Overclocking RAM

[Pasgui9]

I have a memory kit of Corsair Vengeance LPX 3000MHz C15 RAM, and it was bought before I moved to the Ryzen platform, however I heard that 3600MHz C16 was a sweetspot for the R5 3600. I went into the BIOS, turned off XMP and tweaked the frequency and timings, however even on 3600MHz 18 21 21 21 42 any game crashes after about 2 minutes. I know that certain Vengeance kits can get to 3600MHz through XMP and i figured mine can get to that through overclocking but I must be mistaken. What am I doing wrong? The motherboard supports it, it's an MSI B450 Gaming Plus Max.

[OCNewbee]

Hmmm afaik if you ram is rated at 3000, then 3200 is prolly the highest you can hit. Althogh theoretically 3000 cl15 is similar to 3600 cl 18, things dont always work out that way. Why dont u try 3200 cl16 first?

[Pasgui9]

30 minutes ago, OCNewbee said:

Hmmm afaik if you ram is rated at 3000, then 3200 is prolly the highest you can hit. Althogh theoretically 3000 cl15 is similar to 3600 cl 18, things dont always work out that way. Why dont u try 3200 cl16 first?

alright. thank you for your advice

[TechJezus]

What voltage are you using for the oc? 

[Jurrunio]

Going beyond rated specs is always a gamble with only a couple of exceptions. That being Crucial 3200MHz CL16 kits and above or kits with timings so tight, it can only be Samsung B-die. 3000MHz CL15 is the kind of generic bin that you get all sorts of possible dies, the only common factor here is that none of them are good.

 

After all this is why they bin memory, good stuff gets sold for more and bad stuff for less. Sometimes good stuff appear on cheap kits because there's only so much demand for the more expensive good stuff and they need more supply on cheaper kits. Another reason is that the kit responds differently than expected, for example a certain kit improves more than others when pushed to 1.4V, but because the vendor only tested for 1.35V and it did badly, it got binned to a bad bin. Some people get lucky to get their hands on good stuff sold cheaper than usual, but this should never be relied on.

 

Oh and I hope this isn't news to you, but names like "Corsair Vengeance", "Gskill Trident Z Royal" only refers to their heatspreader and possible LEDs. It has nothing to do with the memory dies underneath the heatspreader.

[Pasgui9]

Just now, TechJezus said:

What voltage are you using for the oc? 

It's automatically set by my motherboard. It has a feature where it presets frequencies, timings and voltages based on what the community recommends. I haven't looked at it but the 3600MHz kit is at 1.35V so I shouldn't go above that, right?

 

[Pasgui9]

2 minutes ago, Jurrunio said:

Going beyond rated specs is always a gamble with only a couple of exceptions. That being Crucial 3200MHz CL16 kits and above or kits with timings so tight, it can only be Samsung B-die. 3000MHz CL15 is the kind of generic bin that you get all sorts of possible dies, the only common factor here is that none of them are good.

 

After all this is why they bin memory, good stuff gets sold for more and bad stuff for less. Sometimes good stuff appear on cheap kits because there's only so much demand for the more expensive good stuff and they need more supply on cheaper kits. Another reason is that the kit responds differently than expected, for example a certain kit improves more than others when pushed to 1.4V, but because the vendor only tested for 1.35V and it did badly, it got binned to a bad bin. Some people get lucky to get their hands on good stuff sold cheaper than usual, but this should never be relied on.

 

Oh and I hope this isn't news to you, but names like "Corsair Vengeance", "Gskill Trident Z Royal" only refers to their heatspreader and possible LEDs. It has nothing to do with the memory dies underneath the heatspreader.

The last part is actually news to me. I expected some consistency in their marketing. The part with binning I did know.

[Jurrunio]

9 minutes ago, Pasgui9 said:

The last part is actually news to me. I expected some consistency in their marketing. The part with binning I did know.

for example, this

https://www.gskill.com/product/165/184/1535625724/F4-2133C15D-16GVRRipjaws-VDDR4-2133MHz-CL15-15-15-35-1.20V16GB-(2x8GB)

and this

https://www.gskill.com/product/165/184/1596609585/F4-4400C16D-16GVKRipjaws-VDDR4-4400MHz-CL16-19-19-39-1.50V16GB-(2x8GB)

 

Both Ripjaws V, but the red (more gamery?) one is not even half as fast as the black one.

[OCNewbee]

20 minutes ago, TechJezus said:

What voltage are you using for the oc? 

@Pasgui9 doh. This is by far the most important factor OP. As TechTezus mentioned, voltage is everything for stability.

 

Just thinking about this issue again, the fact that you can boot at 3600mhz means that 3600mhz is very possible. Up the voltage and see where that gets you. Try 1.4v first. Monitor the temps and try lowering if it is stable. IMO I think at 1.4v you should be stable. But be warned, you may end up burning your ram (highly unlikely tho) 

[Jurrunio]

24 minutes ago, Pasgui9 said:

I haven't looked at it but the 3600MHz kit is at 1.35V so I shouldn't go above that, right?

 

Since now you know your kit is rated low because it can only go so far, you have to use more voltage. That said how well it responds to voltage and how far you can go until adding voltage doesnt help with potential needs testing. Good thing with DDR4 is that the point of end of scaling occurs way before the memory die gets damaged by voltage, so feel free to go higher. If you want a limit to follow, I'd say 1.5V for testing and 1.4v for long term use

 

try loosen tRP and tRCD more, so maybe 3600MHz 19-26-26-54

[Pasgui9]

8 hours ago, Jurrunio said:

Since now you know your kit is rated low because it can only go so far, you have to use more voltage. That said how well it responds to voltage and how far you can go until adding voltage doesnt help with potential needs testing. Good thing with DDR4 is that the point of end of scaling occurs way before the memory die gets damaged by voltage, so feel free to go higher. If you want a limit to follow, I'd say 1.5V for testing and 1.4v for long term use

 

try loosen tRP and tRCD more, so maybe 3600MHz 19-26-26-54

sure, I'll try. Any good programs for testing memory?

[Pasgui9]

8 hours ago, OCNewbee said:

@Pasgui9 doh. This is by far the most important factor OP. As TechTezus mentioned, voltage is everything for stability.

 

Just thinking about this issue again, the fact that you can boot at 3600mhz means that 3600mhz is very possible. Up the voltage and see where that gets you. Try 1.4v first. Monitor the temps and try lowering if it is stable. IMO I think at 1.4v you should be stable. But be warned, you may end up burning your ram (highly unlikely tho) 

Alright, I'll give it a go at 1.4V

[Jurrunio]

17 minutes ago, Pasgui9 said:

sure, I'll try. Any good programs for testing memory?

https://www.hcidesign.com/memtest/

 

Each instance uses 1 CPU thread for testing but you should leave 1 core (2 threads) and some RAM for the OS to run on. So for R5 3600 and 16GB memory, I suggest running 1400MB on each instance and 10 instances together. Completing 100% will be "mostly stable" and you can go back to the BIOS for further tuning. Completing 400% will be good for daily use.

[Pasgui9]

On 4/6/2021 at 7:43 PM, Jurrunio said:

https://www.hcidesign.com/memtest/

 

Each instance uses 1 CPU thread for testing but you should leave 1 core (2 threads) and some RAM for the OS to run on. So for R5 3600 and 16GB memory, I suggest running 1400MB on each instance and 10 instances together. Completing 100% will be "mostly stable" and you can go back to the BIOS for further tuning. Completing 400% will be good for daily use.

Thank you for your time!