• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Corsair RAM Frequency Tuning Help

Joined
Dec 31, 2021
Messages
18 (0.01/day)
System Name Nalisse
Processor AMD Ryzen 7 1700X @ 3.40GHz
Motherboard MSI B450 TOMAHAWK MAX
Cooling Arctic Freezer 7X
Memory 32GB (4x8GB) DDR4 CL16 (16-18-18-36) Corsair Vengeance RBG PRO @ 2667MHz
Video Card(s) EVGA NVIDIA Geforce RTX 2070 SUPER BLACK GAMING 8GB
Storage 1x 1TB WD Caviar Black HDD + 2TB Micro 1100 SATA SSD + SKHynix 256GB NVMe SSD
Display(s) Main 32-inch Samsung LCD HDTV + Secondary 19-inch Samsung LCD Monitor
Case Evesky G82 ATX Case
Audio Device(s) CREATIVE 5.1 Channel Home Theater
Power Supply EVGA 600 BQ
Mouse Logitech G305 Lightspeed Wireless Mouse
Keyboard Logitech G510s
VR HMD Someday
Software Windows 11 64-Bits
Hello! I wanted to ask if it was possible for anyone to help me and shed some light on some RAM frequency matters, here's the situation:

Up until now I've had a 16GB (2x8GB) CL16 (16-18-18-36) Corsair Vengeance RGB Pro RAM kit, it's worked great with my rig for the longest time, now being like 3-4years or so? Running constantly at its rated speed of 3200Mhz with XMP.

Today I bought a second kit of similar RAM to upgrade to 32GB total, same model and timings, also 2x8GB (so 4 DIMMs now), and also Corsair Vengeance RGB Pros, only difference is the heatsink being white instead of black I guess.

Up until now both kits have been working okay, installed them all on my rig and all of them are detected and the capacity is shown correctly to be 32GB, only thing is the RAM frequencies are off.

Trying to set the frequencies to 3200MHz via XMP causes my PC to reboot and default back down to 2133MHz, not the end of the world to me, but I would like to know if there's a way to bring all of the 32GB up to their rated speeds of 3200MHz, at the moment I experimented with some tweaking of voltages and frequencies, the highest I could get them was by manually setting the frequency to 2866MHz with a DRAM voltage of around 1.390v, and currently I'm running them at 2800MHz with automatic voltages, I don't think 400Mhz will do a huge difference for me but I would like to know more about this matter.

I'm wondering if giving it more voltage in order to reach 3200MHz is risky? Or if it even is the correct way to do it? Or if it is any sort of hardware limitation of running 32GB of RAM? I'm very inexperienced in RAM tuning after all, any help would be deeply appreciated.

Processor: AMD Ryzen 7 1700X @ 3.40GHz
Motherboard: MSI B450 TOMAHAWK MAX
Memory: 32GB (4x8GB) CL16 16-18-18-36 DDR4 Corsair Vengeance RBG PRO 3200MHz
 

Attachments

  • imagen_2023-12-18_181502372.png
    imagen_2023-12-18_181502372.png
    833.6 KB · Views: 110
  • imagen_2023-12-18_181548250.png
    imagen_2023-12-18_181548250.png
    14.4 KB · Views: 111
Generally you don't want to mix ram kits for this very reason although I have done the same thing once but with kits that likely came off the same lot and purchased at the same time. In your case many years apart and are likely different IC's (chips) between the ram kits causing incompatibility. If the model numbers on the back of the RAM are the same but the version numbers next to the model numbers are different this is likely the case you have mixed IC's. The best thing to do is get a 4 dimm 32GB kit or get a 2 dimm 32GB kit to be sure they will work together. My choice would be to get a 2 dimm 32GB dual-rank kit for the best compatibility and performance however I am not sure if your Zen based CPU will like that as much as Zen2/Zen3 CPU's do.
 
Last edited:
If you want to learn more about the differences between your kits you can download and use the free version of Thaiphoon Burner (https://www.softnology.biz/files.html) to look at the information stored in your RAM's SPD chips. This data contains the characteristics of each RAM module. You can use this to compare and see how close or different they are. However it's not a foolproof tool and relies on the RAM manufacturer to program the SPD chips correctly and accurately.

If you decide to keep your mixed RAM and just run it at the lower speeds it would be a good idea to use PassMark's Free Memtest86 (https://www.memtest86.com/) to get a first chance look to test if your installed memory is likely to encounter errors.

Generally you don't want to mix ram kits for this very reason although I have done the same thing once but with kits that likely came off the same lot and purchased at the same time. In your case many years apart and are likely different IC's (chips) between the ram kits causing incompatibility. If the model numbers on the back of the RAM are the same but the version numbers next to the model numbers are different this is likely the case you have mixed IC's. The best thing to do is get a 4 dimm 32GB kit or get a 2 dimm 32GB kit to be sure they will work together. My choice would be to get a 2 dimm 32GB dual-rank kit for the best compatibility and performance however I am not sure if your Zen based CPU will like that as much as Zen2/Zen3 CPU's do.
In addition to getting a matched kit to replace your current RAM setup, keep in mind that the configuration of RAM you install may influence the actual speed you can run.

Here is an example from ASRock that tries to inform the user about this complexity.

1702940958758.png


I don't recall the general RAM specifications for Summit Ridge CPU's (Zen) like your 1700X but it's very similar to the above chart. Basically the rule of thumb was the more memory ranks you add the lower speed you needed to run that memory. This is because of limitations of the IMC (integrated memory controller) on your CPU. On the flipside Ryzen CPU's in general could often exceed these specification and run overclocked like what you had be doing running your original kit at DDR4-3200 aka XMP speeds.

Regarding voltages... different IC's respond differently to increases in voltage. Some need more and some even work better with decreases in voltage.
There is another voltage called SOC. Since you are inexperienced in RAM tuning be careful to NOT adjust SOC voltage above 1.2v but know that when troubleshooting and overclocking 1.1v is often used.
Also keep in mind when putting in voltages in UEFI/BIOS that you might not get that exact voltage in reality.

Ram tuning takes a lot of time to test for stability and people tend to not want to do that which is why I recommend getting a kit in the capacity you need that "just works" and just keep in mind you might need to reduce the frequency on the chance your CPU can't handle 32GB at DDR4-3200 in a 4 or 2 dimm configuration.
 
Last edited:
Generally you don't want to mix ram kits for this very reason although I have done the same thing once but with kits that likely came off the same lot and purchased at the same time. In your case many years apart and are likely different IC's (chips) between the ram kits causing incompatibility. If the model numbers on the back of the RAM are the same but the version numbers next to the model numbers are different this is likely the case you have mixed IC's. The best thing to do is get a 4 dimm 32GB kit or get a 2 dimm 32GB kit to be sure they will work together. My choice would be to get a 2 dimm 32GB dual-rank kit for the best compatibility and performance however I am not sure if your Zen based CPU will like that as much as Zen2/Zen3 CPU's do.
Yeaaah, I've always made a mental note to try and never mismatch my RAM sticks, but since I usually buy used it has become a little harder for me to do so properly, for the next time I am certainly going to stick to RAMs of the same kit and lot, definitely not years apart.

Do they have different revision numbers on the barcode? If so they aren't the same and cause you lots of headaches
After checking the barcodes they DO seem different, yes, I should have probably added that my first kit of RAM was also a hand me down from a friend, and apparently said kit is where the mismatch came from, it still ran a 3200MHz but I suppose that was just luck?

If you want to learn more about the differences between your kits you can download and use the free version of Thaiphoon Burner (https://www.softnology.biz/files.html) to look at the information stored in your RAM's SPD chips. This data contains the characteristics of each RAM module. You can use this to compare and see how close or different they are. However it's not a foolproof tool and relies on the RAM manufacturer to program the SPD chips correctly and accurately.
Wow that tool Thaiphoon Burner is actually really cool! I had never heard about it but it sure gave me the information I needed, three of my RAM sticks seem to be of the same type, but the other one apparently is different. (Will attach screencaps below)

The mismatch is as follows:

Three of the RAM are essentially identical according to Thaiphoon, all three sticks have the exact same specs, model, identical minimum timing delays, same manufacturer (STMicroelectronics), and all 3 share the following Part Number:
CMW16GX4M2C3200C16.
The utility marks their speed grade at 2133P Downbin.

And ONE of the RAM sticks is different, different timing delays, has a different manufacturer (Giantec), different model, and has the following Part Number:
CMW8GX4M1Z3200C16
The utility marks its speed grade at 2666U.

Not trying to jump to conclussions but I suppose the mismatched single stick of my older kit is causing the issues, hate to see it.

As of now I don't have the means to get a proper matching kit so I may stick to using them as is and running Memtest86 to check their functionality.

Also wow thank you for such a thorough explanation on the way a RAM setup like this would work! And thank you so much for the pointers about overclocking, I might just try it just a little bit more while being very, VERY conservative with the voltages, but realistically I still doubt the drop from 3200MHz OC down to 2800MHz will cause me any real issues performance-wise, I'll mostly just stick to testing for stability in the memory modules as is, and try to stress test the system to see how it behaves!
 

Attachments

  • imagen_2023-12-18_193913634.png
    imagen_2023-12-18_193913634.png
    84.7 KB · Views: 86
  • imagen_2023-12-18_193924403.png
    imagen_2023-12-18_193924403.png
    85.1 KB · Views: 102
Hello! I wanted to ask if it was possible for anyone to help me and shed some light on some RAM frequency matters, here's the situation:

Up until now I've had a 16GB (2x8GB) CL16 (16-18-18-36) Corsair Vengeance RGB Pro RAM kit, it's worked great with my rig for the longest time, now being like 3-4years or so? Running constantly at its rated speed of 3200Mhz with XMP.

Today I bought a second kit of similar RAM to upgrade to 32GB total, same model and timings, also 2x8GB (so 4 DIMMs now), and also Corsair Vengeance RGB Pros, only difference is the heatsink being white instead of black I guess.

Up until now both kits have been working okay, installed them all on my rig and all of them are detected and the capacity is shown correctly to be 32GB, only thing is the RAM frequencies are off.

Trying to set the frequencies to 3200MHz via XMP causes my PC to reboot and default back down to 2133MHz, not the end of the world to me, but I would like to know if there's a way to bring all of the 32GB up to their rated speeds of 3200MHz, at the moment I experimented with some tweaking of voltages and frequencies, the highest I could get them was by manually setting the frequency to 2866MHz with a DRAM voltage of around 1.390v, and currently I'm running them at 2800MHz with automatic voltages, I don't think 400Mhz will do a huge difference for me but I would like to know more about this matter.

I'm wondering if giving it more voltage in order to reach 3200MHz is risky? Or if it even is the correct way to do it? Or if it is any sort of hardware limitation of running 32GB of RAM? I'm very inexperienced in RAM tuning after all, any help would be deeply appreciated.

Processor: AMD Ryzen 7 1700X @ 3.40GHz
Motherboard: MSI B450 TOMAHAWK MAX
Memory: 32GB (4x8GB) CL16 16-18-18-36 DDR4 Corsair Vengeance RBG PRO 3200MHz
The first generation Ryzen (Summit Ridge) is not able to run 4 dimms at 3200MHz. My 1600X wasn't able to run at 3200MHz with 4 dimms of Samsung B-die.
The highest I got was 2933MHz stable. 3066MHz wasn't stable.
If you want to run at 3200MHz, you very likely need to upgrade your processor to a Matisse (e.g. 3700X) or newer.

Note: Your motherboard could have a VRM issue which causes random black screens with a Ryzen 5900X (105W TDP) under full load until you put the processor in ECO mode. Keep this in mind when upgrading the processor.

I might just try it just a little bit more while being very, VERY conservative with the voltages, but realistically
If those stick have Samsung C-die on it, dont push the memory voltage above the rated voltage since those IC's dont like high voltages and will slowly die.
 
The first generation Ryzen (Summit Ridge) is not able to run 4 dimms at 3200MHz. My 1600X wasn't able to run at 3200MHz with 4 dimms of Samsung B-die.
The highest I got was 2933MHz stable. 3066MHz wasn't stable.
If you want to run at 3200MHz, you very likely need to upgrade your processor to a Matisse (e.g. 3700X) or newer.

Note: Your motherboard could have a VRM issue which causes random black screens with a Ryzen 5900X (105W TDP) under full load until you put the processor in ECO mode. Keep this in mind when upgrading the processor.


If those stick have Samsung C-die on it, dont push the memory voltage above the rated voltage since those IC's dont like high voltages and will slowly die.
That makes a lot of sense honestly, yes, my 1700X is at its limit with 4 DIMMs, and the highest these will go is 2800MHz, and I haven't even tested them for games and productivity like video editing or 3D modelling, still gotta do that.

Guess that for my next Ryzen rig I'll be looking more carefully at RAM specs and how those mix with the CPUs capacity, I had always wanted to upgrade to Matisse but at this point I believe I will hold out on the upgrade, and go all the way to Vermeer down the road, unless I happen to find a nice deal on an used Matisse processor or something.

Thanks a lot!
 
That makes a lot of sense honestly, yes, my 1700X is at its limit with 4 DIMMs, and the highest these will go is 2800MHz, and I haven't even tested them for games and productivity like video editing or 3D modelling, still gotta do that.

Guess that for my next Ryzen rig I'll be looking more carefully at RAM specs and how those mix with the CPUs capacity, I had always wanted to upgrade to Matisse but at this point I believe I will hold out on the upgrade, and go all the way to Vermeer down the road, unless I happen to find a nice deal on an used Matisse processor or something.

Thanks a lot!
If you plan on keeping your AM4 system, it's Vermeer all along, in my opinion. It's a whole different architecture than Matisse with 1x8-core compute chips instead of 2x4, which shows in the cache configuration and performance as well. Besides, Vermeer doesn't have Windows power profiles anymore, just a slider when the profile is "balanced", which is also a plus, I think.

You'll have guaranteed 3200 MHz RAM support that solves your problem with your current system. :)
 
The first generation Ryzen (Summit Ridge) is not able to run 4 dimms at 3200MHz. My 1600X wasn't able to run at 3200MHz with 4 dimms of Samsung B-die.
The highest I got was 2933MHz stable. 3066MHz wasn't stable.
If you want to run at 3200MHz, you very likely need to upgrade your processor to a Matisse (e.g. 3700X) or newer.

Note: Your motherboard could have a VRM issue which causes random black screens with a Ryzen 5900X (105W TDP) under full load until you put the processor in ECO mode. Keep this in mind when upgrading the processor.


If those stick have Samsung C-die on it, dont push the memory voltage above the rated voltage since those IC's dont like high voltages and will slowly die.

With a new Crucial F-Die-Kit my System (Ryzen 1700X) is now capable of a Bandwith above 3000 MT/s. I never thougt this is possible, but its luck in the Silicone-Lottery. You should give it a Try. Most People don't consider to use Micron after they quit their Fancy-Military-Style Ballistix-Brand.
 
Yeaaah, I've always made a mental note to try and never mismatch my RAM sticks, but since I usually buy used it has become a little harder for me to do so properly, for the next time I am certainly going to stick to RAMs of the same kit and lot, definitely not years apart.


After checking the barcodes they DO seem different, yes, I should have probably added that my first kit of RAM was also a hand me down from a friend, and apparently said kit is where the mismatch came from, it still ran a 3200MHz but I suppose that was just luck?


Wow that tool Thaiphoon Burner is actually really cool! I had never heard about it but it sure gave me the information I needed, three of my RAM sticks seem to be of the same type, but the other one apparently is different. (Will attach screencaps below)

The mismatch is as follows:

Three of the RAM are essentially identical according to Thaiphoon, all three sticks have the exact same specs, model, identical minimum timing delays, same manufacturer (STMicroelectronics), and all 3 share the following Part Number:
CMW16GX4M2C3200C16.
The utility marks their speed grade at 2133P Downbin.

And ONE of the RAM sticks is different, different timing delays, has a different manufacturer (Giantec), different model, and has the following Part Number:
CMW8GX4M1Z3200C16
The utility marks its speed grade at 2666U.

Not trying to jump to conclussions but I suppose the mismatched single stick of my older kit is causing the issues, hate to see it.

As of now I don't have the means to get a proper matching kit so I may stick to using them as is and running Memtest86 to check their functionality.

Also wow thank you for such a thorough explanation on the way a RAM setup like this would work! And thank you so much for the pointers about overclocking, I might just try it just a little bit more while being very, VERY conservative with the voltages, but realistically I still doubt the drop from 3200MHz OC down to 2800MHz will cause me any real issues performance-wise, I'll mostly just stick to testing for stability in the memory modules as is, and try to stress test the system to see how it behaves!
That seems odd. Literally. I almost want to say perhaps you made an error when switching modules to read in Thaiphoon. I would have expected two to be CMW16GX4M2C3200C16 and the other two to be CMW8GX4M1Z3200C16. Can you verify?

If you indeed did make a mistake and the ram sticks are as I described (2a and 2b) you can try this experiment if both kits rated at the same DRAM voltage. (it's listed on the back of each module) If so you could try manually setting (not using XMP) entering the DRAM voltage and the worst timings of the two kits into the UEFI/BIOS memory configuration (leaving anything you don't know as auto) and 1.1v SOC voltage and 2933MHz frequency as an interesting experiment. You may want to set UEFI/BIOS to default first and reboot to ensure you won't have any stray settings before trying this experiment. If you can successfully POST then run at least one pass of Memtest86 and see if you get any errors before attempting to boot. Reminder memory corruption can kill your OS so there is risk when not using some other test disk when experimenting with RAM overclocking.

(edit)

ok I see now after googling CMW8GX4M1Z3200C16 is from a single stick kit! So all this time prior to your new kit you have been running two different modules thinking it was a matched kit?
 
ok I see now after googling CMW8GX4M1Z3200C16 is from a single stick kit! So all this time prior to your new kit you have been running two different modules thinking it was a matched kit?
Correct! Apparently, upon closer inspection my original set of RAMs, the black ones I've been running for years that were hand me downs from a friend, are not matching modules themselves, one of them is different (Slot 1 on this CPU-Z analysis), I guess the mismatch is not really as important since I ran those at 3200MHz for years?
 

Attachments

  • 1702964280239.png
    1702964280239.png
    28.2 KB · Views: 80
  • imagen_2023-12-19_013811344.png
    imagen_2023-12-19_013811344.png
    23.5 KB · Views: 109
  • imagen_2023-12-19_013825303.png
    imagen_2023-12-19_013825303.png
    23.5 KB · Views: 71
  • imagen_2023-12-19_013836463.png
    imagen_2023-12-19_013836463.png
    23.5 KB · Views: 75
Correct! Apparently, upon closer inspection my original set of RAMs, the black ones I've been running for years that were hand me downs from a friend, are not matching modules themselves, one of them is different (Slot 1 on this CPU-Z analysis), I guess the mismatch is not really as important since I ran those at 3200MHz for years?
My guess is you were running the original mismatched set with the slower timings and that's probably why it worked.
 
My guess is you were running the original mismatched set with the slower timings and that's probably why it worked.
I see, this is about the slower DIMM of RAM usually slowing down the faster one to ensure proper functionality, I suppose?

Very well, at this point what do you suggest I should do? I tried using the PC and running stuff like Fortnite and Cyberpunk 2077, besides some weird behavior and crashing on Fortnite, I downclocked them down to 2400MHz and now it seems more stable, some performance loss but not unusable imo, I also ran Memtest86 and it passed all checks.
 
With a new Crucial F-Die-Kit my System (Ryzen 1700X) is now capable of a Bandwith above 3000 MT/s. I never thougt this is possible, but its luck in the Silicone-Lottery. You should give it a Try. Most People don't consider to use Micron after they quit their Fancy-Military-Style Ballistix-Brand.
Is this with 2 dimms or 4 dimms? I ran 3466MHz with 2 dimms of Samsung B-die with my 1600X :laugh:

@Wolf Lady Louise
Can you share the version number of each module. It should listed like ver #.##.
1702970726303.png
 
Is this with 2 dimms or 4 dimms? I ran 3466MHz with 2 dimms of Samsung B-die with my 1600X :laugh:

@Wolf Lady Louise
Can you share the version number of each module. It should listed like ver #.##.
View attachment 325961
Its with 2 Modules. Asrock stated that more than 3000 MT/s will never be stable with Summit Ridge. Yes, my Mainboard sucks - but there is RAM that could do more with Summit Ridge.
 
@AusWolf
there are still power profiles when you change how win shows stuff.
running a ups i prefer the old power layout (taskbar), and on that i still see all 3 profiles, can adjust and switch.

its more related to the change that now win has 99% for min cpu perf (100% for max),
to let the chips handle stuff, not win.

@Wolf Lady Louise
memtest (any) are good to find defective ram, not for testing stability/tweaks.

use TM5 or HCI (min 1200%) to see if your settings are stable.

besides that, corsair is known to swap chips on the lower tier stuff,
even IF you match the sticker, doesnt mean much, as they could have samsung on one and hynix on the other.

you can make it work, but its usually a pain on amd, where the mem controller is in the cpu,
intel is less sensitive on that part.

and having seen QVLand recommendations from others not always matching reality (+300$ Gb not running Gskill past jedec, but cheapest vengeance kit @3600),
i prefer to get kits showing A(X)MP (e.g. amd optimized).
 
Last edited:
@AusWolf
there are still power profiles when you change how win shows stuff.
running a ups i prefer the old power layout (taskbar), and on that i still see all 3 profiles, can adjust and switch.

its more related to the change that now win has 99% for min cpu perf (100% for max),
to let the chips handle stuff, not win.
Does it actually change anything? I have mine at 0% min and 100% max, but changing the min value doesn't seem to do anything.
 
mismatched sticks will end up giving you lower speeds. i had the same issue on my older DDR3 system. was using 2x8gb and 1x16gb. all supposedly 3600 but couldnt get them to that even with timing tweaks.
 
@AusWolf
not in a big way, more similar to using "prefer max power" on gpus, where some games with low load benefit.
i use 99/100 on balanced, but 0/60 for the power saving profile if im not gaming/encoding, enough for streaming/surfing even older games,
while keeping clocks down (5800 and up dont feel less snappy), saving a bit on power/temps.
the 99 is so win doesnt force max clocks on the cpu all the time.

not needed, but i prefer to have clocks at a steady level, rather then jumping to full every other second just because i "clicked" something.
 
Older ZEN 1 Cpu's will have trouble running XMP at 3200 with 4 dimms. While increasing voltage may get you there its probably better to sell all the sticks
and get a good 16GBx2 DR (dual rank) kit.

This will also serve you well if you upgrade from ZEN 1 to ZEN 3 which I can personally highly recommend after running my
trusty 1700X for several years. We are running similar MSi B450 boards so I know these boards can OC RAM very well.

Example of what can be done if you do increase voltages but I don't recommend this for most end users.
3667 1700X 664 MSi 2023.jpg


Something that was really crazy was getting all the way to 3933MT/s with a single dimm with said 1700X.
CINEBENCH_R15_Multi_Core_1555.jpg
 
@The King
depending on ram/board, even 2xxx and 3xxx series chips have trouble going past 3200 with 4 sticks.
 
@The King
depending on ram/board, even 2xxx and 3xxx series chips have trouble going past 3200 with 4 sticks.
Correct, This is why I recommend anyone who wants 32GB to rather go 16GBx2 than 8GBX4.

I recommended ZEN 3 has an upgrade which is 5xxx series not 2/3xxx but yes they also have much weaker IMC when compared to ZEN 3 so they too struggle with 4xdimm configs.

Looking at mobo QVL list also helps in some cases when buying ram but I found Samsung B-die and Mircon B/E die ram to work extremely well on these MSi B450 boards.
 
Hi,
Corsair memory
Well funny thing I've noticed is about all you have to do is raise the frequency besides enabling xmp profiles 1 or 2
xmp2 being the highest frequency preset.
 
@ThrashZone
past kits i have seen using A(XMP), are using the 1st profile as max clock, the 2nd one is if the MC doesnt wanna run profile 1.
at least thats what i had with +4 kits having bdies (or similar).

ignoring that most are better off doing identical settings manually.
as most lower tier boards will have higher (than needed) voltages for SOC/VDDx using profiles,
cant tell for higher end from Asus/Asrock,a s i started to prefer MSI for builds,
as even the 2nd smallest had used numbers close to what i could set manually (while still stable).
 
Back
Top