2x16GB 6200 CL26 with Ryzen on an ASUS ProART X670E Creator

hey guys buildzoid here and today we're going to be taking a look at another ddr5 overclock with ryzen 7000 uh this time we're on an Asus Pro art x670 Creator motherboard and I bought this motherboard on eBay so big thank you to the channel supporters for making purchases like this possible now the CPU I'm using is a 7900x which was provided by AMD so big thank you to AMD for the CPU and the memory kit is a gskill 7800 CL 30 uh is a 2x16 7800 CL 36 kit uh that was provided by G skill and the ic's that it uses are HX 16 GB a so it's a uh single rank uh memory kit now then the original plan for this set of Parts here was that I wanted to do 6200 cl24 unfortunately um even though this board does allow very high memory voltages which is really cool considering that it's like a workstation type motherboard um but yeah like this board allows the memory like the vdd voltage for the memory to go all the way up like 2 volts and higher um the thing is that even with like 1.9 volts cl24 at 6200 simply wouldn't be stable and it's not an issue of this memory of like this G skill kit uh I tried the Kingston kit that I have as well right the Kingston kit that I used in the previous like 6,000 cl24 video um yeah I tried the Kingston kit I tried this gskill kit the g- skill kit is actually stronger than the Kingston kit for well just in general um but yeah neither of them would do 6200 cl24 and I suspect the reasoning for that is that basically once the voltage gets that high it's actually negatively affecting stability because like I could R like I couldn't run any extended like like it would run for a bit and it would inevitably error out at some point um so I did end up eventually giving up on that um so you know that's it's not really that disappointing cuz the the Practical benefits of like cl24 versus cl26 are basically non-existent it was more like like I it was mostly just a case of like well 6,000 cl24 works at like 1.75 volts so what would it take to do 6200 cl24 uh and the answer is uh it just doesn't work as far as I can tell like I couldn't get it to function um and so I kind of gave up on that and decided to go for something a little bit more uh sensible so 6200 CL 26 um that's that's my definition of more sensible um anyway uh as you can see it's run you know over 2 hours of white cruncher in fact I ran even more white cruncher than this but the there's like there was a fun like a weird issue with the previous white cruncher instance which I think was because I had CaRu open it ran really slow like it wasn't uh like it wasn't doing uh the performance was like really low and I think that was just because of some kind of software conflict because yeah this this ran uh properly the entire time um so yeah like all of these are 1.56 1.57 so yeah that that was evidently just some kind of software issue due to like what order I ran the tests in which is always fun um but anyway so yeah white cruncher passed and that's well over two hours and actually if you included the the run that I scrapped because it was like not spitting out the numbers that I wanted to see that was I think like a 4H hour run so anyway whole bunch of Y cruncher 12 hours of caro um you can see the Zen uh the memory timings over here and we're going to take a look at them more in depth in the actual bios because there's certain settings that don't show up here um it's kind of that also you know I do have a fan on the memory because we are running for cl26 you do need to run some pretty high vdd I don't think 1.65 would actually be completely necessary with these memory sticks it could probably go a bit lower but um yeah I wouldn't be like with a fan on the memory sticks I wouldn't be concerned about this like whatsoever basically um and uh yeah so um I think that's everything that worth looking at in the OS um I don't I don't think I have any viable benchmarks on this system yeah no I don't so yeah um the the thing like here's here's so the thing about like running memory benchmarks to me is like if you punch in good timings and you run a memory Benchmark and you're like me like your performance is bad it probably just means that you have some kind of software issue right like so anyway um like I don't see much point also these timings really aren't that special like the cast latency is pretty low but other than that I'd consider this quite generic hyx timings so um yeah I don't I don't really know why you'd want to see benchmarks it's like this isn't anything that hasn't been done before so let's restart the system I guess because I can't yeah I guess I'll commit to this take at this point now one thing that is somewhat annoying about this board is that it doesn't have a post code but at the same time it is an AMD system so it's not like the post code is that useful um but it is like it's disappointing that like I think the retail price for this motherboard is approaching like 500 pounds and it's not like it doesn't have a post code which is kind of ridiculous especially when post codes used to be a standard feature of the workstation lineup from Asus like if you go back like if you go back through like past Asus motherboards and look at the workstation boards they had post codes that was a standard feature because like you know if you have a workstation and it doesn't boot you might be you might want to know why it doesn't boot so that you can fix your workstation um at least that's the like the the the justification Asus had for like including post codes on their workstation uh series motherboards in the past and so now we have like workstation boards from Asus that don't have post codes which just doesn't really make any sense if you think about it but anyway so let's take a look at the settings here um so I did actually try even higher fclk but I still get eted out by the like like technically this ran at like 2200 but it's really hard like proving fclk stability is a pain and 2233 doesn't work at all so I have like Ser like I find that weird considering that like my 7800 x3d which like works like actually works at like 2200 that chip also kind of works at 2233 so I find it weird that you'd have a chip where it's like oh it doesn't work at 2233 at all but it does work at 2200 so anyway and then I actually ended up backing down the infinity fabric from even 2166 all the way down to 2133 cuz for a while uh I couldn't like I was having these like really rare errors showing up that were showing up like a few hours into the stress test turns out that was either trcd or trp doing that so technically I could probably bring the fclk back up now um so like 2166 um but the performance difference between 2133 and 2166 is like not very big so I you know like I don't feel like dealing with hard to track down stability issues when they don't really make that much of a difference so anyway though in this case the issues were actually caused by me just trying to run like pre like I was originally trying to get this to work and that that was like not like that would just fail the stress test after a long enough runtime um so yeah ended up testing with uh 3630 um yeah cuz I was kind of hoping that like these G skill dims would be some somewhat better for trcd than the than the Kingston Kingston ones but yeah I know like um these also like the trcd on both of these kits actually behaves extremely similarly the cast latency seems to require a bit less voltage on the G SK dims than on the Kingston ones but uh for trcd they're basically the same anyway in terms of voltages uh the so's at 1.23 that's another reason why I was like reducing the uh fclk is because uh in my experience you the FC scales negatively with s so voltage um so the more s so voltage you run the less stable the infinity fabric gets and the lower the speed that you need to run it at for stability and so with the 7900x previous L like when I was doing the 6000 cl24 overclock and also just historically running this this chip wouldn't really go over 2100 uh the new agsa has like drastically improved the infinity fabric stability but I haven't like I I like I need to do more testing on that before I actually trust it because it is very easy to set the infinity to fabric to like settings that aren't actually stable and then have very weird issues because of it so like I yeah like and I'm I'm approaching this very much as like a more like a daily stable settings rather than you know like Benchmark settings cuz for Benchmark settings if it runs and it spits out the score you want then who cares how unstable it is but for a daily system it needs to not crash as as far as I'm concerned right that's like it's not really a daily system if it goes down at random um so yeah that's that's my Approach on that and so that's why the infinity fabric like it's probably not maxed out with with this bios um with these settings right now um but yeah like I need more testing on that also if you're wondering about what's up with the S so voltage reading uh that is like why is it like 30 molts above the uh s so voltage setting there's a very simple reason for that the voltage measurement point for the S so voltage on the motherboard is like really close to the output of the actual so vrm which means it reads really high relative to the actual like voltage that the CPU is getting because thec uh Power plane is relatively like narrow and so relatively High Resistance and also the so power goes into the CPU through very few pins in the socket there's not that many pins dedicated to S so power to uh to the actual CPU so there's a pretty significant voltage drop across the S so power plane and the super iio uh voltage like measurement like voltage sensor that's actually giving us that s so voltage reading yeah that's hooked up like way way far away from the CPU so it reads really high um so that voltage reading right there is just like that's just a measurement error um I might make a video showing that in practice later but yeah like that's like a very very common issue with uh super iio voltage readings uh is that they're really inaccurate because the super iio is really not like it doesn't matter if that chip has good voltage readings cuz the important one is the actual vrm controller itself and the vrm controller gets its uh voltage readings directly from the CPU so uh the vrm knows what the voltage is the super iio has a vague idea of what the voltage is um and that vague idea is 30 molts higher than it what it actually is um cuz yeah we're right now we're and the reason I'm so confident that it is actually 1.23 volts right now is because we're sitting idle in the Bios and the vrm should have absolutely no problem maintaining a steady 1.23 volts while doing nothing in the Bios I've not actually measured it myself but that's based on past experience with motherboards like that really shouldn't be any like error in the uh output voltage while we're just sitting idle on the on the BIOS you know I wonder if the super iio reads higher than the back of the CPU socket cuz that is actually really high yeah so I I might do something to like hook up a bunch of different uh voltage probes to this board um if people are interested anyway um vdd at 1.35 I've just found this to work well on this motherboard there isn't like yeah like this is one of the so vdd is the voltage for the actual like that the memory controller uses to talk to the memory sticks um and as for what is a good like so I would for long-term longevity I probably wouldn't set this over yeah I think I the thing I don't know like with Intel CPUs you have like the Intel equivalent of this voltage would be vd2 but and like but that memory controller also deals with ddr4 whereas this is like a ddr5 memory controller on a completely different manufacturing process so uh I would personally just kind of stick to below like 1.4 volts or less um yeah Asus based on their color coding seems to think all the way up to like 1 55 volts is probably sensible but um there would like I don't know how much I like I don't know how they came to the conclusion that like one up to 1.5 five like that like 1.56 volts uh wait okay 1 Point 58 volts deserves to be red like I don't know how they came to that conclusion for those for the color codings so uh yeah I'm going to on the side of caution and just say like 1.4 Vols or less um also in my experience there's very little reason to use like more than 1.4 volts on vdd iio um yeah I don't think I've ever actually run into a setup where that was necessary so anyway uh Mis miscellaneous voltage is on auto because the miscellaneous voltage look basically it just powers like the vdd voltage Regulators you don't have to really worry about it like by by spec it's supposed to be 1.1 volts and as you can see it too is off by 30 MTS cuz it's being measured by the super iio um but uh yeah so no reason to change that cuz that does like the this is like a power supply for the actual Vol like vdd voltage Regulators uh by specit 1.1 volts the only reason you'd ever want to have like raise this is if you wanted to run like insanely high vdg voltages for some reason and in my experience there is never really a reason to do do that um yeah like the the pddg voltages really don't like actually setting them too high can be more like generally is going to be more detrimental to stability rather than helpful um so no reason to adjust that uh I have the High Dam voltage mode enabled because we're going over 1.43 uh volts on the vdd voltage we're all the way at 1.65 and as long as you have a fan on the memory sticks I think 1.65 Vols vdd is perfect at least for HX memory chips should be perfectly safe uh the one concern with this again is that well the one concern with this is that because we're running all this extra voltage the memory sticks are going to run hotter and so if you're like in a bad airf flow environment or a high Ambient Air Temperature environment and you want to run really high refresh interval this is potentially going to be a problem um like that's potentially going to harm like the highest refresh interval setting that you can run in which case you'd actually probably be better off running slight lower voltage cuz refresh interval probably more important than lowering your cast latency by two ticks um so yeah anyway then we have vddq at 1.4 volts uh this is quite different from how I would approach a gigabyte board right on gigabyte boards I usually run vdo and vddq the same uh I found that with this board uh higher vddq voltage like yeah like it just likes this value um that's like the annoying thing with dd5 is like because the memory speeds are so high uh there's quite a lot of variety in terms of like what like different motherboards will like different configurations of voltages um so yeah um and this just happens to work best with 1.4 volts um so that's why it is where it is anyway now Advanced memory settings I don't have anything set here uh is there anything below that nope um yeah the CPU is on auto so I didn't do anything to the vrm settings uh I think yeah oh I did disable spread Spectrum um which should just make the bclk drift around less um other than that didn't do anything there and so I guess at this point let's take well I did mess with the where is it there so I did mess around with the Nitro mode a little bit so this is this should be faster than like the auto settings as far as I know but uh I don't know how much faster um so and I did try lower values for for these three settings right here uh those don't work so like I tried zero and actually this one doesn't go below one so that one's at one um and one doesn't work on this one so yeah theoretically this should slightly reduce the memory latency this as far as I know is basically amd's version of Intel's roundtrip latency timings um I'm not like I don't like either of these well this makes a lot more sense than the Intel well this is a lot simpler to to play with than the Intel roundtrip latencies because in my experience trying to do like in MO yeah in my experience with messing with roundt trip latencies on Intel CPUs is that 90% of the time it just causes everything to break if you try to manually set them so um yeah anyway so that's that and then for the DDR controller configuration I didn't do anything yeah that's all on auto um so yeah so this is really just like a small that should help the performance a bit but I I don't have like EXA exhaustive like per like data for how much it actually helps um but I did set that and I don't think I changed anything else right yeah I didn't um anyway um so let's take a look at the actual memory timing so primaries we're looking at 26 3630 um so in my experience this is basically as tight as you can go on HX a die uh at 6200 I think I've seen screenshots of tighter timings out there and I'm going to assume that's just because of better silicon on the actual memory sticks themselves um cuz yeah like I I was trying to do trcd 35 for a very long time and I was kept getting like uh errors really really late to stress test which was super annoying and I ultimately decided to go up to 36 after quickly testing 34 and 34 was just completely unusable so I came to the conclusion that I guess like 35 is not you know like if 34 doesn't work like at all and it would post but like you couldn't actually like run anything like you'd get to the desktop you'd try to open a full like before you even really managed to do anything it would just blue screen um so I guess sometimes it wouldn't even make it into Windows depending on how you know long like I I only tried it once and I was like oh this is unusable so if 34 doesn't work basically at all then 35 is probably right on the edge and it would be you know like easier to just set this to 30 takes um so I did and trp was at 28 I ended up bumping it to 30 because I wasn't 100% sure if this point if like trcd or trp were the issue so and adding two more Cycles to trp isn't going to like drastically affect performance um like in terms of order of importance for the primary timings uh trcd is more important than TCL TCL more important than trp and tras Almost Doesn't Matter so anyway um well yeah trass almost doesn't matter cuz like the main reason you lower trass is so that you can set like a low TRC now for TRC I just stick to the rule of trp plus tras equals TRC I am not 100% certain that that rule actually is uh set in like Stone on the am5 memory controller on am4 it was if you set your TRC really high you could set your tr- like imp possibly low and the memory controller would just seemingly just ignore the tr- setting the am5 memory controller doesn't seem to do that um so yeah but um um like for for convenience I just like to stick to that rule because it it does generally work um I'm like it's also like if you want to experiment with lowering the TRC below like trp plus TR you can I'm just like theoretically it shouldn't do anything to anything to Performance but I haven't really tested for that and also testing for that is a pain because generally speaking if you're adjusting any one timing by like well small increments they don't tend to make of a difference right so yeah anyway tww is at 48 now funny thing uh Asus oh they do actually enforce the tww limit so that's interesting I thought they didn't um but uh refresh intervals at 65,000 um again if you're memory runs really hot this is probably not going to work um which is why I mean you know open air test bench with a fan on the ram sort of best case scenario for memory temperatures whereas if you put your system in a case and you have a big thirsty 4090 it's probably going to cook your memory sticks um anyway trfc we have that at 477 think it could actually go lower for HX AI I just generally don't worry about minimizing this timing because the refresh interval is 65,000 clock Cycles it it really doesn't M like it's not a big deal if you spend 400 or 500 clock Cycles out of your 65,000 clock Cycles on a refresh right like just doesn't really matter that much now trfc2 and trfc same bank aren't used by the AMD memory controller so I just leave them on auto because they literally don't affect anything uh rtps at 12 uh on some kits you might be able to set it below that but uh not my like i' I've tried and on my sticks like I've never had success with that trl's a eight this is a side effect of me trying to chase down that like inst ility that was proba like that seemed was probably trcd or trp um with 1.65 volts honestly I'd probably expect this to work even at four on a lot of memory kits but uh it's also worth noting that because ddr5 has a eight cycle like data burst length setting your TRL below eight really like doesn't do much to Performance um it technically does have like give does give the memory controller the ability to schedule activate commands closer together but it doesn't actually allow you to squeeze in like a whole extra read burst so um like the performance impact of setting either of the TRD timings below eight is really really negligible and of those two TRL is actually specifically the timing that the memory controller avoids using as much as possible so yeah um and with ddr5 with how many Bank groups you have that actually means that trdl is less import like on in like trdl on ddr5 is less uh impactful than trdl uh was on uh ddr4 uh in terms of performance and even on ddr4 it wasn't really a priority to run it all the way down at four right like regularly I would recommend people run like 64 for the trds because yeah a lot of the time 44 would work but 64 performs basically the same as 44 and with ddr5 it's like well 84 performs basically the same as 88 actually well 88 performs almost the same as like 44 for ddr5 because of just changes to how the memory works so 84 is like yeah that's fine uh T Falls at 20 now this one yeah Asus does let you set this below the register limit uh I am relatively certain the AMD memory controller does not understand what a TAA of one even is uh so I just stick to the like minimum value from gigabyte motherboards like gigabyte motherboards don't let you don't let you set this below 20 uh as far as I know because that's the actual like register limit for this timing um so yeah and setting it below 20 therefore theoretically shouldn't actually achieve anything in terms of performance so I see no reason to set like you could set it lower um it just shouldn't really do anything um because the memory controller is just going to cor that into the lowest like into the lowest actually possible value which should be 20 um anyway tww TRL uh we're at 164 this is just like you know my lazy hyx timings I think this could probably go a little bit lower but I can't like it's not going to drastically change performance at this point so you know why bother uh read to read is at four um SC is at one so this is just standard single rank HX things then we have right to right and right to right SC 21 actually this could probably work with one one uh but again like it's not going to drastically change anything so yeah and then TWD we at 416 like this is very much generic hyx timings plus some like lower primaries as far as I'm concerned so yeah anyway uh then we have uh uh oh I have the command uh yeah the command bus set to unbuffered which should be ont command Ray um interestingly enough I don't know where the setting for gear down mode is um on this board and I did want to experiment with that because you might have noticed that like with this setup I was like sort of trying to do like the Nitro settings and then also setting the command rate to one cuz I wanted to try something different than just you know cuz I did already do a 6200 cl26 video ages ago on a gigabyte motherboard I wanted to try something a little bit you know more than that uh but yeah I can't find the command I mean can't find gear down mode um not sure it might be this I don't know why Asus insists on always having the wackiest of names for things and then we have like a bunch of extra memory controller settings that Asus exposes that other manufacturers don't expose uh for all of the like Drive strengths uh terminations uh and well no yeah Drive strengths and terminations that's all all this over here is uh those are all on auto uh I left power down enabled because I have memory context in like on auto which is enabled now theoretically what you could do I'm not sure how reliable this would actually be um you could disable power downlo and memory context restore if you go into DDR training options run this enabled and you like so what training runtime reduction does is it means it's not going to like you're still going to retrain memory every boot which is a requirement if you want power down mode to work or like disabling power down mode to to work you you need to retrain the memory every boot but this drastically reduces the training time so that might be viable if the board is consistent in like retraining the same like stable settings every time um is like disabling power down mode disabling context restore and then enabling the runtime reduction so that the training doesn't take forever cuz if you don't turn on the runtime reduction memory training can take like four minutes or I maybe if you had like a really high density setup even more than that but anyway for for this like I wanted this to be like a sensible daily configuration so everything's just kind of like on auto there uh uclk is equal to mclk right because we actually want good performance not terrible performance um and then we have a bunch more memory controller settings that I don't actually know what these do um yeah because I've never seen them on any other motherboards these kind of look similar to the Nitro mode options but the names are very different so yeah I wonder what these do because these seem like well we'll not worry about that that so anyway that's kind of it um yeah 6200 cl26 um at like settings I would be oh I guess like I have vddp set to 1.08 I don't really see like this this is a uh so vddp in my experience um does affect like stability for y cruncher um it doesn't really affect stability for a lot of other things in my experience and you can have it set too high So like um though usually I've run into that being a problem at very high memory speeds rather than at like one to one mode so like in two to one mode if the vddp voltage is really like is too high then like white cruncher complains about it takes a while too like it doesn't like having the vddp too high doesn't mean that white crunch is going to fail in like a couple seconds it it takes like several minutes if not even an hour sometimes um so yeah anyway the board I think was defaulting to like 1.1 or 1.15 I ended up turning this down to 1.08 cuz I was trying to fix the Y cruncher instability which I do believe was originally just the trcd being slightly too low but I'm I'm not going to you know like go back and try to like spend hours trying to trace down where the problem originally came from um by reverting all of the adjustments um anyway um yeah so that's it um for for this combination here and honestly like as far as I'm concerned this is just like generic hyx things almost figure like almost considered not bothering to make the video because really the only special thing that I've done as far as I'm concerned is uh where is it in like yeah as far as I'm concerned the only like extra stuff that I've done here really uh is not this this it's this yeah um that is really the only like oh this is different from what I would usually do with with with hyx memory on on am5 um and I'm not even sure how much of a performance difference this makes so yeah anyway um but the point is like if you do if you are running a one to one mode set up with like HX dims that might be worth trying is running the Nitro settings at like 121 cuz I think on default it tends to default to like 231 or 131 something like that uh which is going to be a little bit slower um so anyway um there that's it um yeah as for like any like generic ddr5 overclock like this is an real issue for me with like making videos about memory is like it it's the same like I don't I don't know why everybody keeps making such a big deal out of it it's like you you you take a board like or at least on AMD I don't know why you'd make such a big deal out of it because for the most part it's like you just punch in hyx timings and then you either have to deal with like like there were AG gsas where like the memory context restore was really unstable um so you just didn't use it right you just turned off memory context restore and that that was the end of the story because the other option was that your system wouldn't be stable reliably um and that's kind of that and like the memory context restore issues in my experience at this point have been fixed um like like I've yeah like I've not run into issues with it um recently that but the thing is at the same time I have seen cases of like AMD motherboards sometimes like spontaneously just deciding to MRA um but I don't know like which specific agsa versions each of those you know examples actually used so I don't know it might have been like an old agsa bug um so yeah I don't know hopefully this is somewhat helpful if you have an Asus motherboard because like this is basically the same thing as the easy HX timings just like new agsa I I guess a lot of people might have thought like oh new agsa will allow like 6,400 to work and for me it really doesn't okay like I've tried on a few different boards at this point with this 7900x um and on the old agsa this 7900x already did 6200 um and on the old agsa it also didn't do 6,400 at all like 6,400 was incredibly unstable whereas on the newest agsa 6400 is just stable enough to take a couple seconds to like you can run a little bit of white cruncher and then it crashes or you can run a little bit of caro and then it errs out um it it's not stable and I haven't found like a fix for it and so I don't really see a like yeah I I don't see a well this CPU evidently doesn't do 6,400 as far as I'm concerned um and this CPU isn't even the worst one that I have in terms of memory controller quality the worst one I have is the 7800 x3d in my daily system that didn't do 6200 on uh well that that hasn't done 6200 yet so anyway I'm going completely off track at this point um so yeah that's that's it for the video hopefully you found it somewhat interesting or helpful or something cuz I didn't this generic hyx timings whoa never seen those before um but anyway that's it so thanks for watching and uh like share sub 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 of a patreon which is how I like funded the purchase of this motherboard and also like a Teespring store and a band camp there's links to that down in the description so if you check them out that would be much appreciated and yeah that's it thanks for watching and goodbye

Channel: Actually Hardcore Overclocking

Views: 12,130

Keywords: Overclocking, PCbuilding, Buildzoid, AHOC, Actually, Hardcore, Hardware, OC

Id: QJKBPJQVIZA

Length: 36min 19sec (2179 seconds)

Published: Mon Jan 01 2024