need advice lowering memory timings

[Morgoth]

so i plan to try to lower my memory timings so i need some guidance on this
dont tell me to google it.. i prefer to ask those who got expierance in this

what software do i need for stability testing?

what settings should i change? and what not?

im running Kingston Technology KSM32ED8/16HD, 16 GB, 2 x 16 GB, DDR4, 3200 MHz, 288-pin DIMM ECC cl 22

 

[ThrashZone]

Hi,
3200c16 is pretty limited
But you should look through the aida64 thread for similar memory post while you wait

Share your AIDA 64 cache and memory benchmark here

This thread is for Aida :) If you are gonna run Pi, at least do the 32m :)

www.techpowerup.com

 

[3x0]

Here's a good guide, everything explained https://github.com/integralfx/MemTestHelper/blob/oc-guide/DDR4 OC Guide.md
Good luck

Edit:

3200 MHz, 288-pin DIMM ECC cl 22

Yeah, I don't think you will have too much success overclocking RAM that supports ECC.

 

[Deleted member 185158]

so i plan to try to lower my memory timings so i need some guidance on this
dont tell me to google it.. i prefer to ask those who got expierance in this

what software do i need for stability testing?

what settings should i change? and what not?

im running Kingston Technology KSM32ED8/16HD, 16 GB, 2 x 16 GB, DDR4, 3200 MHz, 288-pin DIMM ECC cl 22

First take some screen shots of what you have got going on.

Cpuz memory and SPD tabs.
Zen timings.
Could do Taiphoon Burner and screen shot the Eeprom information as well

 

[A Computer Guy]

For quick testing I use passmark memtest86 but that won't catch a lot of other problems. Then I use OCCT.
There are some other programs people use for testing as well.
But you should also test against varying workloads.
Watch your error logs for WHEA errors. ECC corrections should show up in the logs if it happens.
Test your system running at idle and low workloads for spontaneous reboots.

One thing to test is for heat. If your workload generate a lot of ram heat you want to make sure to test for that so you know your overclock is stable at higher temps.

Personally I have found it more difficult to OC ECC (I'm not an expert) and have a terribly difficult time getting something worthwhile to run over DDR4-3200 but have had much more success simply raising voltages to reduce timings. For example from CL22 to CL16 to make it more like a typical CL16 kit. For me the primary timings, tCWL, and tRFC have been the easiest to improve but I don't have your model kit so YMMV.

 

[Morgoth]

First take some screen shots of what you have got going on.

Cpuz memory and SPD tabs.
Zen timings.
Could do Taiphoon Burner and screen shot the Eeprom information as well

i start with installing the reqeusted software "memtest86" "Taiphoon Burner "

 

[ThrashZone]

Hi,
Zen timings utility would also be nice to see.

 

[phanbuey]

The timings on DDR4 that made the biggest difference for me, and ones that I usually go for first after setting primaries are:

TREFI
TRFC

TRRD
TRRDL
TFAW

The the bank / dimm read latencies.

Those usually account for 80-90% of my gains, these alone can boost FPS in bandwidth bottlenecked scenarios by 20%. You can also into a rabbit hole with ram timings for that last 1% boost really easily.

 

[Morgoth]

Hi,
Zen timings utility would also be nice to see.

For quick testing I use passmark memtest86 but that won't catch a lot of other problems. Then I use OCCT.
There are some other programs people use for testing as well.
But you should also test against varying workloads.
Watch your error logs for WHEA errors. ECC corrections should show up in the logs if it happens.
Test your system running at idle and low workloads for spontaneous reboots.

One thing to test is for heat. If your workload generate a lot of ram heat you want to make sure to test for that so you know your overclock is stable at higher temps.

Personally I have found it more difficult to OC ECC (I'm not an expert) and have a terribly difficult time getting something worthwhile to run over DDR4-3200 but have had much more success simply raising voltages to reduce timings. For example from CL22 to CL16 to make it more like a typical CL16 kit. For me the primary timings, tCWL, and tRFC have been the easiest to improve but I don't have your model kit so YMMV.

my ram modules didt came with heatsinks should i invest into heatsinks first??
wats OCCT?
where can i find the log for error corrections in windows 10?

 

[ThrashZone]

Hi,
Yeah 3200c22 is ugly x6

 

[phanbuey]

View attachment 279404


my ram modules didt came with heatsinks should i invest into heatsinks first??
wats OCCT?
where can i find the log for error corrections in windows 10?

You don't need heatsinks. Those timings are brutal... you have a huge potential with these.

 

[A Computer Guy]

View attachment 279404


my ram modules didt came with heatsinks should i invest into heatsinks first??
wats OCCT?
where can i find the log for error corrections in windows 10?

Heat sink not required

OCCT link https://www.ocbase.com/

I create a custom view in windows event viewer

 

[ThrashZone]

Hi,
This is likely a very long shot at 1.35v
3200 MHz CL18-20-20-40

Make sure you create a system image before screwing with timings it can corrupt an os pretty easily.

 

[A Computer Guy]

Even running at CL22 with a remote session to cloud network and a 6 core windows VM running in the background and a browser with 20 tabs open my bench score is on par with CPU-Z.
Just saying you shouldn't expect dramatic gains in performance if you push it down to CL16 but you can improve CPU utilization with better timings.

When I first started ram OC I used to make a spreadsheet of timings in nanoseconds to help suggest timings at different frequencies and track my testing progress.
This is just to give you an idea, I don't expect these timings to work for you. Testing is a laborious process.
Taiphoon Burner will help give you the SPD chip values. Other values I just grab from BIOS/UEFI defaults or what's reported in ZenTimings.

 

[Morgoth]

huge list to choose from.. this is new for me
so i asume i should change CL?

and

TREFI
TRFC

TRRD
TRRDL
TFAW
?? lower by 1 value and test and repeat?

Here's a good guide, everything explained https://github.com/integralfx/MemTestHelper/blob/oc-guide/DDR4 OC Guide.md
Good luck

Edit:

Yeah, I don't think you will have too much success overclocking RAM that supports ECC.

btw i heard that ECC ram is more stable in overclocking than non ecc. not sure if its true or not


i start tomorow with making a fresh back up of my OS tomorow

 

[3x0]

btw i heard that ECC ram is more stable in overclocking than non ecc. not sure if its true or not

Memory chips that are used in ECC RAM don't have speed in mind but stability and error correction. Stable and slow.

 

[ThrashZone]

Hi,
Yeah they might throw up just using dimm at 1.35v pretty much normal for c18/ c16 primary timings.
I can't find any 3200c20 and it's primary timings they might still be 1.2v

Might add zen timings/... to the op

 

[Deleted member 185158]

I'd go for more frequency first...

Try dram frequency to 3466mhz and see if it posts up and test stable.

 

[A Computer Guy]

huge list to choose from.. this is new for me
so i asume i should change CL?

and

TREFI
TRFC

TRRD
TRRDL
TFAW
?? lower by 1 value and test and repeat?

I tend to start with common voltages first 1.35 DRAM and 1.1 SOC before changing frequency or timings but it may depend on your ram IC what voltages you should use.
Finding an existing example for your ram kit (or similar chipped kits) will save you a lot of time and at least might help you know up front what won't work and take a lot of guess work out of the process.

btw i heard that ECC ram is more stable in overclocking than non ecc. not sure if its true or not

ECC gives you reporting. So if you are on the edge of stability causing correctable errors you can find out easily. Uncorrectable errors will cause a reboot.

I've seen a tRFC chart posted in the forums here a few times. Maybe if someone who has it can repost it here that might be helpful.

 

[Morgoth]

this is after 17min 20gb testing not sure if the ram heats the same way up as a cpu and gpu doe's with this test
ambient tempreture of the room is 22c got the heater on i assume these tempratures are good for stock settings
is there a way to run memtest with out running the OS ? or other testing software for ram?

talking about freqenchy i tough ryzen 5 5900x doest support over 3200 mhz ram speeds??

 

[A Computer Guy]

That seems ok to me. If in doubt put a fan on it. For example when I did put a ridiculous heat spreader on my kit with active cooling this is what I got after stress testing for 1hr.

 

[S456]

talking about freqenchy i tough ryzen 5 5900x doest support over 3200 mhz ram speeds??

The CPU memory controller is only guaranteed to work up to 3200, and with JEDEC timings. However, most CPUs can exceed that and work with faster RAM and/or faster timings. Also read about keeping FCLK 1:1.

 

[A Computer Guy]

Since you don't have a heat spreader you have the benefit of seeing the IC numbers on your chips. This might help you search for OC results using those chips.

 

[trikats]

Since you don't have a heat spreader you have the benefit of seeing the IC numbers on your chips. This might help you search for OC results using those chips.

How many correctables are you getting per month, etc?
Gskill doesn't have advance / cross RMA so I decided to jump on the ECC bandwagon.