Original subtitles

hey guys Bill Zoid here and today we're

going to be taking a look at another

ddr5 overclock this time with Horizon

970 900x on an ASRock

b650m--hdv m.2 motherboard the CPU was

provided by AMD so big thank you to them

for sending it over and the motherboard

and the memory kit were purchased by me

so big thank you to my supporters for

making those purchases possible uh now

then the memory kit that I'm using here

is a Kingston Fury 2 by 16 gigabyte uh

memory kit it uses heynix 16 gigabit MDI

memory chips at least the one that I

have here theoretically this uh spec of

like 6000 cl36 at 1.35 volts you could

end up with high Nix a die memory chips

uh which wouldn't really change much

because the memory timings that I'm

using here aren't actually like super

like they're not minimized they're just

kind of generic hynix timings that

should work on both a die and MDI memory

kits so this video will apply to both

even though there are some very slight

differences in terms of what kind of

timings MDI chips will do compared to 8i

chips but anyway that doesn't really

matter that much so yeah that's the spec

of the system that I'm that I'm you know

working with here for cooling there's an

air cooler on the CPU so the CPU did end

up running quite hot especially for lin

pack and there's no extra cooling for

the memory so it the memory is basically

just cooled by whatever air washes off

of the CPU air cooler so anyway

um

let's take a look at the the settings

here so the memory I've pushed it to

6200 megabits per second so that's 3.1

gigahertz actual like memory clock

because DDR Right double data rate the

clock is half the data rate the data

rate is 6200 megabits per second and

then you also map to multiply that by

the bus width which for two sticks of

ddr5 is 128 Bits And if you multiply all

of that together you can actually

calculate your theoretical maximum

memory bandwidth uh anyway

um yeah so 3.1 gigahertz on the memory

3.1 gigahertz on the memory controller

uh you want to have your memory

controller running at the same clock as

your memory because that gives you a

latency Advantage as well as just a

general performance Advantage

technically there is a two to one mode

for the memory controller I see

absolutely no reason to ever use that on

ryzen 7000 in my experience it doesn't

really clock any higher and it does hurt

performance so yeah at least an ambient

it really doesn't clock any higher maybe

with like sub ambient temperatures it

actually helps but if you're on like air

cooling or water cooling there is zero

reason to use the two to one mode

because you're just giving up

performance and yeah there's there's no

benefits like it doesn't clock

significantly higher enough to to like

justify the trade-off

anyway the infinity fabric I have it

clocked at 2100 megahertz the infinity

fabric on ryzen 7000 is completely

desynchronized from the memory

controller because the memory controller

like typically your memory controller

might run at like say uh at like three

gigahertz and your Infinity fabric at

like two gigahertz there is no way to

like

cleanly transition data off of a three

gigahertz like clock region into a two

gigahertz clock region that just doesn't

work right so that whole thing with

ryzen 5000 where you want your Infinity

fabric synchronized to the memory

controller that works because your

Infinity Fabric and the memory

controller actually run at the same

clock if the clock is like you know if

what like three to two requires buffers

in between the two clock regions and

well once you're buffering the data the

you you just have a latency penalty that

that's just permanent that's like the

whole reason that unlike ryzen 5000 the

whole keeping your memory in Infinity

fabric synchronized was a thing was if

you did keep them synchronized you

didn't have to buffer the data on ryzen

7000 it always has to be buffered um

even if you set your memory speed to

like two gigahertz uh it still ends up

getting buffered there's there's no uh

like Infinity fabric to memory

controller synchronization on ryzen 7000

CPUs based on my testing so the good

news about this is that if you just want

to max out your performance you just

push the infinity fabric clock as high

as it goes the bad news about that is

that the infinity fabric causes some fun

instability like yes fun stability

issues depending on how you look at it

mainly in that the infinity fabric will

absolutely not produce at least not in

my experience it will not produce

Hardware errors at all that thing though

on ryzen 5000 if you push the infinity

fabric too far on a lot of CPUs you'll

just get a flood of errors in Windows

you do not get that on ryzen 7000

instead you get random performance

regressions which are proportional to

how heavy the test is and if you push

the infinity fabric way too far it will

actually out like right crash but

basically the infinity fabric on on the

7000 series does have some kind of error

correction with re-transmission

capabilities uh and so if you push it a

little bit too far you start seeing

performance regressions uh and if you

push it way too far it does start

causing outright crashing but the

basically what that means is that you're

going to want to run stress tests which

actually have a performance metric which

is why I use Lin pack for stress testing

this because 2100 for this 7900x is

actually really pushing it most CPUs or

actually I'd go as far as saying that

all CPUs should do at least 2066 on the

infinity fabric but yeah once you start

trying to go past that you'll find that

some CPUs really don't want to run even

like 2100 and this 7900x is one of those

a lot of chips might be able to actually

even do like 2133 but also it depends

like what's really annoying with the

infinity fabric is that it actually

clocks better if your memory clock is

worse so at like six ddr5 6000 you can

generally bulk the infinity fabric a bit

higher however the performance uh the

overall performance is actually better

if you push your memory clock first so

you should first max out your memory

while keeping the infinity fabric at

like two gigahertz and then once that

you've maxed out your memory clock

that's when you start adjusting the

infinity fabric doing it the other way

around doesn't work you end up with less

performance in the end so

anyway uh yeah so 2100 uh I got that

stable here linpak is running at the way

it should be basically what you want to

be doing is as you're raising your

Infinity fabric clock you want to run

Lin pack and check that the gigaflops

aren't going down so if like two

gigahertz you get

um I don't know like 620 gigaflops and

then at like 2033 you get 620 you know a

little bit more than that it doesn't

scale that much with the infinity fabric

unfortunately linpak on a on ddr5 is

actually uh like kind of latency limited

because ddr5 just has so much memory

bandwidth which like on ddr4 systems

linpak was like a really memory

bandwidth Limited Test and now it's

really more limited by like memory

timings which is kind of interesting

that like ddr5 just has so much

bandwidth available for for something

like linpak

that it like completely changes how the

like performance scaling behaves but

anyway so you'll you know see like small

performance increases as you push the

infinity Fabric and at some point you're

either going to start seeing calculation

errors or an actual reduction in

performance and that basically means

that the infinity fabric that like

you've pushed the infinity fabric too

far there are certain voltage

adjustments you can make to try uh fix

that

um which is what I've done here and

that's sort of one of the main things

that I wanted to show in this video

um

but uh yeah if you're not gonna like if

you don't want to fiddle with any of the

voltages manually then uh or you don't

want to push any of the voltages uh

further than the sort of lazy option is

to just stop at like 2066 because that

generally doesn't run into any weird

issues and also 2066 and 2100 perform

kind of they're the same because they're

not really that different from each

other right like going from 2000 to 2100

is a five percent clock speed increase

going from 2066 to 2100 is like well

less than five percent so

uh you know you'll you'll have to weigh

up how much effort you want to put into

that yourself

um

anyway uh we're gonna take a look at the

memory timings once we actually get into

the BIOS I only have ryzen Master here

for the clock readouts

so let's take a look at all of the

different stress tests I Ran So I did

run linpack uh this is 77 Loops of Lin

pack which 45 seconds per Loop works out

to I think just under an hour I also ran

over you know 5 000 seconds of Y

cruncher so that's like over an hour

like yeah that's well over an hour of Y

cruncher why cruncher is actually really

good at picking up on Infinity fabric

instability

um it will just outright crash which is

kind of neat though I'm not 100 certain

that it's completely consistent about

the whole like outright crashing thing

because some CPUs might be more tolerant

to Infinity fabric instability than

others so

yeah anyway but why cruncher passed

completely stable no problems there Lin

pack is running at like the gigaflops it

should be running at also no calculation

errors whatsoever we don't have any

errors down here though I really don't

expect to see any errors even if Lin

pack and white crencher were crashing uh

just based on my experience with pushing

the infinity fabric so far and I've also

run three thousand percent of HCI mem

test though the there's one thread that

only ran to two thousand percent and

this is a general issue with the Dual

CCD 7000 series chips in my experience

uh basically there is not enough memory

bandwidth to go around and windows isn't

good at balancing the threads so

basically what's going on here is that

like the first 22 uh 23 threads because

we have thread zero over here right so

they're started from zero so that's 23

threads over here so the first 23

threads basically have access to pretty

much all of the memory bandwidth they

need and this one's getting kind of

neglected and that's why that one that

one you know is only two thousand

percent while a lot of the other ones

are well over three thousand percent

right like this one was only getting

like 800 megabyte like this one was

getting around 900 megabytes per second

of memory bandwidth uh sort of provided

to it while a lot of the other ones were

running over a gigabyte of memory

bandwidth

um

well over a gigabytes per second of

memory bandwidth and the issue here is

that there's just not enough memory

bandwidth across the entire memory

controller to feed all of the threads

evenly at the same time and because

Windows is a bit dumb you end up with

one thread that's just lagging way

behind this is even worse on a 7950x on

a 7950x you'll have like eight threads

that are just miles behind all of the

others uh the 7900x is actually not the

worst example because the thing with the

7950x is like it has you have even more

threads that pull the same amount of

bandwidth and the basic issue with with

ryzen 7000 is like the memory system

here really doesn't actually have that

much memory bandwidth you can really see

that reflected in the like overall

bandwidth readout over here it's only 30

gigabytes per second like this is

high-end ddr4 territory

um now admittedly this is a memory you

know like relatively sequential workload

I would assume whereas something like

linpak is very random which is why

linpak like loves ddr5

um but yeah uh and like Lin pack even on

ryzen 7000 just loves ddr5 also it might

be that ryzen 7000 just isn't that

bandwidth heavy when running linpak but

either way in mem test you can actually

really see that like amd's memory system

is is really not maximizing the memory

bandwidth that's available from from uh

ddr5 I would suspect that it's mainly

because the infinity fabric is at like

2.1 gigahertz while the memory is at

like 3.1 gigahertz so you know you have

this really fast memory hooked up to a

not that fast CPU interconnect

um so yeah there's like with linpack

that's not a not really an issue it

doesn't use as much bandwidth as memtest

does but mem test uses a ton of

bandwidth and so it just kind of gets

bottlenecked on the infinity fabric from

but based on how it behaves

anyway here's the uh

Ida uh performance results 55

nanoseconds of Licensing which is pretty

solid and then I don't think these

bandwidth readouts are at all accurate

but they're there

um and uh then for a more useful

Benchmark or at least in my opinion a

more useful Benchmark we have Pi Prime

2.0 over here uh 9.64 uh

yeah 9.649 seconds now the thing with pi

Prime is it does scale a little bit with

CPU clock and I'm obviously on a 7900x

so this result would not necessarily be

comparable for say lower clocked CPUs

like the 7600x but if you have a 7950x

or something uh you should be like sub

10 seconds is like the goal as far as

I'm concerned if you're getting over 10

seconds in pi Prime on a stock 7000

series with overclocked memory you're

over you could you could overclock the

memory harder

um

also even lower results are possible and

x3d CPUs actually I think run Pi Prime

even faster because this this Benchmark

is like crazy memory latency limited and

so it also benefits a lot from L3 cache

um

so

anyway uh oh no I did the I did the

windows thing uh that's what I get for

using a new OS install but anyway uh so

yeah all of the stress tests passed I

didn't bother running test mem 5 just

because it's so Jank and it's so clunky

it's a good stress test it's a good free

stress test it's just so clunky that I

don't like using it

um

anyway so yeah that's that's all of the

stress has passed I mean at this point

I'm going to be restarting the system

anyway so

um also that means I have now committed

to this this take of this video so

yeah actually I don't know why I'm

manually closing everything down I

should just hit restart

now I do have the memory uh context

restore disabled so this is going to

take a while to post

um because the memory context uh restore

with over like with yeah if you're

manually overclocking it can cause some

stability issues

um and it depends it kind of depends on

the motherboard as well like some

motherboards are less bad with the

memory context restore enabled but on

some other boards it's really bad where

if you have it enabled for like if you

have it enabled after a couple restarts

you might find that your memory is no

longer stable

um because during every boot like if

something slightly changes on the system

uh you want the memory control

controller to basically retrain the uh

memory also please oh I missed the BIOS

it's actually not that bad a boot up

like considering I have memory context

restore disabled that really didn't take

that long but uh

yeah which is how how also like the

capture card doesn't help because the

the system's actually like the capture

card actually like lags behind the

system like initializing the GPU

so I'm just gonna have to mash the

delete button

now there we go and it's up

yeah but still I wouldn't consider this

an egregiously long post time

considering that I have memory context

restore uh disabled over here I do

appreciate the fact that ASRock put this

into the OC tweaker menu now then here

you can see my settings I am pushing the

memory voltage relatively like this is

about as high as I would go for daily uh

for the vdd voltage now vddq and vddio I

have these at 1.25 volts the reason for

this is the vddq voltage and the vddio

these are basically the signaling

voltages so this you can sort of so dram

vdd voltage is like the voltage that

powers the internals of the memory chips

uh vddq voltage Powers the input output

portion of the memory chip so this this

is the voltage that's like actually used

for talking to the cpu's memory

controller and the vddio mem S3 over

here that's the same deal but from the

CPU side so the CPU uses vddio mem to

talk to the memory and the memory uses

video ddq to talk to the CPU

and in a lot of cases what you will find

is that

um running these low cons in well

depends on the motherboard and the BIOS

version but running these lower can

sometimes help with stability

just because driving the signals with

higher voltage doesn't necessarily mean

that the signals are going to be easier

to read on both ends so

um yeah now you can also just use the

the VD dvdq equals like the all of them

being equal mode

um I'm not entirely certain that what

I've done here is actually necessary but

it works and I'm like the the thing is

uh depending on the motherboard and the

settings you're running you might find

that there's actually a whole bunch of

different voltage combinations that do

work

um because with say this like on some

other boards having high signaling

voltages is just fine right so like

running everything at 1.42 volts might

have actually worked just fine

um but uh on say I think they're like

there's a specific set of bioses for

like gigabyte boards where you actually

want your signaling voltages at 1.25

volts because above that it just causes

issues with memory training atheroc

doesn't like this the BIOS that I have

on this board doesn't seem to suffer

from that

um

but uh wait where does it say well bile

swirl oh I guess on Main yeah so we're

on the 1.21 bios also this bios does

have the 1.3 volt SOC voltage

restriction which uh uh well you can see

that sort of sort of down here it's just

like yeah uh also yeah I have that

completely maxed out because the cpu's

memory controller is not great

um

but anyway

um yeah so the like

so the vddio end of this this is

completely safe for the CPU

um right the memory vdd voltage that's

not a problem for the memory sticks

themselves the vddq voltage also in and

of itself wouldn't be a problem for the

memory stick I do have a slight concern

because according to the ddr5

documentation these two voltages should

be kept relatively close together

but on the other hand I've also seen a

lot of people you know running at least

High Nix chips with very big gaps in

between these two voltages because the

vdd voltage is sort of what really

allows you to push your memory timings

down whereas the vddq voltages of

voltage you basically want to set up

like set optimally for the like

signaling behavior of the CPU and the

and the motherboard that you're using so

yeah this one sort of affects your

timings and this one really affects your

your clocks basically is the way you can

think about it

um and they don't necessarily like and

you won't necessarily get the best

results by having them both at the same

value so uh yeah I wouldn't be too

concerned about this as it's set up

right here like yeah I I would I would

I'd be willing to daily this if you have

doubts about doing this then ASRock does

have the option to just have all of the

voltages set equal to each other in

which case uh I mean I don't think 1.4

volts vddio would be dangerous for the

CPU as there are uh there are you know

Expo memory kits with the 1.4 volts as

their Expo voltage I would assume AMD

wouldn't certify those if 1.4 volts of

eddio was actually a problem so

though there are some motherboards that

actually default your vddio and vddq to

1.25 volts even if you're like vdd

voltage is at 1.4 so

yeah this this is the thing is just

like

I don't have like there's not enough

data on this like I don't work at SK

High Nix I don't know how the memory

chips behave internally at this from all

everything I've seen and my own

experience this shouldn't cause any

issues but I can't say like with a

hundred percent certainty that this

won't cause any issues because I just

don't know

um

anyway dram VPP voltage is at 1.8 volts

which is the stock voltage for the VPP

voltage it should be at 1.8 volts that's

that's what it defaults to

um anyway then I have the infinity

fabric set to 2100 uclk equals mem CLK

uh also speaking of UCL k equals mclk

this is a really annoying thing about

the azroc file so if we set ddr56000 you

can see that uclk mclk is is the same if

we set 6200 it automatically divides

your mem CLK by two

um

this I think is mainly a mainly like

ASRock is mainly doing this because if

you have an XMP memory kit that's like

6200 or 6400 it's more likely to work if

they down clock the memory controller

the obvious issue with this is like well

if I set my memory clock myself I have

every basically every time you adjust

your memory clock upwards you need to

manually set your memory controller to

memory clock ratio which is kind of

annoying

um

yeah anyway below that we have SOC

voltage which I have at 1.3 volts which

is as high as it goes on this BIOS

version right I can lower it

I cannot raise it I do appreciate how

quick asrock's bios refreshes like there

are some bioses that are really laggy

convert to this so this is quite nice

um

anyway then we have vdd misc voltage

which is at 1.1 volts and that's the

actual default voltage for that I have

not raised that at all

um or I do think I've manually like set

it to 1.1 volts but this is what it like

defaults to on every single motherboard

even on Full Auto settings this voltage

is used to basically power the voltage

Regulators that generate the vddg CCD

voltage as well as the vddg IOD voltage

now these two voltages these affect your

Infinity fabric clock stability

um most boards once you enable an expo

profile will default these to 0.95 volts

I think at all like bone stock I think

is like 0.85 at like ddr5 5200 or

whatever

um

but what I found with some CPUs if

you're pushing the infinity fabric like

basically like if you're trying to max

out the infinity fabric raising the CCD

voltage can help with this uh can help

with stability so that's why I have the

CCD voltage at 1.05 volts instead of

just 0.95 because usually you would just

use 0.95 because that's just sort of

what it tends to default to

um but yeah I've ended up bumping this

up to 1.05 one thing I did run into with

my 7600x and this was on a different

motherboard so it might be tied to the

motherboard to some extent is raising

the vddg voltages past one volt can

cause weird stability issues at idle now

the system that we're looking at here

was literally idling for over a hundred

hours at this point without any crashes

or anything so this seems to be working

at least on whatever BIOS version this

is

um but uh yeah so that that is something

to watch out for if you're messing with

the vddg voltages and especially if

you're pushing them like past one volt

uh you might find that like your

stability in like Lin pack and why

cruncher VST gets better but then the

system like just resets itself at idle

which is super weird I have no idea why

that happens but it is something that

I've run into with a 7600x when pushing

these voltages for the same reason in

order to try get like uh I think I was

trying to get 2133 or 2166 Infinity

fabric to work

um and that's actually the CPU where I

first discovered that hey raising the

CCD voltage actually kind of helps

stability at least under load it didn't

help but there was the the idle issues

anyway with with this chip on this board

this actually works for getting 2100 to

to be stable

so that's pretty cool then we have the

vddp voltage this voltage is used for

actually uh powering the Phi of the

memory controller uh

ddr5 bus signaling

item in it

that doesn't make any sense

oh right okay well the just yeah that's

what I figured

anyway so yeah if you read the

descriptions for these you have vddp is

a voltage for the ddr5 bus signaling the

the Phi so that's the actual physical

interface to the memory chips but if you

read the vddio voltage it says the same

basically the same ddr5 bus signaling

five uh yeah

um so that's fun

all right why would vddi have to be

greater than

SOC voltage

like I just noticed that they have that

rule there but it's like

is it a reverse biasing issue

because if that was a reverse biasing

issue I'd expect the CPU to be dead at

this point

um

also that just doesn't make a ton like

there's like some relatively low voltage

high speed memory kits out there

not sure how like I'm pretty sure you

can find like 1.2 volt ddr5 6000 kits

I I can't imagine that the board would

default that to like 1.35 volts on that

one of those

um

man it would be really cool if AMD like

provided actual public documentation

like Intel does

um except they don't so uh yeah

man I don't want to reshoot this video

because like this works

this is stable

um

I really wonder what that like SOC

voltage rule is supposed to be about

though

like that vddio has to be above SOC

voltage

wait I can't be right because stock so

oh no because stock SOC voltage is one

volt and if you use a jdeck memory kit

you'll be at 1.1

I really wonder about that

that is such a weird way to Def like

I I really wish I had a like and you

know what's really annoying AMD back in

the am3 plus days you know when they

made CPUs that weren't competitive with

apps like anything

they published like full-on overclocking

like guides for their CPUs I guess to

try compensate for the fact that they

were really slow out of the box and

those guides would include a bunch of

voltage recommendations

AMD hasn't done that since ryzen 7 uh

since ryzen 1000.

um yeah they kind of did that for ryzen

1000 but since like ryzen 1000 I haven't

seen them like do anything similar to

the the old like FX processor tuning

guides that they had which is really

quite annoying because it did include a

bunch of like uh voltage setting uh

recommendations and I'm I'm looking at

this wondering like wait like is this a

stability concern or is this like a

longevity concern because like if it's a

longevity concern at this point I'd

expect the CPU to be dead

um

if it's a stability concern well

evidently that's not true

because this is stable

um

so anyway that's fun

um

yeah honestly I think I'm just going to

leave this video up as is

um

like I'm not like yeah no screw it like

if you if you you know if you want to

set this voltage higher like that's up

to you

um I'm gonna leave it at 1.25 volts

um

honest and I would do that even with my

7800 X 3D

um there uh like I I know because like

this is a CPU provided by AMD so it's

like oh builds what you're just saying

that because the CP was in Texas the

risk no if like when I set up my well

the thing is with my 7800x3d I probably

wouldn't want a daily it at 1.3 volts

SOC

um but I'd probably daily it at 1.25 and

I would probably still run my vddio at

1.25 because there's a pretty decent

chance I'm going to end up dailying a

gigabyte motherboard at this right

um just because well yeah it's like yeah

there's a decent chance I'll be dealing

a gigabyte motherboard like I haven't

settled on what motherboard I eventually

want to move into my daily system but

right now it really looks like it's

going to be a gigabyte board even though

they're pcie like well

we'll see

because I I kind of want to have an

external clock Jen anyway

um

yeah so these are my voltage settings

here if you don't like them you don't

have to use them

um

I haven't adjusted anything in the

external voltage settings because

there's just no reason to do anything

here there's not even any LLC settings

which is a funny thing with this board

I'm not sure if that's necessarily a bad

thing or a good thing because on some

motherboards uh having LLC settings

doesn't mean that they're actually

useful so

um yeah like well like I'll decide

whether or not I'm happy with the lack

of LLC one LLC settings once I measure

this board with the oscilloscope I

haven't done that yet anyway now you

might be looking at this over here and

thinking oh this motherboard has an

external clock gen uh no it doesn't yeah

it unfortunately doesn't

um

ASRock basically this is just a thing

that they do they will regularly forget

to remove settings from their lower end

boards for functionality that they have

on their higher end boards I am pretty

sure this is not the first time I've run

into an ASRock board that had a bclk

setting that doesn't actually do

anything like the first one actually

works that's your like regular vclk that

overclocks absolutely everything but the

external clock gen that you can get on

some high-end boards that avoids

overclocking the PCI and stuff yeah this

board doesn't have that so the BCL K1

setting actually just causes the board

to like not post if you try to change

that well no if you do like this

nothing happens and if you do

this the board fails to post and if you

do this you get a two percent bclk

overclock so yeah

um welcome to ASRock motherboards anyway

then we have the spread Spectrum setting

which I have disabled because that just

basically uh spreads the frequency of

the bclk around a little bit to reduce

Emi

um if you disable it your bclk ends up

being more consistent which if you're

you know pushing your frequencies to the

limit is kind of nice that the bclk

isn't floating around by a by a small

amount

um

but anyway this can potentially cause

Emi issues which is why this is a thing

um but I've never run into those myself

and I've been disabling spread Spectrum

on motherboards for basically as long as

I've been overclocking because that is a

very yeah that is a and like you have

spread Spectrum for vrms and for cl like

for bclk and I've always been disabling

it and never ran into any issues so

anyway but your experience uh may vary

with that so yeah that's basically all

of the adjustments

um that I've made on on this board

there's really

oh yeah so that's like the voltage

settings and now we need to go over the

timing settings I have gear down Mode

still enabled because disabling it is

not worth the effort in my experience

um you just call like you get free

stability at the slight like a

negligible reduction in performance like

I I'll take the free stability thank you

very much uh if I'm pushing benchmarks

then yeah I'll go and disable it but for

for a like stable daily ish setup uh I

can't be bothered anyway then we have uh

Castle agency 30 trcd 38 this could

probably go a little bit lower like 37

or maybe even 36 if you're lucky

actually I think with this member well

this memory kit will probably not do 36

but 37 might work if you're lucky with

your memory kit you might be able to do

36

um at 1.42 volts but this kit I like I

tried that it didn't work uh trp's at 38

this could also potentially go lower

depending on how lucky you get with your

memory kit uh t-ras is at 30. now ASRock

doesn't enforce the

so ASRock does enforce a timing limit

but like I'm not sure if this one is

real because if you compare it to say

gigabyte boards gigabyte boards enforce

a t-ras limit of 30.

again if AMD had public documentation

for their CPUs like Intel does I

wouldn't have to wonder about this but

yeah so I'm going with the higher like

the gigabyte timing limit

um because uh

it like I'm not sure if um

if you can actually set tras below 30

also it's worth noting that the TRS

timing doesn't affect performance that

much so like pushing it below 30

probably won't make much of a difference

now TRC does affect performance quite a

bit however the way TRC works is if you

set it below your trp plus trass it

doesn't do anything

um so this is just set to 68

um because my trp and traps are 30 and

38 so t-ros 68. if I set it to 67 uh it

shouldn't really improve performance in

any way shape or form it also shouldn't

affect stability I haven't retested that

since the ryzen 5000 chips but unlike

ryzen 5000 chips you could set your like

TRC impossibly low and then your t-ras

would sort of actually be what would

govern your actual like real TRC or you

could set your t-ras impossibly low and

then your TRC would govern your TRC

um

so yeah that's that's what's going on

with that anyway like this is not

necessarily a super tight TRC setting

for for heineck stims at 6200 but also

there's it's not really going to go much

lower than this I don't think uh then we

have TWR which is at 48 which oh as rock

apparently and for okay I think they

just like have arbitrary timing lists

because 42 is just the most random TWR

no as a 48 like I can I can understand

why it would be 48 because that at least

lines up with some of the ddr5 jdec

documentation

42 is just like a completely random

number as far as I'm concerned so anyway

48 at least this should be the register

limit on the CPU the thing is if you

push

um if a motherboard lets you set timings

below the register limit of the CPU the

CPU will just round up to the actual

register limit usually so setting the

timings even lower will just kind of not

do anything a lot of the time

um anyway I have the refresh interval

completely maxed out at 65 535 the trfc

one is at 477. uh this could potentially

go a little bit lower but um yeah

slightly below 500 tends to be you know

a good stopping point for heinek's based

memory sticks uh then trfc2 and trfc

same bank are not currently used by the

AMD memory controller so I have them at

1337 because that's just kind of funny

um I would have them at 69.69 but they

won't let me do that and if I do

actually well actually it would let me

have them at 69. so I guess I should

have set one of them to that but yeah

neither of these timings is actually

used by the memory controller I have

another video sort of testing this

um these are an alternative refresh mode

AMD just hasn't implemented it or if

they have implemented it it's not

enabled by default

um

on like any motherboard in my experience

so yeah anyway that's why I just like

you could also just leave these on auto

like the point is these two timings

don't actually do anything right now

because they're just not being used

um and we have RTP which I have set to

12. this could probably go a little bit

lower especially on MDI memory kits

it'll usually go as low as like eight

maybe even below that but actually yeah

maybe even below that uh but uh on ADI

memory kits you will find that generally

it doesn't like going below 12.

um at least not without tons and tons of

voltage

anyway up next we have trdl and trrds

which I have an eight and four TR RDL at

6200 on an MDI kit could probably do

four instead of just eight but the

performance difference from going like

going from eight to four on TR RDL with

ddr5 is relatively negligible because

there's a lot of different bank groups

so the memory controller shouldn't be

using this timing very often uh and so

yeah that's like this timing isn't like

super performance critical

um and trds is literally at the register

limit it cannot be one okay I assure you

it can't be one

uh T Falls at 20. uh ASRock for some

reason allows you to set it to 12 as far

as I know the AMD memory controller

doesn't doesn't allow it to be less than

20 but again

um that's mostly based off of the timing

limits of gigabyte boards

um

not based off of like you know register

documentation

um which for Intel CPUs is accessible

but isn't for AMD CPUs because reasons

um

anyway now we have the right to read

timings uh TW uh wtrl is at 16.

I don't think you're you'll be able to

push this any lower than this not at

least without like tons of extra voltage

and wtrs is at four this might actually

go a little bit lower

um but it won't really do much to

performance at this point because it's

already very low

um

anyway then we have read to read scl

which is F4 uh this won't yeah this

probably won't go any lower than this uh

same uh SC is at one this is as low as

this timing goes uh then we have the

right to rights these are high Nix

memory sticks so they do very low right

to right timings so scl is a two and SC

is also at one because it doesn't go any

lower than that and then we have the

right to read uh the other set of right

to read timings and read to write

timings which are at 4 and 16. uh this

might go a bit lower than this but again

it won't really drastically change

performance at this point and read to

write is not really going to get much

lower than 16. uh actually I don't think

it'll be stable at all below 16. so yeah

these are sort of like these are

basically the memory timings I will

punch in on any AMD system with high Nix

memory sticks at around 1.4 volts on the

memory if you're lucky it might even

work at like 1.35 volts and by 1 like by

on the memory I mean the like vdd

voltage as as I've mentioned earlier the

vddq and the vddio voltage those are

more about the memory clock than they

are about the actual timings

um

your timing scale basically with just

vdd voltage because like the vdd voltage

Powers the circuitry that's affected by

the various uh memory timings that we

have over here whereas like vddq just

Powers the part of the memory chip that

actually outputs the data

um and the the data or like receive the

data actually it doesn't really receive

the data but anyway

um

yeah so

that's it for this um so this video was

a bit of a mess but uh I'm gonna I'm

gonna leave it this way because I mostly

wanted to show that this ASRock board is

actually like capable as far as like

memory overclocking and infinity fabric

overclocking goes like this is as high

as I've ever had the infinity fabric

clocked on this 7900x and it's also as

high as I've ever had the memory clocked

on this 7900x so uh

yeah the Amazon

b650m-hdv is you know perfectly capable

of like memory overclocking also it ran

Lin pack for like

uh it actually ran an impact for much

more than just the 77 Loops that you saw

in the initial stress test here because

I ran will impact multiple times in the

process of like getting to these

settings so yeah this board has

absolutely no problem handling a you

know 7900x even a 7950x really shouldn't

be any any uh like you shouldn't give

this motherboard any any difficulty

um especially because the 7900x and the

7950x actually have the same at least at

stock settings they have the same power

limit

um and current limit as well so

yeah anyway

um

that's it for the video so hopefully

this is somewhat helpful if you have an

ASRock uh B6 really any b650 motherboard

because yeah like the the thing is a lot

of these settings you can just kind of

like like on one hand you're technically

not supposed to copy other people's

settings blindly that doesn't tend to

well the issue with copying other

people's settings is if you don't know

what you're doing and you copy

somebody's settings and it doesn't work

you're not going to be able to fix it on

the other hand I do actually just kind

of use the same settings on basically

every motherboard

so I like copy paste my own settings

um so make of that what you will

um

but uh oh actually I should mention that

there are some CPUs out there and you

have to be incredibly unlucky for this

but if you get incredibly unlucky 6200

might not work

um yeah but that is like very very rare

every single CPU I have can do over 6200

um not 6400 6400 is very rare but 6200

is very doable in my experience so

um

yeah now with the fclk I mentioned like

2066 is very doable more than that tends

to depend on how lucky you get and then

maybe if depend you know also can depend

on how much you decide to push your like

vddg voltages

um

anyway so yeah that'll be it for this

video it went way longer than I was

hoping it would but oh well here we are

so thanks for watching uh like share

subscribe leave any comments questions

suggestions down in the comment section

below if you'd like to support what I do

here with actually hardcore overclocking

I have a patreon there's a link to that

down in the description below there's

also the hoc Teespring store where you

can pick up shirts uh hoodies posters

you know the usual YouTuber merch and

I've also got a band camp uh there's a

link to that down in the description

below as well

um and uh yeah if you'd like to support

the channel That would be much

appreciated and that's it for the video

so thank you for watching and goodbye