Overclocking guide:

- Do not use IMLC/AIDA and similar tools if you have a normal platform (without defects/problems at stock, etc.)

- Use JEDEC rules to avoid ECC errors

- Disable Self Refresh in BIOS

COOLING:
--------
The best cooling currently is water cooling. Among worse options is cooling from Alseye (or similar). The minimum cooling option is a 120/140mm fan on the GPU backplate.

DDR4 INTEL OVERCLOCKING:
------------------------
First, you need to find out what kit you have by installing Thaiphoon Burner. Look at the chip, then check XMP in the editor tab.

Then you should make sure everything is fine with your memory by testing with Karhu 10k% + OCCT GPU test 8 shaders. If your temperatures exceed 40-45 degrees (higher temperature = higher resistance => worse signal integrity), and if you have errors even at stock, I don't recommend doing anything. If everything is normal, continue.

I also recommend finding out which version of JEDEC you have - criteria may vary, as your overclocking depends on this standard and when your kit was created.

After you've identified your chips, you can set voltages.
I personally recommend setting maximum voltages first, then lowering them to the limit.

You can start with 1.45 - 1.55 DRAM Voltage

On K-series processors you can change SA IO Voltage. I recommend starting at 1.35 and decreasing to the limit, but in some rare cases, high starting voltage can make things worse, so you can also try setting 1.2, 1.25

VDDQ I recommend starting at 1.4 and going down

DRAM Reference Clock - Set to 133, this will help you achieve better frequencies without changing voltage. On 12-14 gen it makes sense to set 100 (since in many cases people had higher frequencies with multiplier 100)

Gear - Set gear 1

Command Rate - 1,2,3 I recommend testing this parameter with Y-Cruncher/Prime large fft + Furmark

Next are primary timings, their changes depend on voltage scaling, which can vary on different chips.

Tcl - As low as possible
trcd - As low as possible
trp - As low as possible, ideally should equal trcd
tras - I recommend setting it higher than trp + trcd, as low as possible

Secondary timings:

trc - Does not exist on Intel (fake timing)
Twr - Don't touch, set it through twrpre.
tWRPRE - 4 + tWR + tCWL - Set as low as possible, twr >15
Trtp - Don't touch, set it through tRDPRe
tRDPRE = tRTP - Set as low as possible, I don't recommend less than 8 on any chips and frequencies
tfaw - Any value greater than 15
trrds - Any value greater than 3
trrdl - Any value greater than 4
twrtl - Set through tWRRD_sg
twrrd_sg - 6 + tcwl + twtrl
twtrl - Should be 2 less than trtp or greater by any value
twtrs - Set through twrrd_dg
Twrrd_dg - 6 + tclwl + twtrs
twtrs - Any number greater than 3
Trefi - Any value, depends on your temperatures. After a certain value, trefi/trfc lose meaning
trfc - Any value divisible by 16. To calculate it in ns, you need to use this calculator, I don't recommend setting less than 120-140ns. Use this table
Tccd - Greater than 3
TCCDL - Greater than 5
trdwr sg and dg - I recommend setting greater than 9
twrrd_sg - 6 + tcwl + twtrl
twwrd_dg - 6 + tcwl + twtrs
trefix9 - Any value up to 127 (up to 11th gen) and 255 (12+ gen) or you can use the formula trefix9 = (trefi + 1):1024

RTL IOL:
--------
You can configure them up to 11th gen Intel. You need to disable Turn Around Timing Optimization and enable Round Trip Latency. You can also experiment with ODT Training, DFE Training. Disable Fast Boot or set it to Slow Training, and after overclocking you can return it.

At high frequencies up to 11th gen, you can try RTL IOL manually on ASUS and MSI boards. To configure it on Gigabyte, experiment with Memory Enhancement Settings. Also on Gigabyte disable SA GV, Dynamic Memory Boost, Realtime Frequency, set Memory Timing Mode to Fixed.

RTL should equal IOL + IOL Offset <= 28
RTL Init set = 2*CL + (35 - 38)
IOL Offset 21-21
RFR Delay 14-14 (For the channels you're using)

Then experiment with values and remember that RTLs and IOLs should not differ from each other by more than 1 value.

RTT Configuration:
------------------
Try all values (Main thing is RTT Wr should be highest), while lowering DRAM Voltage. For testing use GSAT from a flash drive.

Also you should disable PPD (Precharge Power Down) through BIOS or through MemTweakIt

If you can't disable PPD, lower txp, tcke, txpdll

Testing:
--------
Karhu 20000%, testmem5 test absolute, extreme, pcbdestroyer, universal, testmem pro 10 hours, MemTest86 V11.0 Pro, Y-Cruncher vst/vt3/fft/n64/hnt/c19 5+ hours, Prime large fft/blend - 20 hours, Linpack/Linx (Watch how much flops differ, if more than 4-10, it's bad), OCCT sse/avx2 large variable

In parallel with all these tests, run a GPU test (Furmark\Kombustor\OCCT gpu test 8 shader)

DDR4 AM4 OVERCLOCKING:
----------------------
FCLK = frequency/2
Mem Mode/UCLK DIV1 Mode - uclk = MEMCLK or UCLK = MEMCLK/2 at high frequencies (4400+)

Set SOC Voltage 1.08 - 1.15 and then lower it
VDDG - 1.0-1.07
VDDP - 0.95-1.0
DRAM Voltage - 1.37-1.55 depending on your cooling

Disable PPD and Geardown Mode

To disable geardown you need to:
Set CAD Bus - 60/40 20 20 24
CAD Bus Setup - 56 0 0 or 63 0 0 or 56 56 56, or 63 63 63

Always set maximum frequency

Set primary timings as low as possible

When setting tcl, watch PHYRDL, it should always be 26
tras - ignored on AMD, but I recommend setting it according to JEDEC (>trp + trcd)
twr - Any value >11 (I recommend >15)
trtp - >5 (I recommend >7)
tfaw - Any value >15
trrds - 4/5
trrdl - >4
twtrs - >3
trfc - Any value divisible by 16. To calculate it in ns, you need to use this calculator, I don't recommend setting less than 120-140ns. Also use this table
twrtl >5
trdwr - >9
twwrd - 1-4
tcwl - -1 from tcl / -2 from tcl (even)
trdrdscl - >3
trdrdscl - >3
Command Rate - 1
trc - Any value >28 (trp + tras)

Resistances:
------------
For two 8GB sticks - Disabled/Off/RZQ5
For two-four 16+GB sticks RZQ7/RZQ3/RZQ1
For four 8GB sticks RZQ7/Off/RZQ5
ProcODT - I recommend setting minimum to reduce memory heating

Testing:
--------
Karhu 20000%, testmem5 test absolute, extreme, pcbdestroyer, universal, testmem pro 10 hours, MemTest86 V11.0 Pro, Y-Cruncher vst/vt3/fft/n64/hnt/c19 5+ hours, Prime95 large fft/blend -20 hours, Linpack/Linx 8/10+gb, large number of cycles (watch how much flops differ, if more than 4-10 = bad), OCCT sse/avx2 large variable, ram test

In parallel with all these tests run a GPU test (Furmark\Kombustor\OCCT gpu test 8 shader)

Open HWInfo and scroll to the very bottom and watch your WHEA during testing

DDR5 INTEL OVERCLOCKING:
------------------------
Find out what memory chips you have, based on this you'll need to do overclocking

Preferably set DRAM Reference Clock 133, but possibly in your case this will give poor results and then set 100

On 4-DIMM boards maximum frequency is 6933

Set Gear Mode - Gear 2, but for very high frequencies set Gear 4

Then disable Memory Fast Boot, or set it to Slow Training

Disable Lucky Mode

On Gigabyte you can test Memory Enhancement Settings

Enable Round Trip Latency and disable Turn Around Timing Training. For frequencies above 7000 I would recommend experimenting with Late Command Training, Early Command Training, ODT Training, DIMM DFE Training, VDDQ Training

Voltages:
---------
I recommend setting DRAM VDDQ up to 1.5, then lowering it to minimum stable value
DRAM VDD - same as with vddq
CPU VDDQ TX - up to 1.4, lower to minimum stable value
CPU SA - set up to 1.35, then lower it to minimum working stable value
CPU VDD2 - set up to 1.45 then lower it to minimum working stable value

Also you can enable high voltage mode (some boards have problems with it)

Timings:
--------
Tcl - Always even, lower it to minimum working value
Tcwl - 2 less than tcl
Trcd - Any value, lower it to minimum working value, should equal trp
Trp - Any value, lower it to minimum working value
tras - Calculate by formula trp + trtp
Twr - I recommend setting through twrpre - minimum working value twr according to JEDEC is 48, increase by 6 if unstable
Twrpre - Calculate by formula tWR + tCWL + 8
trtp - I recommend setting through tRDPRE - minimum working value according to JEDEC is 12(18), preferably even, any value
trdpre = trtp
tfaw - 32
trrds - 8
trrdl - 8/12
trfc - Any value divisible by 16. To calculate it in ns, you need to use this calculator and table. On some boards you'll need to set it through Granularity Mode
trfc2 - Minimum working value divisible by 16 and less than trfc
trefi - Any value, depends on temperature
trefix9 - Any value up to 255, or you can use the formula trefix9 = (trefi+1):1024
tccdl - Any value greater than 15
tccd - 8
twtrl - I recommend setting through twrrd_sg, any value greater than 15
Twrrd_sg - tcwl + twtrl + 6
twtrs - I recommend setting through twrrd_dg, minimum value 4
Twrrd_dg - tcwl + twtrs + 6
trdwr sg/dg - I recommend setting both timings even, they should equal each other, minimum value 16
twrwr_sg - tccdl*2
Twrwr_dg - tccd

Try to achieve cr1
PPD - 0 / DISABLED

Resistances:
------------
I recommend trying absolutely all values and along with this lowering voltage + passing testing

Testing:
--------
Karhu 20000%, testmem5 set channel 1 in config, testmem pro 10 hours, MemTest86 V11.0 Pro, Y-Cruncher all tests 10-30 hours, Prime large fft\blend 20 hours, OCCT avx2/sse large variable, ram test

In parallel with all these tests run a GPU test (Furmark\Kombustor\OCCT gpu test 8 shader)

DDR5 AM5 OVERCLOCKING:
----------------------
Disable Memory Fast Boot
DFE READ TRAININGS - ENABLE
Set PPD - Disabled (try it, but most likely you'll get instability)
Set Geardown Mode disabled
Disable Nitro Mode, or enable it and select values that add minimum latency (for example: 1 2 0 / 1 3 1 / 1 2 1 etc.) to achieve higher frequencies.

Mem Mode/UCLK DIV1 Mode - UCLK = MEMCLK or UCLK = MEMCLK/2 at high frequencies (7600+)

8000>7800>6400(1|1) Memory Frequency - Maximum working + stable value

FCLK - Information, video

BCLK Frequency - 100, or you can do a small bus overclock thereby increasing memory frequencies

[IMAGE 3 PLACEMENT: UCLK/SOC voltage scaling guide]
If you can't get UCLK to work at 1.3V SOC, it's just not gonna work and you should give up.
Really bad CPUs need more than 1.2VSOC for 3000MHz UCLK
Really good CPUs run 3300MHz at less than 1.3VSOC
100MHz more UCLK usually needs 100+mv more
Example:
1.05VSOC >> 3000MHz UCLK stable
~1.15VSOC >> 3100MHz UCLK stable
~1.25VSOC >> 3200MHz UCLK stable
1.3VSOC >> 3300MHz UCLK UNSTABLE

On 9000, you can configure PMU Trainings, set to 0: PHY vrefdac0 P0, PHY vrefdac1 P0, PMU dqvref P0, select stable value for ardptrinitval P0

Voltages:
---------
DRAM VDD - Up to 1.6 and lower to minimum working stable value
DRAM VDDQ - Up to 1.65 and lower to minimum working stable value
CPU SOC - (FCLK depends on it) - up to 1.35 and lower to minimum working stable value. By setting VDDG (900-1000 mv), VDDP (1.05-1.15), you can achieve the same results as with high SOC, so I recommend lowering it later
CPU VDDIO - up to 1.4 and lower to minimum working stable value
MVDD - 1.65
MVDDQ - 1.45

Also you can enable high voltage mode (some boards have problems with it)

Resistances:
------------
Combine different variations and lower voltage/increase frequency

For RTT Wr RTT Park RTT Dqs normal is 40 48 40
RTT Nom is not used for single rank

If resistances are strangely set to Auto, that's bad. For example, from my observations on some boards it's advisable to enable EXPO and set settings yourself for correct operation (However your EXPO may be unstable due to its settings)

In many cases, what you set in resistances won't display correctly (for example, in ZenTimings)

Lower ProcODT

Timing Configuration:
--------------------
tcl - Even value, lower to minimum working stable value
trp = trcd - Lower to minimum working value
tras - 28(30) (since it's ignored), or set by formula trcd + 12 (recommended)
trefi - Any value, higher is better, but remember about temperatures
trfc - Any value divisible by 16. To calculate it in ns, you need to use this calculator and table
twr - Minimum value 48, increase by 6 in case of instability
trtp - Minimum value 12, any number, preferably even
twtrl - Any value, preferably even, minimum value 16 (I recommend setting 24)
twtrs - Any value, preferably even, minimum value 4 (I recommend setting 12)
trdwr - Any even value greater than 15
trc - tras + trcd
tfaw - 32
trrds - 8
trrdl - 8
twrwrscl - Any value greater than 7
trdrdscl - Any value greater than 7
twrrd - Set 4 or higher value (if you have instability), or 1 depending on rank

Testing:
--------
Karhu 20000%, testmem5 (for x3d use special configs) set channel 1 in config, testmem pro 10 hours, MemTest86 V11.0 Pro, Y-Cruncher all tests 10-30 hours, Prime large fft, blend-20 hours, Linpack/Linx 8/10+gb, large number of cycles (watch how much flops differ, if more than 4-10, it's bad), OCCT sse/avx2 large variable, ram test

In parallel with all these tests run a GPU test (Furmark\Kombustor\OCCT gpu test 8 shader)

INTEL CPU OVERCLOCKING:
-----------------------
Disable E-Cores
Set P-Cores Ratio/CPU Clock Ratio to desired value
Package Power Limit, Core Current Limit I recommend expanding if you have good temperatures, if not, you can configure them
Disable everything related to TVB
Set Max and Min Ratio to one desired value
Vcore Voltage Mode set to Manual or Override Mode, or Fixed and set desired value (1.2-1.5 depending on cooling, at stock the processor has random voltage), but remember about LLC (Best is where minimum Vdrop)
Test Current Capability at maximum/pre-maximum value
CPU VRM Switching Frequency set to Manual and set Fixed CPU VRM Switching Frequency to maximum value (If you have many power phases and generally good board, as well as good temperatures)
CPU Power Duty Control - set to Extreme (If you have many power phases and generally good board, as well as good temperatures)
CPU Power Phase Control - set to Extreme (If you have many power phases and generally good board, as well as good temperatures)

Testing:
--------
Y-Cruncher, Prime95 blend, OCCT avx2/sse large variable, Geekbench, Cinebench23, Linpack

TVB OC:
-------
Enable everything related to TVB
Configure all TVB Point Temp, Ratio Offset based on temperatures
Set desired frequencies
Don't disable any functions related to processor power/power saving (Like C-States)

Alternative Overclocking:
------------------------
Disable Voltage Reduction Initiated TVB and Enhanced TVB
Set Active Turbo Ratio to Manual, setting desired values below
Do the same with Turbo Per Core Limit Control
Set Vcore Voltage Mode to Adaptive Vcore
VF Offset Mode to Selection
Set Internal CPU Vcore to desired value
Adjust all other VF Point Offset based on temperatures

AMD CPU OVERCLOCKING:
---------------------
CPU Core Ratio set to desired value
Disable Core Voltage Suspension
CPU Vcore set to desired value (1.2-1.5, depending on cooling, at stock the processor has random voltage), but remember about LLC (Best is where minimum Vdrop)
CPU Power Duty Control - set to Extreme (If you have many power phases and generally good board, as well as good temperatures)
CPU Power Phase Control - set to Extreme (If you have many power phases and generally good board, as well as good temperatures)
Test Current Capability at maximum/pre-maximum value
CPU VRM Switching Frequency set to Manual and set Fixed CPU VRM Switching Frequency to maximum value (If you have many power phases and generally good board, as well as good temperatures)

Testing:
--------
Y-Cruncher all tests, Prime95 blend, OCCT sse/avx2 large variable, Geekbench, Cinebench23, Linpack

PBO (PRECISION BOOST OVERDRIVE):
---------------------------------
I recommend this configuration method if you have temperature problems

Don't disable any functions related to processor power/power saving (like PSS Support, C-States) for PBO configuration

Enable Precision Boost Overdrive
Configure limits based on stock values in PBO2tuner. Limits cannot be configured on X3D processors, also PBO2tuner is unavailable on ZEN4

Also for PBO configuration you can use Hydra and read its manual

Precision Boost Override Scalar set to Manual and set to either 1 or 2
CPU Boost Clock Override set to Enabled Positive
Max CPU Boost Clock Override set to desired value

Run BoostTester and open HWinfo, look at maximum boost frequencies in it and then, after finding out at what maximum value there is boost, set this value

Curve Optimizer set to Per Core

Then from OS in PBO2tuner/Hydra set value for each core, for example -30 is Core Curve Optimizer Sign - Negative and Core Curve Optimizer Magnitude at value 30

Testing, select values for each core, then set them in BIOS

Choose LLC as in the instruction

Also use the power plan from this instruction

Testing:
--------
Y-Cruncher all tests, Prime95 blend, OCCT sse/avx2 large variable, Geekbench, Cinebench23, Linpack, Hydra, CoreCycler avx2/avx/sse

During testing I recommend enabling WHEA-Logger (Open "Event Viewer" right-click on "Custom Views" and click "Create Custom View". Check "Warning" and "Errors" and choose "By source", set Event source to "WHEA-Logger" and click "OK" then "OK" again and watch these errors during testing.

GPU OVERCLOCKING (NVIDIA):
--------------------------
You'll need MSI Afterburner

Set desired Power limits, I recommend setting them to maximum
Go to settings in Fan and turn all sliders to maximum
In General tab check: Unlock Voltage Control, Unlock Voltage Monitoring, Force Constant Voltage
Check for available product update set to "never"

Increase Core Voltage, Core Clock, Memory Clock to maximum possible values

During testing watch values in HWInfo, as in benchmarks they may be fake (Frequency values + voltages)

Testing:
--------
3dmark, Memtest Vulcan, Superposition, Heaven, Furmark, Kombustor, OCCT GPU Test (8 shaders)/VRAM Test (Set 95-100%), watch GPU errors in HWInfo

Save your profile and set it to load in Windows

GPU UNDERVOLTING (NVIDIA):
--------------------------
You'll need MSI Afterburner

I recommend doing this configuration only on GPUs with high temperatures

Set desired power limits, I recommend setting them to maximum
Go to settings in Fan and turn all sliders to maximum
In General tab check: Unlock Voltage Control, Unlock Voltage Monitoring, Force Constant Voltage
Check for available product update set to "never"

Increase Memory Clock to maximum possible values, watch real performance increase in different tests from this action

Press ctrl+f and configure curve to values you want. When you set values with 1 point, pull others to it so there's no big frequency drop

During testing watch values in HWInfo, as in benchmarks they may be fake (frequency values + voltages)

Testing:
--------
3dmark, Memtest Vulcan, Superposition, Heaven, Furmark, Kombustor, OCCT VRAM (Set 95-100%)/GPU Test (8 shaders), watch GPU Errors in HWInfo

Save your profile and set it to load in Windows

AMD GPU OVERCLOCKING:
---------------------
Open AMD RADEON SOFTWARE - Performance - Tuning - Tuning Control - Manual

Click on sliders with GPU TUNING, ADVANCED CONTROL, FAN TUNING, VRAM TUNING, POWER TUNING

Set desired Power Limit

Increase memory frequency to maximum possible values, watch real performance increase in different tests from this action. Also after you set your frequency, you need to right-click on it and select "set as min"

Select maximum GPU frequency, while raising voltage. Also after you select them, right-click and choose "set as min" (On old GPUs). On GPUs starting from 5000 you need to set frequency difference of 100

After all operations click "Apply"

Testing:
--------
3dmark, Memtest Vulcan, Superposition, Heaven, Furmark, Kombustor, OCCT VRAM (set 95-100%)/GPU Test (8 shaders), watch GPU Errors in HWInfo

[IMAGE 2 PLACEMENT: tREFI/tRFC diminishing returns explanation]
Why raising tREFI or lowering tRFC has diminishing returns on performance:

This recently came up on the AHOC discord but I'm pretty sure I've mentioned this in a video or something but anyway:

As tRFC = tREFI approach 0, further reductions in tRFC or increases in tREFI will impact performance less and less.

tREFI increasing:
- 8000tRFC with 8000 tREFI = 10% of clock cycles spent on refresh
- 8000tRFC with 16000 tREFI = 5% of clock cycles spent on refresh
- 8000tRFC with 32000 tREFI = 2.5% of clock cycles spent on refresh
- 8000tRFC with 64000 tREFI = 1.25% of clock cycles spent on refresh
- 8000tRFC with 128000 tREFI = 0.625% of clock cycles spent on refresh

So going from 8K tREFI to 32K tREFI gives 7.5% more useful clock cycles >> big performance increase
Going from 64K to 128K gives you an extra 0.625% >> basically no performance increase

tRFC decreasing:
- 8000tRFC with 8000 tREFI = 10% of clock cycles spent on refresh
- 4000tRFC with 8000 tREFI = 5% of clock cycles spent on refresh
- 2000tRFC with 8000 tREFI = 2.5% of clock cycles spent on refresh
- 1000tRFC with 8000 tREFI = 1.25% of clock cycles spent on refresh

As I hope you can see this leads to the same diminishing returns that raising tREFI has.

This is also why I don't really bother with minimizing tRFC on daily memory OCs because once tRFC > tREFI the benefit for stability is not worth the tiny extra bit of performance but can harm stability.

Things you don't know what tRFC and tREFI are:
DDR memory needs to be "refreshed" in order to retain data because it stores data as charge on capacitors that slowly discharge over time.

tRFC is how long the memory controller sends REFRESH commands before sending other commands like ACTIVATE, READ, WRITE, PRECHARGE...

tREFI is how often the memory controller sends REFRESH commands.

The timings for RAM will depend on what RAM you have. Some memory sticks are incapable of running tRFCs of less than 600 at speeds higher than 1.6GHz (3200Mbps / MT/s)

VBIOS FLASHING:
---------------
Flashing is needed to improve your overclocking situation (for example: power limits)

The method itself is dangerous. If you do something wrong, you'll need a programmer

Download desired firmware from TechPowerUp along with GPU-Z, nvflash

After you downloaded firmware and nvflash, move it to C drive root

In cmd type:
cd C:\
nvflash --protectoff
nvflash -6 (your VBIOS).rom
press Y twice