tuning fclock on ryzen

[rippy4500]

Is it better to keep fclock at half ram speed or to push it as far as possible seperately from ram speed?

some people recommend setting it to half while others suggest pushing it as far as possible

 

ive kept it at 1600mhz for a while now since my ram speed is 3200, but ive tried pushing it further and got 1900mhz which feels faster.

i could get it further and it wasnt unstable but i noticed strange behavior like stuttering, audio popping and the desktop randomly being really slow.

 

should i keep it at 1900 or 1600?

i have r5 5600x.

 

edit: i just realized this is kind of a dumb question, i should just keep it at the faster one.

[Bombastinator]

The definition of “stable” seems to be different for different people.  That wouldn’t be my definition of stable. 

[RONOTHAN##]

Ryzen 5000 series CPUs overall tend to have issues running FCLK above 1900MHz. There are a couple chips that can do more (5800Xs and 5600Xs are more likely to do better since they're single CCD, though it's still not common), but for the most part you do get negative scaling past 1900MHz because there is ECC built into the CPU and you have to utilize it a lot to run higher speeds. In Y-Cruncher 1B, for example, my 5900X will do 26 seconds with tuned B die at 3800MHz (1900FCLK). With slightly loosened timings at 3866MHz or 1933MHz FCLK (a frequency this kit is rated for) it will take over 5 minutes to run that same test. 

 

Just make sure that you try and get the RAM kit and the FCLK to run at the same frequency since running them out of sync has a pretty big latency disadvantage. For some workloads it doesn't really matter and the extra multicore performance will help, but in some stuff it will hurt performance. If you memory kit, for example, will only do 3600MHz, running 1800FCLK would likely be the best of both worlds. If you've got a good RAM kit that does 3800MHz though, run that. 

[Tegneren]

If I remember correctly, there was some performance gain in some cases on earlier Ryzen at least, but nothing major. It all comes down to stability, do some benchmarking and check.

You can also try just OC'ing the RAM. I also have 3200 MHz RAM (G-Skill FlairX 4x8GB CL14) that I'm running at 3600 CL16. Completely stable.

[rippy4500]

9 minutes ago, RONOTHAN## said:

Ryzen 5000 series CPUs overall tend to have issues running FCLK above 1900MHz. There are a couple chips that can do more (5800Xs and 5600Xs are more likely to do better since they're single CCD, though it's still not common), but for the most part you do get negative scaling past 1900MHz because there is ECC built into the CPU and you have to utilize it a lot to run higher speeds. In Y-Cruncher 1B, for example, my 5900X will do 26 seconds with tuned B die at 3800MHz (1900FCLK). With slightly loosened timings at 3866MHz or 1933MHz FCLK (a frequency this kit is rated for) it will take over 5 minutes to run that same test. 

 

Just make sure that you try and get the RAM kit and the FCLK to run at the same frequency since running them out of sync has a pretty big latency disadvantage. For some workloads it doesn't really matter and the extra multicore performance will help, but in some stuff it will hurt performance. If you memory kit, for example, will only do 3600MHz, running 1800FCLK would likely be the best of both worlds. If you've got a good RAM kit that does 3800MHz though, run that. 

im currently running 3200 cl14, so getting it to 3800 would require me to greatly reduce timings, should i go for 3800 and then reduce timings from there?

[Bombastinator]

8 minutes ago, rippy4500 said:

im currently running 3200 cl14, so getting it to 3800 would require me to greatly reduce timings, should i go for 3800 and then reduce timings from there?

The SOP as I understand it is to do neither, but start from the next highest speed and find out which timings work, if any.

OCing memory is famous for being a laborious process.  Whether you particular set of memory will even do 3400 is unknown.  It might.  It might not. Traditionally jumping to the highest possible number is more or less a guarantee of failure. For one we don’t even know what memory you have and while the pc number of the thing might give a rough idea of what kind of timings would be needed to get to that speed it still might not do it at all.  Part of the issue is heat sinks today seem to be used by manufacturers to hide the quality of the components used.  

[freeagent]

You could try 3600 flat 14s with 1.45v, 3800 and flat 15s should be just as easy, maybe with a little more voltage, 1.475-1.5v. You will have to play with it them to see what they can do.

[RONOTHAN##]

42 minutes ago, rippy4500 said:

im currently running 3200 cl14, so getting it to 3800 would require me to greatly reduce timings, should i go for 3800 and then reduce timings from there?

Yeah, loosen out the timings and go for as close to 3800mhz as you can (the last kit of 3200c14 I owned maxed out at 3733 with the voltage I was comfortable using, 1.5v). 3800mhz CL16 should be fairly achievable, and with the way FCLK works, even though it might have slightly worse calculated latency compared to 3600mhz cl14, it will actually be better with 3800c16.

 

Also, don't be afraid to tune your subtimings, they can offer fairly large performance improvements. 

[rippy4500]

17 minutes ago, RONOTHAN## said:

Yeah, loosen out the timings and go for as close to 3800mhz as you can (the last kit of 3200c14 I owned maxed out at 3733 with the voltage I was comfortable using, 1.5v). 3800mhz CL16 should be fairly achievable, and with the way FCLK works, even though it might have slightly worse calculated latency compared to 3600mhz cl14, it will actually be better with 3800c16.

 

Also, don't be afraid to tune your subtimings, they can offer fairly large performance improvements. 

my ram was actually 3600 cl18 with stock xmp, i just liked how 3200 cl14 felt so i stuck with that. its not samsung b die so idk if i can do 3800 cl16, or 3600 cl14. i remember 3600 cl15 being possible for mine tho and also 3200 cl13, but not anymore because i dont remember what timings i used. 3733 and 3800 worked but idk at what timings.

[RONOTHAN##]

Just now, rippy4500 said:

my ram was actually 3600 cl18 with stock xmp, i just liked how 3200 cl14 felt so i stuck with that. its not samsung b die so idk if i can do 3800 cl16, or 3600 cl14. i remember 3600 cl15 being possible for mine tho and also 3200 cl13, but not anymore because i dont remember what timings i used. 3733 and 3800 worked but idk at what timings.

Still, most RAM kits aren't on the edge enough that you can't raise the memory frequency a little bit. I woould go up to 3800MHz and see how tight you can get the timings while still being stable, it should be the best performance. As long as you don't need to loosen the timings out to like 3800CL24 to be stable, it should give you the best performance. 

[rippy4500]

9 minutes ago, RONOTHAN## said:

Still, most RAM kits aren't on the edge enough that you can't raise the memory frequency a little bit. I woould go up to 3800MHz and see how tight you can get the timings while still being stable, it should be the best performance. As long as you don't need to loosen the timings out to like 3800CL24 to be stable, it should give you the best performance. 

ok with some quick tuning i got 3800 cl17, cl16 boot looped

 

edit: i left the voltage at the 1.46 it was already at, could i increase it to 1.5 or would that decrease the lifespan of my components? (1.5 is the max my mobo will let me set)

[Bombastinator]

44 minutes ago, RONOTHAN## said:

Yeah, loosen out the timings and go for as close to 3800mhz as you can (the last kit of 3200c14 I owned maxed out at 3733 with the voltage I was comfortable using, 1.5v). 3800mhz CL16 should be fairly achievable, and with the way FCLK works, even though it might have slightly worse calculated latency compared to 3600mhz cl14, it will actually be better with 3800c16.

 

Also, don't be afraid to tune your subtimings, they can offer fairly large performance improvements. 

3200 cl 14 is pretty good memory.  You likely have a shot.  3600 cl16 seems possible.  3800 would also need an OC of the infinity fabric to push it high enough that 3800 memory wouldn’t drop the thing to 2:1 from 1:1.

[RONOTHAN##]

1 minute ago, rippy4500 said:

ok with some quick tuning i got 3800 cl17, cl16 boot looped

 

edit: i left the voltage at the 1.46 it was already at, could i increase it to 1.5 or would that decrease the lifespan of my components? (1.5 is the max my mobo will let me set)

It depends on the memory chips that are present on the memory stick you have. If you've got Micron 16Gb Rev. B, Micron 8Gb Rev. E, Samsung B die, Hynix DJR, and a couple others, 1.5V is the low end of what people use for overclocking. If you've got Hynix CJR, Samsung C die, and most other memory chips, running at 1.46V is probably a bit too high for daily voltage. 

 

There are 2 ways to tell what memory chips you have. 

1. If you have Corsair memory, this is the easiest. On the sticker with the model numbers, there's a version number, something like v4.31 or v3.41. There's a way to decode that, but the easiest way is to just google "corsair ddr4 vX.XX" and see what memory chip comes up. 
2. If you have anything else, download Thaiphoon Burner and use it to read the SPD information. Unfortunately, this isn't perfect since some manufacturers don't properly program the SPD and thus it doesn't report the memory chip, but this is where everything should show up. 

Spoiler

 

Also, because of the way AMD's memory controller works, by default you cannot run odd numbered CAS latency and a couple other timings. It'll let you input them and it will boot, but it will just bump it up a tick so in reality, 3800MHz CL17 is really running at 3800MHz CL18. You can get it to work with odd CAS Latency by disabling Gear Down Mode (sometimes, there are boards like the Unify X where it is broken), but that's more effort than it's generally worth to get 1 tick lower on your latency. 

[rippy4500]

4 minutes ago, RONOTHAN## said:

It depends on the memory chips that are present on the memory stick you have. If you've got Micron 16Gb Rev. B, Micron 8Gb Rev. E, Samsung B die, Hynix DJR, and a couple others, 1.5V is the low end of what people use for overclocking. If you've got Hynix CJR, Samsung C die, and most other memory chips, running at 1.46V is probably a bit too high for daily voltage. 

 

There are 2 ways to tell what memory chips you have. 

1. If you have Corsair memory, this is the easiest. On the sticker with the model numbers, there's a version number, something like v4.31 or v3.41. There's a way to decode that, but the easiest way is to just google "corsair ddr4 vX.XX" and see what memory chip comes up. 
2. If you have anything else, download Thaiphoon Burner and use it to read the SPD information. Unfortunately, this isn't perfect since some manufacturers don't properly program the SPD and thus it doesn't report the memory chip, but this is where everything should show up. 

  Reveal hidden contents

 

Also, because of the way AMD's memory controller works, by default you cannot run odd numbered CAS latency and a couple other timings. It'll let you input them and it will boot, but it will just bump it up a tick so in reality, 3800MHz CL17 is really running at 3800MHz CL18. You can get it to work with odd CAS Latency by disabling Gear Down Mode (sometimes, there are boards like the Unify X where it is broken), but that's more effort than it's generally worth to get 1 tick lower on your latency. 

its samsung c die, and gskill trident z rgb, gear down mode is off.

edit: wouldnt high voltage be fine as long as the temps are low

[RONOTHAN##]

2 minutes ago, Bombastinator said:

3200 cl 14 is pretty good memory.  You likely have a shot.  3600 cl16 seems possible.  3800 would also need an OC of the infinity fabric to push it high enough that 3800 memory wouldn’t drop the thing to 2:1 from 1:1.

Yeah, that kit would do 3800MHz if I upped it to 1.65V, I just didn't particularly want to leave it at that voltage for more than just running a benchmark or two. 

 

besides, OP is using a kit of 3600MHz CL18 that he just downclocked to 3200CL14. I would be very surprised if it can't do 3800MHz CL18. 

[rippy4500]

1 minute ago, RONOTHAN## said:

Yeah, that kit would do 3800MHz if I upped it to 1.65V, I just didn't particularly want to leave it at that voltage for more than just running a benchmark or two. 

 

besides, OP is using a kit of 3600MHz CL18 that he just downclocked to 3200CL14. I would be very surprised if it can't do 3800MHz CL18. 

I got it to 3800 cl17 with gear down off. i got 3200 cl13 and 3600 cl15 in the past but i dont remember the timings for those.

[RONOTHAN##]

4 minutes ago, rippy4500 said:

its samsung c die, and gskill trident z rgb, gear down mode is off.

OK, Samsung C die is weird. That tends to get very unstable when you add more than 1.4V to it and generally can't do more than 3600MHz CL16, plus it also tends to sweet spot somewhere between 1.3 and 1.4V. If you have a kit that does 1.46V stable, you hit the jackpot and have the weirdest Samsung C die I've ever seen. 

 

Either that or something is just misreporting it and it's actually Samsung B die. 

[rippy4500]

5 minutes ago, RONOTHAN## said:

OK, Samsung C die is weird. That tends to get very unstable when you add more than 1.4V to it and generally can't do more than 3600MHz CL16, plus it also tends to sweet spot somewhere between 1.3 and 1.4V. If you have a kit that does 1.46V stable, you hit the jackpot and have the weirdest Samsung C die I've ever seen. 

 

Either that or something is just misreporting it and it's actually Samsung B die. 

i set it to 1.5 out of curiosity and it boots up just fine, is that unusual?

edit: check hwinfo to see temps and its in the low 40s for my ram, is that bad?

[RONOTHAN##]

Just now, rippy4500 said:

i set it to 1.5 out of curiosity and it boots up just fine, is that unusual?

Yes. 

 

 

 

[rippy4500]

3 minutes ago, RONOTHAN## said:

Yes. 

 

 

 

so is it actually b die or do i just have the perfect kit of c die?

[Bombastinator]

1 minute ago, rippy4500 said:

so is it actually b die or do i just have the perfect kit of c die?

The only thing I know to do To find out you can either download that app mentioned and hope the manufacturer filled out the numbers right (which I didn’t know before) or the puppy method: turn it over and look.  This would Involve removing the heat sink which could be quite a production. 

[RONOTHAN##]

1 minute ago, rippy4500 said:

so is it actually b die or do i just have the perfect kit of c die?

I honestly don't know, it could be either one. I'd say it's more likely to be B die, just cause every kit of C die I've seen just craps itself above 1.4V, but anything is possible. If it didn't blue screen 1.5V when you tried to do some sort of benchmark that touches the memory, I would be extremely impressed if it was C die. 

[RONOTHAN##]

3 minutes ago, Bombastinator said:

or the puppy method: turn it over and look.  This would Involve removing the heat sink which could be quite a production. 

AFAIK G.Skill gets the chip manufacturers to put their logo on it instead of their own, so even that isn't guaranteed to work. 

[freeagent]

Set 3800 flat 14s 1.5v and if it boots it’s bdie lol.

[rippy4500]

50 minutes ago, freeagent said:

Set 3800 flat 14s 1.5v and if it boots it’s bdie lol.

what does 14s mean? did you mean cl14?

i was only able to get 3800 cl16 at 1.5. cl15 or less didnt work.

 

1 hour ago, RONOTHAN## said:

AFAIK G.Skill gets the chip manufacturers to put their logo on it instead of their own, so even that isn't guaranteed to work. 

Ive been trying to tune the impedances and procodt, ik they help stability but have no idea to test which value works best. rn i have odt: 40ohm, and the other ones at 40,20,24,24

i didnt set them back to auto when i entered 3800 cl16. maybe that can get me to cl15 or even cl14