I do mostly productivity tasks around CG and 3D modeling apps and casual gaming (Shadow of Tomb Raider).
I currently have 3000Mhz memory that runs at advertised speed via XMP profile.
All is fine, runs perfectly stable.
However, I wonder how much tangible benefit would I get if I move to faster RAM. I am currently on X570 board and plan to upgrade to 3900x sometime in near future.
I get it that benchmarks would improve and I could imagine getting some more FPS but quite frankly neither of those mean much to me. Would it really matter like in rendering ambient occlusion pass in Substance Painter, decimating mesh in ZBrush, or opening huge Maya file?
(I got a really sweet deal in the 32GB Sniper X kit, it’s about 150 dollars and this is by far the cheapest deal on 3000Mhz DDR4 memory here in Japan, memory as all parts are always so damn expensive here. If the fact would be that having more Mhz has limited potential to improve my workflow, I will happily get another kit to get my ram to 64)
IMHO, memory speed won’t make a substantial difference. From benchmarks I’ve seen, you’ll get a few percent increase. But nothing that would make me go “wow i got to buy a faster kit and try to sell my old set”.
Are you sure you are memory speed limited in Substance Painter, ZBrush or Maya?
you may want to check disc activity vs memory consumption vs cpu utilization to see if there are any other bottlenecks
In Substance Painter I am pretty sure I am vram limited, and ram limited, likely so on ZBrush too.
Some operations in Maya or Zbrush just take time and during those operations CPU usage is never anywhere near 100%.
Watching Youtube everyone seem say that memory speed makes large difference in Ryzen but I am yet to see any benefit of faster ram in practice when doing actual work. I am not even sure if I would notice the difference of faster speed vs running at 2333Mhz…
I can confirm, stability is more important than timings with the 2700x. I have some Teams 3200 C14 ram. Even with tighter timings suggested by the Ryzen DRAM Calculator the difference was nominal. Actually, couldn’t get the XMP C14 to run stable on an X470 Gaming 5 and stuck with the auto detected C15 timings, manual 3200Mhz with 1.35v/0.675v termination. All the benches I ran were within margin of error for C14 to C16 latencies.
If you currently have 3000mhz stable, i would try upping the frequency a bit more until it gets unstable, and then give it a tad more voltage! if you get it stable at a good speed, lets say 3200mhz ish, and then start tightening your timings from there.
Since you already have 3000mhz running smoothly, a upgrade would be more like a “sidegrade”, so overclock what you have, basically free performance!
And yes, ram speed and timings DO matter on Ryzen, so the more you can get out of your current kit, the better!
It can make a big difference for a lot of things, but once you are past 3000mhz and 14-16 primary timings, any gains quickly become academic and not worth the time unless you have a kit that you can REALLY pump up.
For anyone sitting below 3000, then it’s generally worth giving it a go.
Going from 3000 to 3200 = bleh
Going from 3000 to 3600 = probably be worth it if timings don’t go to shit.
If I was you, I’d play around a bit to see if there was some quick easy gains. If not, I’d leave it be and go back to not worrying about it. I have 2400mhz ECC ram sitting nicely at 3000mhz 14-15-15-16 (very little time invested this go around, thanks to past efforts) thanks to the latest and greatest bios for the taichi x399. This is a overclock that’s totally worth it that shows in benchmarks and even games. Of course, early threadripper systems ESPECIALLY like fast ram.
However thanks to the power of error correction, I can actually run a bit higher to 3200mhz, correcting errors as I go and you know what I gain? Fuckall that’s what. I could probably stabilize 3066 but why bother.
Basically, beyond some curiosity, don’t burn more time/value than you are going to gain back from overclocking ram.
Thanks for your replies, I really appreciate it.
It seems that my memory seems to be able to do 3200hz without having to touch the voltage. So far running Prime95 memory torture test, seems no problem.
At present it’s CL16-18-18-38. Should I try lower these values?
My question is, does voltage increase help to achieve tighter timings? Then for G.Skill Sniper X, what would max safe voltage?
It depends, some timings scale with voltage and other don’t, but it depends on the type of RAM chips you have.
This is the guide I used and it’s pretty helpful.
I also used the Karhu RAM Test tool that the guide mentioned and it’s definitely worth the money as it saves so much time over other tools. I would get instability in the first couple of minutes of testing most of the time using this tool.
I assume you are running “XMP” profile now? so the ram already has some voltage put into it over stock, i wouldnt go past 1.35v for a 24\7 stable usecase.
Timings is a wierd one, some scale really good with voltage, and some doesnt, but if i were you, i would be happy with 3200mhz at those timings, and just enjoy the stability and extra 200mhz you got for free