[SOLVED] Best RAM for Ryzen CPUs ?

[heitorbonfim]

It's often said that Ryzen processors are for fast RAM, but my experience shows me that RAM at low clocks and short timings are best for them. Overclocking make processing very unstable for Ryzen's. Look:

https://valid.x86.fr/s0vynj

 

[drea.drechsler]

It's often said that Ryzen processors are for fast RAM, but my experience shows me that RAM at low clocks and short timings are best for them. Overclocking make processing very unstable for Ryzen's. Look:

https://valid.x86.fr/s0vynj

Overclocking makes any system unstable if not done properly.

RAM mfr's often use very loose timings in their XMP settings for "fast" RAM to be confident of better stability. So running fast timings well means tweaking the timings in BIOS to tighten them up to lower latency. People don't do that so that's what happens.

 

[MeanMachine41]

Hi heitorbonfim and Welcome to the Forum

OK however not the best for them!!
CPUz validation is not an indication of stability and the system should be tested using Memtest86 and four full passes.

Did you conduct a stress test at all, which one and results?

I consider Realbench a real-world tester and if you can run that for a few hours your stable.

 

[logainofhades]

It's often said that Ryzen processors are for fast RAM, but my experience shows me that RAM at low clocks and short timings are best for them. Overclocking make processing very unstable for Ryzen's. Look:

https://valid.x86.fr/s0vynj

Performance will suffer, with that slow ram speed. Running a good DDR4 3600 CL16 kit, and just turning on PBO, would perform better than what you are running at now. Overclocking Zen 3 is a waste of time, and can actually give you worse overall performance.

 

[heitorbonfim]

Hi heitorbonfim and Welcome to the Forum

OK however not the best for them!!
CPUz validation is not an indication of stability and the system should be tested using Memtest86 and four full passes.

Did you conduct a stress test at all, which one and results?

I consider Realbench a real-world tester and if you can run that for a few hours your stable.

a good DDR4 3600 CL16 kit

I tried to download RealBench V2 Leaderboard, but windows 11 didn't allow the download (is it unsafe ?). Cinebench, perhaps?

The only test I've done is my daily usage and gaming performance.

But it is true that I need a good DDR4 3600 CL16 kit LOL

 

[drea.drechsler]

I tried to download RealBench V2 Leaderboard, but windows 11 didn't allow the download (is it unsafe ?). Cinebench, perhaps?

The only test I've done is my daily usage and gaming performance.

But it is true that I need a good DDR4 3600 CL16 kit LOL

If you want optimal performance you need DDR4 3600, at least CL16. But you can do it with a lesser kit...if you know how to tweak timings and aren't afraid of using higher DRAM voltage.

There are a couple dozen timings and other settings that affect latency and stability so just randomly attacking the Advanced Timings section is not likely to get you anywhere. Go get the Ryzen DRAM Calculator. It's proven the one of the best tools to help determine timings and other settings to get the best performance out of memory on Ryzen. Using it right you have to determine the exact DRAM chips in your memory so you'll also need Typhoon Burner to read the memory module.

The settings it offers are also a bit conservative, especially using the 'SLOW' option. But if you change things a bit you get a feel for how it affects timings, which can give you ideas how to tweak at least the primary timings for even lower latency. It also has a latency test, and stability test, you can run.

DO NOT be afraid of trying higher voltages: DDR4 spec requires RAM to operate up to 1.5V but it's suggested to not go above 1.4 for most kits. Some kits (Samsung B-Die) likes higher voltage and is considered safe all the way up to 1.5V but most people stop at 1.45.

Right now I'm using a GSkill 3200 B-die kit on my 3700x at 3600, CL14, 1.46V. Full stability, over two years now.

 

[heitorbonfim]

For two months I have been changing values in BIOS, profile xmp etc.
The only change I've succeeded in maintaining stability and high performance was setting the CPU clock to 4.65 GHz and using fast memory timings and slow clock 800-933MHz, but in the BIOS auto mode to set the timings.

Right now, Cinebench gave me a good score of 11830.

Bio's default mode or overclock in XMP/Profile mode doesn't work well, no more than a score of 10600.

And games work very well. Have no blue screens and no kind of crashes.

Unfortunately, I don't have the skills to change power and timings safely wisely. There are several dozen changes to the BIOS. Have no skills LOL. AUTO is my favorite command.

A good kit of memory is a damn good solution. But where i live is not easy to be found.

 

[MeanMachine41]

I tried to download RealBench V2 Leaderboard, but windows 11 didn't allow the download (not safety(???)). Cinebench, perhaps?

Cinebench is not a RAM tester but more as a benchmark for your GPU.

Pose a question at ROG forum re: your inability to download Realbench
I use version 2.5.6 from Techspot

 

[drea.drechsler]

Cinebench is not a RAM tester but more as a benchmark for your GPU.
..

Maybe a typo but Cinebench is a benchmark test for CPU...not GPU. Cinebench 1.5 had a GPU test included but it was pretty worthless even for the time.

And it doesn't seem very memory sensitive though, especially for memory latency.

 

[drea.drechsler]

For two months I have been changing values in BIOS, profile xmp etc.
The only change I've succeeded in maintaining stability and high performance was setting the CPU clock to 4.65 MHz and using fast memory timings and slow clock 800-933MHz, but in the BIOS auto mode to set the timings.

Right now, Cinebench gave me a good score of 11830.

Bio's default mode or overclock in XMP/Profile mode doesn't work well, no more than a score of 10600.

And games work very well. Have no blue screens and no kind of crashes.

Unfortunately, I don't have the skills to change power and timings safely wisely. There are several dozen changes to the BIOS. Have no skills LOL. AUTO is my favorite command.

A good kit of memory is a damn good solution. But where i live is not easy to be found.

You can make a straight up simple test to see if you have any possible chance of getting it to work.

First things first is save a BIOS settings profile of what you're using now so you can get back to it easily.

Change these timings: TCAS to 18, Trcd to 18, Trp to 18, Tras to 40, Trc to 58.

Change your DRAM voltage to 1.45V. This is JUST to get started, you'll lower it later if it works at all. Leave all the other timings and settings defaults, auto.

To start, set memory clock to 3400 (1700Mhz) and restart. The first hurdle is to get through boot looping and into the OS. If you get there, then try a memory stability test. If you can't get through boot looping or the test, you might as well just reset CMOS and return to your saved settings as you'll need to fiddle a whole lot more to get it to work.

If it does pass: raise it to 3500 (or whatever's close) and try again. Then 3600. If it gets this far with stability your RAM has possibilities. But you're not fiddling with other settings, such as ProcODT or GearDown mode, that can help a lot with boot looping with tighter timings and performance at higher clocks.

Now test with a long, inclusive memory test. If it passes now you can lower voltage to 1.40 and try again. If it passes you can try lowering Tcas to 16. If it fails raise voltage a little more (1.41 or 1.42), if it passes lower voltage to 1.38 and try again.

The main thing is to not be afraid of using voltage: DDR4 standard memory is required to be capable of operating at 1.5V so your safe. The second thing is you're no where near optimized, so latency may not be as good as it could be. But you'll know your RAM has possibilities if you ever should want to try more after finding out what DRAM Calculator suggests for your setup.

And BTW: using XMP defaults your RAM timings are more than likely no where near being optimized anyway so you probably don't have anything (except time) to lose trying this.

 

[logainofhades]

I tried to download RealBench V2 Leaderboard, but windows 11 didn't allow the download (not safety(???)). Cinebench, perhaps?

The only test I've done is my daily usage and gaming performance.

But it is true that I need a good DDR4 3600 CL16 kit LOL

I am using a 2x16gb Crucial RGB kit, myself.

 

[heitorbonfim]

You can make a straight up simple test to see if you have any possible chance of getting it to work.

First things first is save a BIOS settings profile of what you're using now so you can get back to it easily.

Change these timings: TCAS to 18, Trcd to 18, Trp to 18, Tras to 40, Trc to 58.

Change your DRAM voltage to 1.45V. This is JUST to get started, you'll lower it later if it works at all. Leave all the other timings and settings defaults, auto.

To start, set memory clock to 3400 (1700Mhz) and restart. The first hurdle is to get through boot looping and into the OS. If you get there, then try a memory stability test. If you can't get through boot looping or the test, you might as well just reset CMOS and return to your saved settings as you'll need to fiddle a whole lot more to get it to work.

If it does pass: raise it to 3500 (or whatever's close) and try again. Then 3600. If it gets this far with stability your RAM has possibilities. But you're not fiddling with other settings, such as ProcODT or GearDown mode, that can help a lot with boot looping with tighter timings and performance at higher clocks.

Now test with a long, inclusive memory test. If it passes now you can lower voltage to 1.40 and try again. If it passes you can try lowering Tcas to 16. If it fails raise voltage a little more (1.41 or 1.42), if it passes lower voltage to 1.38 and try again.

The main thing is to not be afraid of using voltage: DDR4 standard memory is required to be capable of operating at 1.5V so your safe. The second thing is you're no where near optimized, so latency may not be as good as it could be. But you'll know your RAM has possibilities if you ever should want to try more after finding out what DRAM Calculator suggests for your setup.

And BTW: using XMP defaults your RAM timings are more than likely no where near being optimized anyway so you probably don't have anything (except time) to lose trying this.

I tried something like that. 1.35V and AUTO timing. I clocked the memory up to 3000 MHz, but there was no performance gain in games.

But I'm not sure what Bios did, because the Artificial Intelligence probably acted so that I wouldn't screw up my computer.

I am using google translator, i am not native english. Be patient.

 

[heitorbonfim]

Maybe a typo but Cinebench is a benchmark test for CPU...not GPU. Cinebench 1.5 had a GPU test included but it was pretty worthless even for the time.

And it doesn't seem very memory sensitive though, especially for memory latency.

Indeed, agree.

 

[drea.drechsler]

I tried something like that. 1.35V and AUTO timing. I clocked the memory up to 3000 MHz, but there was no performance gain in games.
...

I've no doubt you're right. Games aren't usually memory sensitive as they're so GPU bound.

But I have read there are SOME games that are memory latency sensitive. IRRC they're e-sports, super high FPS type games. I'm sure you need a super tweaked system with extreme top-end components, and super human golden-eyes for FPS, to notice any difference. Not something a boring mortal such as I would notice.

 

[heitorbonfim]

I've no doubt you're right. Games aren't usually memory sensitive as they're so GPU bound.

But I have read there are SOME games that are memory latency sensitive. IRRC they're e-sports, super high FPS type games. I'm sure you need a super tweaked system with extreme top-end components, and super human golden-eyes for FPS, to notice any difference. Not something a boring mortal such as I would notice.

Indeed, i measure performance with my Chicken Dinners on PUBG and lags on gameplay. If has no LAG when i shoot, that is good.