[SOLVED] RAM underclock with tighter timings or stock?

I have a Ryzen 5 3600, B550 Aorus Pro AC.

I am trying to squeeze the most performance out of my system, just OC'd my 2070 Super a little, +250mhz memory, +15mhz core clock. (I'm scared, never even had a pc before lol)

I read somewhere that the ultimate criteria for RAM speed is the absolute latency ie 3200C14 > 3600C18

I have a Ripjaws V 2X8GB kit rated at 3000Mhz, 16-18-18-38. Should I underclock to 2666Mhz and tighten the timings? If yes what should I keep the timings? How much performance increase will this give me?

Also how do I even do this? I can probably just watch a video on this one but still anything important?
 
Solution
If you reduce the frequency (Clock speed) then you are USUALLY going to undo any gains you might get from tightening the timings. Gains come from tightening the timings but NOT having to reduce the clock speed. There are indeed some cases where a very low latency reduces the true latency to the point where the memory is faster than another kit with a higher speed, but not a higher speed with the SAME latency.

So yes, 3200mhz CL14 is MINIMALLY faster than 3600mhz CL16, but it's not ENOUGH faster that you'd want to pay a lot more for one over the other.

And MORE important is the fact that the second you begin messing around changing ANYTHING in regard to deviations from the default configuration or the XMP profile, you MUST then put in...
If you reduce the frequency (Clock speed) then you are USUALLY going to undo any gains you might get from tightening the timings. Gains come from tightening the timings but NOT having to reduce the clock speed. There are indeed some cases where a very low latency reduces the true latency to the point where the memory is faster than another kit with a higher speed, but not a higher speed with the SAME latency.

So yes, 3200mhz CL14 is MINIMALLY faster than 3600mhz CL16, but it's not ENOUGH faster that you'd want to pay a lot more for one over the other.

And MORE important is the fact that the second you begin messing around changing ANYTHING in regard to deviations from the default configuration or the XMP profile, you MUST then put in the work to validate that those changes have not induced any instability, because memory instability is VERY serious. not a joke. It will corrupt your entire data structure. OS, personal files, settings, configuration, registry, movies, music, games, documents, everything, sooner or later. Usually sooner.

My advice, the gains are generally not worth the amount of effort you have to put in to get the gains when it comes to making major changes to your memory configuration. If anything I'd probably say that maybe you leave the timings at their advertised values and see if you can push the memory to 3200mhz instead. You are likely to gain more from doing that than you are by trying to drop the latency down to CL14, which honestly probably won't work anyhow because if the configuration on those sticks was capable of a CL14 configuration then G.Skill would have done it themselves and charged more for the kit like they do with all their CL14 3000-3200mhz kits. Even so, if you DO increase the memory speed you still need to do the same testing to establish stability.

That testing consists primarily of four FULL passes of Memtest86 followed by a rather extensive run of a custom Prime95 configuration and then additionally it would be recommended that you can pass an 8 hour run of Realbench with half your memory capacity selected on the Realbench stress test.

Full instructions can be found here for testing:



And about the best information you will find on making adjustments to memory speed and timings, that I've seen, can be found here:

https://www.overclock.net/threads/comprehensive-memory-overclocking-guide.1630388/#post-26096937
 
  • Like
Reactions: Prad_Bitt
Solution
If you reduce the frequency (Clock speed) then you are USUALLY going to undo any gains you might get from tightening the timings. Gains come from tightening the timings but NOT having to reduce the clock speed. There are indeed some cases where a very low latency reduces the true latency to the point where the memory is faster than another kit with a higher speed, but not a higher speed with the SAME latency.

So yes, 3200mhz CL14 is MINIMALLY faster than 3600mhz CL16, but it's not ENOUGH faster that you'd want to pay a lot more for one over the other.

And MORE important is the fact that the second you begin messing around changing ANYTHING in regard to deviations from the default configuration or the XMP profile, you MUST then put in the work to validate that those changes have not induced any instability, because memory instability is VERY serious. not a joke. It will corrupt your entire data structure. OS, personal files, settings, configuration, registry, movies, music, games, documents, everything, sooner or later. Usually sooner.

My advice, the gains are generally not worth the amount of effort you have to put in to get the gains when it comes to making major changes to your memory configuration. If anything I'd probably say that maybe you leave the timings at their advertised values and see if you can push the memory to 3200mhz instead. You are likely to gain more from doing that than you are by trying to drop the latency down to CL14, which honestly probably won't work anyhow because if the configuration on those sticks was capable of a CL14 configuration then G.Skill would have done it themselves and charged more for the kit like they do with all their CL14 3000-3200mhz kits. Even so, if you DO increase the memory speed you still need to do the same testing to establish stability.

That testing consists primarily of four FULL passes of Memtest86 followed by a rather extensive run of a custom Prime95 configuration and then additionally it would be recommended that you can pass an 8 hour run of Realbench with half your memory capacity selected on the Realbench stress test.

Full instructions can be found here for testing:



And about the best information you will find on making adjustments to memory speed and timings, that I've seen, can be found here:

https://www.overclock.net/threads/comprehensive-memory-overclocking-guide.1630388/#post-26096937
Okay! So if anything at all, I should be just increasing the clock speed to 3200. I'll take a look at the links you provided. Thanks!
 
Do you have Precision boost overdrive enabled in the BIOS?

What CPU cooler are you running, the stock cooler or an aftermarket solution?

What case and what does your case fan configuration look like?

The reason I ask these things, is because on Ryzen, this is where significant performance gains can be had. That is, aside from simply BUYING very fast low latency memory.

Anyhow, just a thought, because I've seen some pretty significant differences in overall performance between identical systems running the stock Ryzen coolers (All of them) and a decent aftermarket cooling solution. It all has to do with where the system peaks in terms of maximum boost AND how long it maintains that boost, and cooling plays a big role on that with Ryzen. Or it can anyhow.
 
  • Like
Reactions: Prad_Bitt
Do you have Precision boost overdrive enabled in the BIOS?

What CPU cooler are you running, the stock cooler or an aftermarket solution?

What case and what does your case fan configuration look like?

The reason I ask these things, is because on Ryzen, this is where significant performance gains can be had. That is, aside from simply BUYING very fast low latency memory.

Anyhow, just a thought, because I've seen some pretty significant differences in overall performance between identical systems running the stock Ryzen coolers (All of them) and a decent aftermarket cooling solution. It all has to do with where the system peaks in terms of maximum boost AND how long it maintains that boost, and cooling plays a big role on that with Ryzen. Or it can anyhow.
PBO enabled. No stock cooler that thing was giving me 95° with pubg lol. I have a hotbox corsair 275R and I was using it with stock fans while building. Then I bought Hyper 212 RGB, used included thermal paste (I think I used a tad too little), bought 2x sp140 rgb icue front intake (removed front panel, only dust filter there now). 1x noctua ippc 3000rpm 120mm as exhaust. The stock fans are on top, one pushed as far back as possible exhaust (slightly restricted by VRM heatsink on mobo, B550 Aorus pro ac) and one is as far front as possible intake. The reason I did this even though it sounds stupid is somehow I get lower temps like this 🤷🏻‍♂️ at least on cpu. I didn't really check gpu but it's winter and my gpu barely even crosses 55°C lol. RTX 2070 Super Windforce OC 3X 8G.
 
Last edited:
No, that seems right to me from what I could tell. The 212 EVO isn't definitely the greatest choice out there, but certainly better than the stock wraith cooler for sure.

PBO sometimes makes a mess of things though if you don't have VERY good cooling. I'd recommend that you make a test.

With PBO enabled, run a series of benchmarks. Which ones are up to you but I'd recommend something like 3dMark, Passmark performance test, Uningine superposition, Uningine Heaven and maybe also Cinebench, and note all the scores.

Then disable PBO in the BIOS, but don't change anything else. Run the same benchmarks again and see what the difference is. On some systems that didn't have exceptional high end cooling I've seen some of them actually score BETTER without PBO enabled than with it, simply due to the system maintaining a slightly lower peak boost but maintaining it for LONGER without PBO enabled. So it might be worth checking both ways just to see and all of those benchmark tools should be free as well.
 
  • Like
Reactions: Prad_Bitt
No, that seems right to me from what I could tell. The 212 EVO isn't definitely the greatest choice out there, but certainly better than the stock wraith cooler for sure.

PBO sometimes makes a mess of things though if you don't have VERY good cooling. I'd recommend that you make a test.

With PBO enabled, run a series of benchmarks. Which ones are up to you but I'd recommend something like 3dMark, Passmark performance test, Uningine superposition, Uningine Heaven and maybe also Cinebench, and note all the scores.

Then disable PBO in the BIOS, but don't change anything else. Run the same benchmarks again and see what the difference is. On some systems that didn't have exceptional high end cooling I've seen some of them actually score BETTER without PBO enabled than with it, simply due to the system maintaining a slightly lower peak boost but maintaining it for LONGER without PBO enabled. So it might be worth checking both ways just to see and all of those benchmark tools should be free as well.
Will do! The problem here is this cooler which is supposed to be $35 is $60. And the Freezer 34 didn't exist here before last month. NH-U12S was $90, and the 212 is now out of stock lol. Just after I bogiut the 275R for $100 (crazy I know, prices were sky high then) Lian Li released the L2Mesh for $120. My luck can never be underestimated. Also isn't the rgb version a bit better? I have the rgb one I could swear I read reviews this is quite good. Anyway, when I did Cinebench, I used to hit 76°-78° in multi core, single core was around 60 I think, can't really remember. Score was 3591 on multi and 484 or 494 on Single. This was during the summers with ambient temps around 28
 
My advice on that would be to just use the cooler you have, for now. Keep your eye open for any deals on a decent 140mm cooler. The 120mm coolers are just not going to give you the same kind of performance no matter what fans you put on them nor how many fans are on them. It's the design of the heat pipes and the total surface area of the fin array that makes the most difference, and then we follow that with the static pressure and CFM of the fans, and then at the end, maybe drop another 1-3°C by the addition of a second fan. Primarily though, for your CPU, any GOOD 140mm single fin stack cooler will do a good job.

I'd try to stick to units sold by Thermalright, Noctua, Deepcool, Phanteks, Cryorig, BeQuiet or Scythe, in general. I'm sure there are some other brands with a few decent coolers, but those are the companies that we KNOW for sure have models that are well reviewed and perform up to the hype. So maybe at some point you can find a deal on a better cooler and make a change at that time. Maybe sell the 212 EVO to a friend or somebody else, or use it for another system.
 
My advice on that would be to just use the cooler you have, for now. Keep your eye open for any deals on a decent 140mm cooler. The 120mm coolers are just not going to give you the same kind of performance no matter what fans you put on them nor how many fans are on them. It's the design of the heat pipes and the total surface area of the fin array that makes the most difference, and then we follow that with the static pressure and CFM of the fans, and then at the end, maybe drop another 1-3°C by the addition of a second fan. Primarily though, for your CPU, any GOOD 140mm single fin stack cooler will do a good job.

I'd try to stick to units sold by Thermalright, Noctua, Deepcool, Phanteks, Cryorig, BeQuiet or Scythe, in general. I'm sure there are some other brands with a few decent coolers, but those are the companies that we KNOW for sure have models that are well reviewed and perform up to the hype. So maybe at some point you can find a deal on a better cooler and make a change at that time. Maybe sell the 212 EVO to a friend or somebody else, or use it for another system.
I was thinking I'd either wait for the Noctua NH-U12A Chromax when it comes out , which is a 120mm but has 7 heat pipes and seems to have consistently better results than even the D15. Or an AIO 240mm when I have the money, which probably won't be before next September
 
  • Like
Reactions: Darkbreeze