Overclock DDR4 2400 MHz RAM, currently at 3000 MHz but I've got no idea what I'm doing

[Reonu]

First of all my specs:
i7 6700k oc'd at 4.4 GHz (terrible silicon lottery) 

GTX 1070

16 GB DDR4 RAM. Specifically: Kingston HyperX Fury DDR4 2400 PC4-19200 16GB 2X8GB CL15. Or HX424C15FB2K2/16.

Asus Maximum VIII Hero motherboard

 

Sooooo I just bumped the RAM speed to 2600 MHz. Worked. Bumped to 2800 MHz. Oof, crash. Bumped voltage to 1.25V. Worked. Bumped to 3000 MHz. Oof, blue screen. Bumped voltage to 1.28V, worked.

 

And that's where I'm at. AC Odyssey runs better now, yay. I haven't touched the timings at all and I have no idea how that works. Any tips? Is 1.28V a safe voltage? Should I touch the timings, is it worth it? Where do I go from here? 

[Eastman51]

CL15 is already pretty low, I doubt you'd need to tweak that much.

 

1.28v is totally safe.

To put it in context, my Corsair Vengeance LPX 3000Mhz has to run at 1.35v for anything higher than 2133Mhz or the system won't boot (usually boot looping for a while and then restoring previous settings).

[aisle9]

Cooling is an important factor. If you have pretty direct airflow across your RAM, you can crank it up higher. That's part of why I like top-down coolers. My NH-C14S is dumping cool air onto my RAM.

 

You probably have Hynix NAND, in which case 1.45V is your absolute maximum. 1.35V is a safe number for everything but bare-PCB OEM stuff. Those HyperX Fury DIMMs will really fly if you let them.

[Reonu]

8 minutes ago, Eastman51 said:

CL15 is already pretty low, I doubt you'd need to tweak that much.

 

1.28v is totally safe.

To put it in context, my Corsair Vengeance LPX 3000Mhz has to run at 1.35v for anything higher than 2133Mhz or the system won't boot (usually boot looping for a while and then restoring previous settings).

Does it run at CL15 by default though? I thought it only ran at CL15 if I use the XMP profile. CPU-Z screenshot:

 

It's running at CL16 right? How do I change that?

6 minutes ago, aisle9 said:

Cooling is an important factor. If you have pretty direct airflow across your RAM, you can crank it up higher. That's part of why I like top-down coolers. My NH-C14S is dumping cool air onto my RAM.

 

You probably have Hynix NAND, in which case 1.45V is your absolute maximum. 1.35V is a safe number for everything but bare-PCB OEM stuff. Those HyperX Fury DIMMs will really fly if you let them.

"Probably"? Can I check which NAND I have?

[aisle9]

Just now, Reonu said:

"Probably"? Can I check which NAND I have?

Anything faster than 3200MHz is B-Die.

Anything 3200MHz with CL of 15 or lower is B-Die.

Anything 3200MHz with CL of 16 or higher is probably Hynix, maybe E-Die.

Anything slower than 3200MHz is probably Hynix.

 

There's a master list on Reddit (sorry, not sure of the URL) that lists RAM SKUs and their NAND type. There's also a program called Thaiphoon Burner that will tell you what type of NAND you have, but it's tricky to use and can damage things if you're not careful about where you click.

[Reonu]

2 minutes ago, aisle9 said:

Anything faster than 3200MHz is B-Die.

Anything 3200MHz with CL of 15 or lower is B-Die.

Anything 3200MHz with CL of 16 or higher is probably Hynix, maybe E-Die.

Anything slower than 3200MHz is probably Hynix.

 

There's a master list on Reddit (sorry, not sure of the URL) that lists RAM SKUs and their NAND type. There's also a program called Thaiphoon Burner that will tell you what type of NAND you have, but it's tricky to use and can damage things if you're not careful about where you click.

Thanks! Can you give me tips on the latency thing? I think my RAM is running at CL16, should I try to drop it to CL15? 

[nick name]

34 minutes ago, aisle9 said:

Anything faster than 3200MHz is B-Die.

Anything 3200MHz with CL of 15 or lower is B-Die.

Anything 3200MHz with CL of 16 or higher is probably Hynix, maybe E-Die.

Anything slower than 3200MHz is probably Hynix.

 

There's a master list on Reddit (sorry, not sure of the URL) that lists RAM SKUs and their NAND type. There's also a program called Thaiphoon Burner that will tell you what type of NAND you have, but it's tricky to use and can damage things if you're not careful about where you click.

There are plenty of kits faster than 3200 that aren't Samsung b-die and I believe there are 3200CL16 kits that are lesser Samsung b-die.  

 

And like he said -- use Thaiphoon Burner to READ your DIMMs.  It's super simple. 

[nick name]

38 minutes ago, Reonu said:

Thanks! Can you give me tips on the latency thing? I think my RAM is running at CL16, should I try to drop it to CL15? 

Honestly, you could save yourself a little time and set your DRAM voltage to 1.35V and then adjust timings and speed.  As mentioned before the majority of DDR4 XMP profiles run at 1.35V.  There are some that run lower and some that run higher.  So 1.35V is absolutely safe and even higher is fine and I wouldn't worry about using more voltage if you wanna push your kit a little further.  

[Reonu]

2 minutes ago, nick name said:

There are plenty of kits faster than 3200 that aren't Samsung b-die and I believe there are 3200CL16 kits that are lesser Samsung b-die.  

 

And like he said -- use Thaiphoon Burner to READ your DIMMs.  It's super simple. 

[aisle9]

9 minutes ago, Reonu said:

Thanks! Can you give me tips on the latency thing? I think my RAM is running at CL16, should I try to drop it to CL15? 

My first step in overclocking RAM is usually to take those first four configuration numbers (usually 16-16-16-36 or something similar for DDR4) and dial each down by one, then verify with Intel XTU's memory test for about ten minutes. I've never had that be an issue except on the crappiest of the crappy old DDR3 sticks (ELPIDA!). After that, it's a lot like overclocking a CPU. Dial the RAM speed up from 2400 to 2666, let the XTU memory test run for an hour or so. If it passes, jump up a step in speed and see if that's stable. If it errors out on you, bump the voltage up from 1.2 to 1.225 (or whatever that rounds to) and try again. Lather, rinse, repeat until you've reached a speed you're happy with or have taken the voltage up to the max you're comfortable with.

 

In my experience, RAM is more susceptible to the "ceiling" than CPUs are. The ceiling is the point at which the voltage jump necessary to move one increment up on speed is no longer worth it. For example, on my old i7-4790K, I could cruise right on up to 4.8GHz at 1.28V with total stability and OCCT temps in the 70s on my Kraken X61. When I tried to get to 4.9GHz, I couldn't even get the damn thing to post until I was well north of 1.3V, and it was only stable for a few minutes at a time. On my original G3258, I could get to 4.4GHz with my eyes closed at a totally safe 1.36V, but getting to 4.5GHz took 1.4+, which is outside of the safe range even for something as cool and efficient as a Pentium. That's the ceiling. You can probably stabilize the part, but it requires so much more voltage than the previous step that it isn't worth it. I find that ceiling to be lower and more concrete with RAM than with CPUs, generally speaking.

[aisle9]

5 minutes ago, nick name said:

There are plenty of kits faster than 3200 that aren't Samsung b-die and I believe there are 3200CL16 kits that are lesser Samsung b-die.  

When did that happen? A couple of years ago, the safest way to tell if a RAM kit was B-Die was to look at the speed and CL and use those rules.

[nick name]

1 minute ago, aisle9 said:

My first step in overclocking RAM is usually to take those first four configuration numbers (usually 16-16-16-36 or something similar for DDR4) and dial each down by one, then verify with Intel XTU's memory test for about ten minutes. I've never had that be an issue except on the crappiest of the crappy old DDR3 sticks (ELPIDA!). After that, it's a lot like overclocking a CPU. Dial the RAM speed up from 2400 to 2666, let the XTU memory test run for an hour or so. If it passes, jump up a step in speed and see if that's stable. If it errors out on you, bump the voltage up from 1.2 to 1.225 (or whatever that rounds to) and try again. Lather, rinse, repeat until you've reached a speed you're happy with or have taken the voltage up to the max you're comfortable with.

 

-snip-

I've never seen anyone use XTU for memory testing.  There are more common programs used by RAM overclockers that I would recommend.  Some are paid and some are free.  The worst part of RAM overclocking is the amount of time it takes to truly test RAM for stability.  Most folks run the tests overnight.  

 

TM5 with 1usmus testing profile.  Run as Administrator:

https://www.overclock.net/forum/27577522-post2594.html

 

Another program with an adjustment to spawn test windows automagically:

https://github.com/integralfx/MemTestHelper

[Reonu]

5 minutes ago, aisle9 said:

My first step in overclocking RAM is usually to take those first four configuration numbers (usually 16-16-16-36 or something similar for DDR4) and dial each down by one, then verify with Intel XTU's memory test for about ten minutes. I've never had that be an issue except on the crappiest of the crappy old DDR3 sticks (ELPIDA!). After that, it's a lot like overclocking a CPU. Dial the RAM speed up from 2400 to 2666, let the XTU memory test run for an hour or so. If it passes, jump up a step in speed and see if that's stable. If it errors out on you, bump the voltage up from 1.2 to 1.225 (or whatever that rounds to) and try again. Lather, rinse, repeat until you've reached a speed you're happy with or have taken the voltage up to the max you're comfortable with.

 

In my experience, RAM is more susceptible to the "ceiling" than CPUs are. The ceiling is the point at which the voltage jump necessary to move one increment up on speed is no longer worth it. For example, on my old i7-4790K, I could cruise right on up to 4.8GHz at 1.28V with total stability and OCCT temps in the 70s on my Kraken X61. When I tried to get to 4.9GHz, I couldn't even get the damn thing to post until I was well north of 1.3V, and it was only stable for a few minutes at a time. On my original G3258, I could get to 4.4GHz with my eyes closed at a totally safe 1.36V, but getting to 4.5GHz took 1.4+, which is outside of the safe range even for something as cool and efficient as a Pentium. That's the ceiling. You can probably stabilize the part, but it requires so much more voltage than the previous step that it isn't worth it. I find that ceiling to be lower and more concrete with RAM than with CPUs, generally speaking.

Apparently my ceiling is 3000 Mhz. 3000 Mhz works at 1.28V, but 3200 MHz won't post even at 1.39V. 

 

I tried to drop the CL to 15 but it won't post. It doesn't post at 1.3V either. I'm gonna try at 1.35V.

[nick name]

2 minutes ago, aisle9 said:

When did that happen? A couple of years ago, the safest way to tell if a RAM kit was B-Die was to look at the speed and CL and use those rules.

I'm not certain, but I always use matching timings as a quick indicator a kit might be Samsung b-die.  It holds true to about 3600MHz but there are some faster speed kits that follow as well.  14-14-14, 15-15-15, 16-16-16 and for faster than 3600MHz I've seen 17-17-17 and 18-18-18.  There are some kits like the Dominator Platinums that will use b-die and use 16-18-18 timings and those kits could also be something like Hynix CJR which can get pretty fast.  

[nick name]

2 minutes ago, Reonu said:

Apparently my ceiling is 3000 Mhz. 3000 Mhz works at 1.28V, but 3200 MHz won't post even at 1.39V. 

 

I tried to drop the CL to 15 but it won't post. It doesn't post at 1.3V either. I'm gonna try at 1.35V.

If your kit is 2400MHz at 15-15-15 then I can't imagine you'll find luck at 3200 with those same timings.  I would try 3200MHz at 16-16-16 first and if that fails then 16-18-18.  And if you're goal is to shoot for the stars then bump your DRAM voltage to 1.4V and fire away.  

[Reonu]

So it booted at CL15 3000 MHz at 1,35V. This is the timings screen in the BIOS. What else should I change?

4 minutes ago, nick name said:

I'm not certain, but I always use matching timings as a quick indicator a kit might be Samsung b-die.  It holds true to about 3600MHz but there are some faster speed kits that follow as well.  14-14-14, 15-15-15, 16-16-16 and for faster than 3600MHz I've seen 17-17-17 and 18-18-18.  There are some kits like the Dominator Platinums that will use b-die and use 16-18-18 timings and those kits could also be something like Hynix CJR which can get pretty fast.  

Did you check my screenshots? My DRAM is from Micron right?

[Reonu]

6 minutes ago, nick name said:

If your kit is 2400MHz at 15-15-15 then I can't imagine you'll find luck at 3200 with those same timings.  I would try 3200MHz at 16-16-16 first and if that fails then 16-18-18.  And if you're goal is to shoot for the stars then bump your DRAM voltage to 1.4V and fire away.  

But how do I enter these values? Check my BIOS photo.

Do you mean like this?

DRAM CAS# Latency - 16

DRAM RAS# to CAS# Latency - 16

DRAM RAS# ACT Time - 16

 

Current latencies:

[nick name]

1 minute ago, Reonu said:

So it booted at CL15 3000 MHz. This is the timings screen in the BIOS. What else should I change?

Did you check my screenshots? My DRAM is from Micron right?

Yeah, your kit is Micron so I'm not sure where you'll end up.

 

And your BIOS doesn't give a ton of options.  Is that the latest BIOS?  

 

The timings I would try (in the order they appear in BIOS) @3000MHz 15-15-32-5-7-350.

 

Again, my strategy would be to set voltage as high as you're comfortable and then find what works after.  So if 1.4V is as high as you want to go then set that and see what speed and timings will run with that voltage.  Don't hunt and peck at voltage, speed, and timings all together.  

[nick name]

If you want to try 3200MHz then enter 16-18-Auto-Auto-Auto-Auto.  If that runs then we can work on something like 16-18-Auto-6-8-350.  If that runs then 16-18-Auto-5-7-350.  If that runs then 16-18-Auto-4-6-350.  If that runs then 16-18-Auto-4-6-300.  

[Reonu]

9 minutes ago, nick name said:

Yeah, your kit is Micron so I'm not sure where you'll end up.

 

And your BIOS doesn't give a ton of options.  Is that the latest BIOS?  

 

The timings I would try (in the order they appear in BIOS) @3000MHz 15-15-32-5-7-350.

 

Again, my strategy would be to set voltage as high as you're comfortable and then find what works after.  So if 1.4V is as high as you want to go then set that and see what speed and timings will run with that voltage.  Don't hunt and peck at voltage, speed, and timings all together.  

There are actually a lot of options, I recorded a video so that you can see them all. And yes my BIOS is updated. I don't want to go past 1.35V btw (that is my current voltage). Currently I'm at 1.35V, 3000 MHz, CL15 and the rest on auto. So the idea now would be to drop the timings a bit I guess.

 

[nick name]

https://www.overclock.net/forum/5-intel-cpus/1569364-official-intel-ddr4-24-7-memory-stability-thread.html

[aisle9]

8 hours ago, nick name said:

I've never seen anyone use XTU for memory testing.  There are more common programs used by RAM overclockers that I would recommend.  Some are paid and some are free.  The worst part of RAM overclocking is the amount of time it takes to truly test RAM for stability.  Most folks run the tests overnight.  

Once I'm to a point where I think I've found my numbers, I'll let it run for a much longer time. I like XTU because it specifically has a memory stress test, and XTU is basically just a toned-down Prime95 so as to avoid cooking your components if you turn it on and walk away. The real Prime95, OCCT and AIDA64 all do it just fine, but XTU gives me lower temps, and unlike pushing a high CPU overclock, I'm not looking to bake my RAM to a crisp.

 

8 hours ago, nick name said:

I'm not certain, but I always use matching timings as a quick indicator a kit might be Samsung b-die.  It holds true to about 3600MHz but there are some faster speed kits that follow as well.  14-14-14, 15-15-15, 16-16-16 and for faster than 3600MHz I've seen 17-17-17 and 18-18-18.  There are some kits like the Dominator Platinums that will use b-die and use 16-18-18 timings and those kits could also be something like Hynix CJR which can get pretty fast.  

Interesting. That throws out the rule I was told back when Ryzen was new and memory compatibility was suddenly a problem for the first time since DDR2 Optiplexes.

[Reonu]

Okay so I reached the conclusion that I can't get any meaningful OC at the latencies of the 2400 MHz XMP profile (15-15-35). The max I can do is 2500 MHz.

 

So now the question is: What is better? The following options appear to be stable:

a) 2400 MHz 15-15-35 1.2V (this is a copy and paste of the XMP profile specs)

b) 3000 MHz 16-18-36 1.28V

c) 3000 MHz 15-18-36 1.35V

 

I have yet to test if other options are stable, but out of these options which one should perform the best? Basically what is better, lower latency or higher speed?