Memory Try-it technical question

[jamie.jameleousunicor15a002e6]

Before I try it - is your MSI TryIt based on MSI's own testing of Micron F-Die chips, or are these too recent to bring up reasonable choices of speed/timings. Or will it get it fairly correct. I do note some voltages are suggested at 1.4v but would that be unwise to go that far and just stick with the suggestions which offer 1.35v. Reason being is I've seen F Die to need 1.38v maximum on forums. I've tried 1 slower underclock but with tighter timings which actually worked and Cinebench completed (2000MHz clock). XMP is 3200MHz (2x32Gb sticks of HyperX Fury Beast). The next step was I think around 2400MHz which crashed on bootup. All these were tighter timings.

As all the higher clocks than 3200 are suggested with tighter timings, is it going to be a waste of time going with those, or going with those and loosening up the timings to worse than XMP suggested ones?

I'm just debating whether I should just be happy with XMP with this?

Since trying this I have updated to the latest March BIOS. It's a Gaming Pro Carbon Max Wifi B450 board.

 

[citay]

I never use Memory Try-It!, for me it's one of those "last resort" settings for people who can't get a proper memory configuration on their own (no offense).

You never want to underclock your RAM, at least not with RAM that's up to DDR4-3600. With faster RAM like DDR4-4000+, there's sometimes a reason to underclock, mainly when you want to avoid a divider for the memory controller coming into play. But DDR4-3600 or slower, you always want to keep the XMP speed or even increase it to up to DDR4-3600 perhaps (depending on the memory IC and other factors), and then just work on improving the timings manually.

Frequency comes before timings optimizations, at least up the DDR4-3600 speed i mentioned. The timings translate to different absolute latencies, for example DDR4-3200 CL14 equates to roughly the same latency as DDR4-3600 CL16, because one clock cycle takes less time at the higher speed. But you get the additional benefit of the higher thruput from the higher frequency. So you always prefer the higher speed first, then you go after the timings. I'm not saying, set your DRAM Frequency to 3600, because your particular modules probably won't like that. Just giving an example of how frequency is more important.

I have to be honest, RAM timing optimization is something you can spend days with (especially with the necessary stability tests). So i would say, once you get the frequency to a nice value, then maybe dabble a bit with the primary timings, but for anything more, you need to know what you're doing. Learning that alone can take days as well. Plus your big 32 GB modules won't have a lot of room for tuning as it is. Steer clear of "Memory Try-It!", those are OC profiles for some commonly used modules, but you don't have very commonly used modules. You need to do it the DIY way.

 

[jamie.jameleousunicor15a002e6]

Thanks for the info it's helped a lot.

My strategy is to try something on auto for a short time then if it seems "capable" and doesn't crash to try manual. I've successfully overclocked the APU (core and gfx) in both PBO and static overclocks to settings beyond what I expected when I bought the 5700G. All core is excellent at 4.6 and because I have a Scythe Fuma 2 and plenty of case fans I don't get much of a droop and that's without LLC, though for MSI I hear about 4 is a good setting for LLC. But because I tweaked the CPU on PBO, I get at least +800 better Cinebench scores just by doing CO and toning down some of the EDC TDC settings. I can confirm that anyone trying overclocking that less heat and toning down some voltages can lead to better sustained results.
For instance if I do all core I get a good high result, but by tweaking PBO I can get a resepectable result too.
Use case is the most important factor, and as I do compiling in Gentoo Linux but play the occasional games when I get time, I have set up a few saved profiles in the MSI BIOS. If I know that I'm going to be doing a huge amount of compiling I'd set it to all core and let it rip. If it's general use I'd do PBO + CO. If I was to play games, I might still try all core as the FPS is generally better and benchmarks turn very smooth with much better FPS. But, it depends on the game, some are better with 1-2 cores so PBO with perhaps a 4.85 boost would benefit those. So I have a profile set for that too.
One thing I did was a BIOS update to the end of March version and re-ran the CO test in Ryzen Master to get some figures to save in the BIOS. I guessed that a BIOS update does have significant changes which would mean if I ever do it I would be best to re-run CO tests. I had an older BIOS previously but it wasn't capable of finding my storage on Windows 11 (I dual boot). I may be tempted to upgrade to the June version to sort the TPM issues but I haven't experienced any yet, and I could just get a module, as I am guessing I might need to test stability again on the RAM CPU and GPU overclocks if i did upgrade it. So I have the option of staying on this BIOS or updating, but.. we'll see.

So back to RAM - you'll be pleased to see that I tried a semi-manual overclock by keeping auto voltages (but keeping an eye in the BIOS as to what voltages are being used), and first upping the frequency one step at a time. I actually got from 3200 to 3533 on default timings. I did 3600 on default and got some crashes, but loosened timings and 3600 was perfectly stable. That is the latest, it took time to do some memtest tests but nothing crashed no errors. Voltage seems to be fed at the default still. I think the tricky part will be as I go to more extreme speeds, but overall transfer bandwidth has increased according to memtest. I should get 4100ish max out of the chips but I reckon as a guess I'll get 3800ish before I hit a wall and will probably have to manually set voltages by that time. Also, I might need to change the SOC voltage but so far the board is surprisingly capable of managing whatever I throw at it. I say suprisingly not because I doubt it but, because I must be lucky with my silicon to get any overclock on anything - a lot of people sometimes can't.

So thanks for your help and I'll be going more manual as I get a feel for the capabilites.

It is addictive though! I don't think I could cope with not having an overclock option now..

 

[citay]

You may also want to read my RAM thread, forgot to link that.

Usually, DDR4-3600 is considered the sweet spot. You don't want to loosen your timings too much to reach that (meaning, not set them ridiculously loose), but i don't know what timings you use, i mean just in general. You can raise DRAM Voltage to around 1.4V with most kits that have 1.35V XMP voltage no problem. Sometimes, even a slight voltage increase to, say, 1.37V may be enough to avoid having to loosen the timings too much. All in all, i would stay at DDR4-3600 and try to get the timings in order. As i said, a couple millivolts more are no problem for most RAM, they don't draw much power anyway, maybe up to 5W per module.

 

[jamie.jameleousunicor15a002e6]

I'll probably try that - I did a few experiments now I have HWInfo installed.
Curiosity got the better of me and I thought I'll see how far I can push the RAM frequency-wise first anyway. I got to 3866 with the same timings as 3600, then I got to 3933 with loosened by 1.
Bandwidth increased (though I realise when I've saved enough dual channel will give me much more performance but it's not an option right now). Only by 3866 voltage was automatically occasionally requested at .1.40V only on the detailed fabric on Cinebench then it would go back to 1.35V. This happened for milliseconds, twice in the render.
On 3933 I got regular 1.40V though it would still drop back, and the occasional 1.60 spike. Still - it passed memtest, I've been using the games which crashed at 3600 with tighter timings, I've ran compiles and everything is rock solid at that frequency.
I did try 4000MHz and loosening timings by 1 or even 2 again, but I got a black screen on post, however sometimes the HDD light would show for my boot drive and I'm just thinking as it's integrated graphics, the loose timings were just too loose for it to keep a display and it was probably functioning, and maybe even an actual graphics card would have resulted in a display. I think the absolute maximum on DRAM calculator was 4133 which was probably fairly accurate in that case, only 2 tweaks off that, and the latency increases won't make it worth it. I just wanted to see the magic number 4000 but... it wouldn't have gained anything over 3933 other than to show off and would have probably been the same or worse throughput.
Still.. after reading that F die RAM might not be good - a lot of those posts were from a year or two back, and maybe the process has been refined a bit as I wasn't expecting anything over 1 increase in frequency before instability.
Of interest to you will be the fact it is PC25600, 2400T downbin, so for it to reach these heights is probably actually an achievement.
I think the reason it has done so is that the chips don't use as much voltage generally. The motherboard has done a really good job on auto voltage, or the memory stick has done a really good job of only requesting what it needs.
I only have one question if I could've prevented a black screen (but system working by the sounds of it) by a SOC increase, as I've never increased it only the iGPU SOC. Maybe that would have given the memory controller a bit more but... I'm not going to bother with 4000 now, 3933 and a very stable overclock of RAM. Considering it is playing well with a +300MHz iGPU overclock and a PBO with no crashes - none, not the faintest data corruption yet. I realise longevity may be a sacrifice but, again with voltages being very good (I've seen posts where people have had to push 1.25 to the iGPU to get +300MHz or even +200Mhz), and with Core Optimisation enabled, I think I am very lucky overall. The only thing I haven't tried is LoadLine Calibration for the CPU to even out things, I've seen MSI boards like the number 4. I might give that a go next before anything else.

 

[citay]

And at DDR4-3933, is it running in "UCLK==MEMCLK" in the BIOS, meaning the memory controller is running at the same frequency of 1966 MHz, and not 983 MHz (shown in HWinfo64)?

 

[jamie.jameleousunicor15a002e6]

all three are running the same speed according to the Zen timings app at 19xx whatever it is (can't see the exact figure at the moment), the BIOS has auto but I think they're all synchronized as hwinfo shows FCLK the same as the memory controller clock speed aswell.