Can I squeeze more out of 2x F4-3600C14D-16GVKA kits?

[SkyStreaker]

Gigabyte x570 UD rev 1.0 (Bios: F36e)
Ryzen 5800X
2x F4-3600C14D-16GVKA (Samsung B-die)
XFX Speedster QICK 308 6600 XT
be quiet! Pure Power 11 600W
be quiet! Pure Loop 240mm (push)
Corsair Scythe Kaze Flex 120mm PWM (1x pull back, 2x push front)
Fractal Meshify C

So, I hear good news about G.Skill (and my previous 3000's I could OC up to 3266 on standard profile, stable). Here's me wondering if I can get gain out of OC'ing the four sticks. Mind that my knowledge of this is marginal at best, and with a x570 chipset I can tune the living daylights out of everything of course.
Some research I could fathom is an OC to 4000 with some "roomier" timings and a slightly higher voltage would do. Ran a full memtest86 at "profile 1" speeds on defaulted BIOS had no problems, just FYI.

 

[tabascosauz]

• If you run just one of the kits, 4000 16-16-16 should be easily doable barring unforeseen complications from it being a low end board.
• If you run both kits (4x8GB), even booting 4000 might be a little optimistic as it is 2 ranks per channel just as 2x16GB Bdie is. The X570 UD is an average 4-layer 4DIMM board, and your average 4-layer board tops out somewhere between 3733-3866 on 2 ranks per channel. Generally would not count on even 1:1 3866 for 2x16 and 4x8 kits on 4-layer boards.

And even if you had a 6-layer board on which dual rank 4000 is not an issue, your 5800X needs to feel up to the task of 2000MHz IF stable and without WHEA errors. 5600X and 5800X have a much better chance at 4000 error-free than 5900X/5950X, but it's far from a guarantee.

If all goes well, 3800CL15 is plenty fast and should be relatively easy for a 3600CL14 kit, probably in the range of 1.4-1.5V. For 3800CL14, you're looking at 1.5-1.55V, and with 4 DIMMs that's going to be nigh-impossible to cool in RAM testing and games without setting up a small fan dedicated to blowing at the RAM, and still challenging with a fan.

 

[Ruru]

Hard to say. When I started OC'ing my HyperX 3200s, I went with a 33MHz bump and ran TestMem5 for several hours. I ended with 3466MHz at 1.38 Volts.

Try for yourself? We're here to help you

 

[SkyStreaker]

If it's a case of WHEA errors, then I think I'm already a bit in trouble. Got cache hierarchy errors when system goes in standby. Despite memtest86 results with 0 errors.

 

[tabascosauz]

If it's a case of WHEA errors, then I think I'm already a bit in trouble. Got cache hierarchy errors when system goes in standby. Despite memtest86 results with 0 errors.

At 3800+ we're talking about Bus/Interconnect WHEAs here. If you don't have any of those and you can boot and be stable otherwise then your CPU is doing fine on Infinity Fabric.

Cache hierarchy has nothing to do with RAM. Cache hierarchy means your cores aren't stable, in your case at idle, that's a separate problem. If you've messed with PBO and Curve Optimizer then you've got bad settings, if you're at stock settings then bad BIOS/bad board/bad CPU.

No one serious about RAM uses MT86, the CPU is literally at rest during that program. Use HCI Memtest, Karhu, or preferably Testmem5 on anta777Extreme1 config or 1usmusv3 config.

 

[phanbuey]

have you OC'd the tertiary timings? -- those get a ton of performance above and beyond what a frequency bump will do -- especially with 4 dimms.

primarily trrd_s / trrd_l /tfaw to [4 / 6 / 24] (loose) then [4 / 5 / 20] then [4 / 4 / 16] (super tight ) or 4 / 5 / 16

Try something like

trrd_s / trrd_l /tfaw - [4 / 6 / 24] (to start)

and:

tRDRD_sg = 7
tRDRD_dg = 4
tRDRD_dr = 7
tRDRD_dd = 7

tWRWR_sg = 7
tWRWR_dg = 4
tWRWR_dr = 7
tWRWR_dd = 7

And then run a test -- i think you will be surprised at the performance increase.

 

[SkyStreaker]

have you OC'd the tertiary timings? -- those get a ton of performance above and beyond what a frequency bump will do -- especially with 4 dimms.

primarily trrd_s / trrd_l /tfaw to [4 / 6 / 24] (loose) then [4 / 5 / 20] then [4 / 4 / 16] (super tight ) or 4 / 5 / 16

Try something like

trrd_s / trrd_l /tfaw - [4 / 6 / 24] (to start)

and:

tRDRD_sg = 7
tRDRD_dg = 4
tRDRD_dr = 7
tRDRD_dd = 7

tWRWR_sg = 7
tWRWR_dg = 4
tWRWR_dr = 7
tWRWR_dd = 7

And then run a test -- i think you will be surprised at the performance increase.

Eh...

I did say "marginally experienced".
I'll try all the suggestions when home.

 

[AlwaysHope]

At 3800+ we're talking about Bus/Interconnect WHEAs here. If you don't have any of those and you can boot and be stable otherwise then your CPU is doing fine on Infinity Fabric.

Cache hierarchy has nothing to do with RAM. Cache hierarchy means your cores aren't stable, in your case at idle, that's a separate problem. If you've messed with PBO and Curve Optimizer then you've got bad settings, if you're at stock settings then bad BIOS/bad board/bad CPU.

No one serious about RAM uses MT86, the CPU is literally at rest during that program. Use HCI Memtest, Karhu, or preferably Testmem5 on anta777Extreme1 config or 1usmusv3 config.

So if MT86 is not worthy, then how does one test ram before booting into OS? The authors of that program have done some nice clean ups with the user interface in v9.3

 

[tabascosauz]

So if MT86 is not worthy, then how does one test ram before booting into OS? The authors of that program have done some nice clean ups with the user interface in v9.3

You don't. There's no magical benefit to "testing outside the OS". You don't use your computer outside of the OS.

Even if there was an argument for testing literally 100.0% of memory capacity, and these theoretical instabilities only show at 100% capacity, it would be incredibly irrelevant in that there is no way you are filling 100.0% of memory capacity while using your computer normally without suffering major consequences.

If you pick up a brand new kit you can run it while doing something else so that you don't waste further time on a physically faulty kit, but that's pretty much it. TM5 automatically allocates nearly all available RAM and hammer it within an inch of its life. Especially when running more aggressive OCs, RAM doesn't exist in a vacuum; the Infinity Fabric needs to be happy, and so does the UMC (IMC for intel). All of these OC favourite in-OS memtests have appreciable CPU and memory controller load to some degree (TM5 probably the heaviest, and LinpackXtreme depending on config) - MT86 basically just has the CPU sit there.

Yes, Windows is sensitive to unstable memory, but it's also not that immediate as long as you are even remotely aware of what you're doing.

 

[SkyStreaker]

You don't. There's no magical benefit to "testing outside the OS". You don't use your computer outside of the OS.

Even if there was an argument for testing literally 100.0% of memory capacity, and these theoretical instabilities only show at 100% capacity, it would be incredibly irrelevant in that there is no way you are filling 100.0% of memory capacity while using your computer normally without suffering major consequences.

If you pick up a brand new kit you can run it while doing something else so that you don't waste further time on a physically faulty kit, but that's pretty much it. TM5 automatically allocates nearly all available RAM and hammer it within an inch of its life. Especially when running more aggressive OCs, RAM doesn't exist in a vacuum; the Infinity Fabric needs to be happy, and so does the UMC (IMC for intel). All of these OC favourite in-OS memtests have appreciable CPU and memory controller load to some degree (TM5 probably the heaviest, and LinpackXtreme depending on config) - MT86 basically just has the CPU sit there.

Yes, Windows is sensitive to unstable memory, but it's also not that immediate as long as you are even remotely aware of what you're doing.

Save Windows 2000. I remember installing 2 sticks back in the day and one was badly seated. Win95, 98 etc crashed at install. Windows 2000 went it's merily way. That one is stable as a rock.

 

[phanbuey]

TM5 for ram and Linpack for Cache OC (on intel) works wonders for me.

 

[Deleted member 215115]

tRDRD_sg = 7
tRDRD_dg = 4
tRDRD_dr = 7
tRDRD_dd = 7

tWRWR_sg = 7
tWRWR_dg = 4
tWRWR_dr = 7
tWRWR_dd = 7

I don't think you can mess with these on AM4. I've never seen them in any submenu.

 

[phanbuey]

I don't think you can mess with these on AM4. I've never seen them in any submenu.

It depends on the motherboard and what they’re called but all of them have a section for turn around timings. Might be labeled something else tho. The sg/dg/dr/dd at the end should give them away.

with 4x ranks tuning those should give some very nice boosts after he is able to find his max frequency.

1) get highest 1:1 ratio mhz,
2) tighten cl and primaries
3) tighten trrd(s)/l and tfaw
4) turnaround timings
5) last and most painful: trfc and trefi tuning
6) watch your cpu-bottlenecked fps go up by 20%

 

[SkyStreaker]

I don't think you can mess with these on AM4. I've never seen them in any submenu.

I've been through all of them, and if memory serves, I have these options and more.

 

[Deleted member 215115]

It depends on the motherboard and what they’re called but all of them have a section for turn around timings. Might be labeled something else tho. The sg/dg/dr/dd at the end should give them away.

with 4x ranks tuning those should give some very nice boosts after he is able to find his max frequency.

1) get highest 1:1 ratio mhz,
2) tighten cl and primaries
3) tighten trrd(s)/l and tfaw
4) turnaround timings
5) last and most painful: trfc and trefi tuning
6) watch your cpu-bottlenecked fps go up by 20%

I've been through all of them, and if memory serves, I have these options and more.

Ah yeah, my bad, these are always called something else for every manufacturer.

 

[SkyStreaker]

Just for arguments sake, what would be a proper board within a budget of about 250 Euro's?

 

[Deleted member 215115]

Just for arguments sake, what would be a proper board within a budget of about 250 Euro's?

There are none. Proper boards start at 300 EUR. ( MSI X570 Unify/Unify-X)

But then again, I've pushed B-die kits to 3800 CL14 on 100 EUR X470 boards so who knows anymore?

For Ryzen, the board doesn't matter much if your memory is great. 3800 is what you want and 99% of boards should be capable of that with some B-die.

B550/X570S boards will do better memory OC because they are newer. 2-DIMM boards will get you the best results.

 

[freeagent]

Proper boards start at

6 layers.

 

[Deleted member 215115]

6 layers.

CMOS reset button.

 

[freeagent]

CMOS reset button.

Just use your reset switch?

Edit:

You said a proper board starts at 300 euro.. I disagreed and said 6 layers..

 

[Deleted member 215115]

Just use your reset switch?

If only that worked

Or do you mean plugging the reset switch into the CMOS reset pins?

 

[freeagent]

If only that worked

Or do you mean plugging the reset switch into the CMOS reset pins?

Yes onto the pins

 

[Deleted member 215115]

Yes onto the pins

Alright, fair enough.

 

[SkyStreaker]

Yes onto the pins

GOOD ONE, I'm going to do that...

@tabascosauz -
3800CL15 seems a good start with 1.45v, but what timings are advised? 15-14-14-??? Unclear on that one.

 

[AlwaysHope]

You don't. There's no magical benefit to "testing outside the OS". You don't use your computer outside of the OS.

Even if there was an argument for testing literally 100.0% of memory capacity, and these theoretical instabilities only show at 100% capacity, it would be incredibly irrelevant in that there is no way you are filling 100.0% of memory capacity while using your computer normally without suffering major consequences.

If you pick up a brand new kit you can run it while doing something else so that you don't waste further time on a physically faulty kit, but that's pretty much it. TM5 automatically allocates nearly all available RAM and hammer it within an inch of its life. Especially when running more aggressive OCs, RAM doesn't exist in a vacuum; the Infinity Fabric needs to be happy, and so does the UMC (IMC for intel). All of these OC favourite in-OS memtests have appreciable CPU and memory controller load to some degree (TM5 probably the heaviest, and LinpackXtreme depending on config) - MT86 basically just has the CPU sit there.

Yes, Windows is sensitive to unstable memory, but it's also not that immediate as long as you are even remotely aware of what you're doing.

Thanks, I wasn't implying there is a magical benefit in testing outside of the OS but the thing is MT86 will check for physical integrity of the DIMMS & obviously before the OS loads. If the end user is attempting an OC on the ram they don't want to load the OS & get errors when its loading before the desktop completely loads. That's the whole point of MT86.