaming and statistical analysis.

[Revioli]

A 4 stick RAM config is gonna be hard on the CPU's memory controller, so it's much harder to get a stable oc than with 2. On top of that, if even one of those sticks aren't stable at that speed, then even if all the other's are fine, it still won't work. You can try 1usmus's DRAM Calculator for Ryzen, and see if the suggested settings are stable.

 

But taking a step back, memory latency is very unlikely to be a bottleneck in your system, so if it's for improved performance, it probably doesn't matter.

[Waifu_Lewds]

On 1/3/2022 at 12:54 AM, Revioli said:

A 4 stick RAM config is gonna be hard on the CPU's memory controller, so it's much harder to get a stable oc than with 2. On top of that, if even one of those sticks aren't stable at that speed, then even if all the other's are fine, it still won't work. You can try 1usmus's DRAM Calculator for Ryzen, and see if the suggested settings are stable.

 

But taking a step back, memory latency is very unlikely to be a bottleneck in your system, so if it's for improved performance, it probably doesn't matter.

Thanks for the suggestions and insight.

[freeagent]

Running four sticks is not a problem. It’s only a problem if you are pushing for top clocks.

 

Corsair bins their chips pretty hard, chances are you aren’t going to be able to get too far with them.

[Mark Kaine]

please if you're overclocking make comparison tests, stock / standard XMP vs all your overclocks... in my experience gains are minimal going from 2933mhz up to 3600mhz,  the thing that changes most is latency,  and certainly not in a good way the "faster" your ram goes.

 

no im not saying its impossible to gain performance,  im saying you should fact check this each step... (regression is a thing as well)

 

5 hours ago, freeagent said:

Running four sticks is not a problem. It’s only a problem if you are pushing for top clocks.

 

Corsair bins their chips pretty hard, chances are you aren’t going to be able to get too far with them.

true, my corsair ram wouldn't oc one bit above the speed that's on the sticker lol... cant complain i guess, still disappointing,  because this ram was and still is recommended on internet message boards by so called self-proclaimed "experts" (and wasn't cheap on top of it lol, probably unsurprisingly...)

 

On 1/3/2022 at 8:37 AM, Waifu_Lewds said:

2. How to reduce RAM latency (I'm at 61.0-62.0ns).

thats definitely a good question... @freeagentshould know? (probably more voltage ☠)

 

well, as said, its likely the higher your frequency the higher the latency will get, i guess it also depends on what you're using to measure this...

 

 

but my cpu is doing really well, 98th percentile holy moly, not that I'd notice much 

 

 

5 hours ago, freeagent said:

It’s only a problem if you are pushing for top clocks.

even 2 sticks too much,  apparently 

 

 

[freeagent]

2 hours ago, Mark Kaine said:

please if you're overclocking make comparison tests, stock / standard XMP vs all your overclocks...

Here is just a quickie. I kept the CPU at 4600MHz, all sub timings are identical, at least the ones that I can see and control. I don't use XMP, its profile set is too loose.. I use that profile on my 5600X to run them at 4000 and up. So really all I did was change the fclk and mclk frequency. I didn't want the CPU to boost because it is a large variance between 2800MHz and 5150MHz to skew some results 

 

 

And my 5600X:

 

 

And 1 last one of my 5900X

 

 

[Mark Kaine]

28 minutes ago, freeagent said:

Here is just a quickie. I kept the CPU at 4600MHz, all sub timings are identical, at least the ones that I can see and control. I don't use XMP, its profile set is too loose.. I use that profile on my 5600X to run them at 4000 and up. So really all I did was change the fclk and mclk frequency. I didn't want the CPU to boost

yeah that makes sense (dude in the video I posted also turned off cpu boost... probably for other reasons but still)

And i see, well your latency got lower but you also kept the timings (largely)  that's interesting but also kinda my point most ram doesn’t do that, or at least it would be difficult,  and once you loosen timings you add latency (to my understanding?) 

 

For example when i did my ram oc i wouldn't get it to boot at 3600 c14, had to set it to c16 and tbh i didn't test for latency but i also didn't really see any notable differences in what i actually tested (cinebench 100 points more , maybe, games would do 1-2 fps more, if that,  etc) still i kinda expect latency to go up, as you can see in that ub test its not great at least...

prob needs more voltage i might get it to 3600 c14... but, i was just happy it boots and well my mobo is still dodgy and generally doesn't like change so just left it at that, at 1.36v.

 

 

ps: my ub results definitely went up, cpu specifically, with faster ram... too bad ub is unreliable trash,  according to the internet. 

 

[Waifu_Lewds]

On 1/4/2022 at 5:23 PM, Mark Kaine said:

please if you're overclocking make comparison tests, stock / standard XMP vs all your overclocks... in my experience gains are minimal going from 2933mhz up to 3600mhz,  the thing that changes most is latency,  and certainly not in a good way the "faster" your ram goes.

 

no im not saying its impossible to gain performance,  im saying you should fact check this each step... (regression is a thing as well)

 

true, my corsair ram wouldn't oc one bit above the speed that's on the sticker lol... cant complain i guess, still disappointing,  because this ram was and still is recommended on internet message boards by so called self-proclaimed "experts" (and wasn't cheap on top of it lol, probably unsurprisingly...)

 

thats definitely a good question... @freeagentshould know? (probably more voltage ☠)

 

well, as said, its likely the higher your frequency the higher the latency will get, i guess it also depends on what you're using to measure this...

 

 

but my cpu is doing really well, 98th percentile holy moly, not that I'd notice much 

 

 

even 2 sticks too much,  apparently 

 

 

I finally learned how to calculate real world latency (1 / GhzFrequency x Cas Latency = Actual Latency). This has helped understand how to evaluate different types of ram and overclocking potentials.