Ok, not the most exciting title, but bear with me. So I got some g.skill F4-4000C19D-32GTZR for my MSI Z390 Gaming EDGE AC. It's fully in the QVL as compatible at 2 dimms (2x16GB) on this board, at supported speed of 4000Mhz. And yet it's not passing the test I always use (reliably) to test RAM and system stability at 4GHz. The test ALWAYS unearths anything wrong hardware wise and my usual RAM passes fine (HX440C19PB3AK2/16) at 4000MHz. I know 32GB @ 4000Mhz is a little harder to get working reliably sometimes, but this board is supposed to be happy with this RAM at 4000MHz. Am I missing something? I tried to up the DRAM voltage to 1.37 and the RAM errors actually got worse. I'm currently testing each stick to ensure there's no faults and expect to find none - but it's supposed to work as a pair on this board. Ideas?
Yup and yup. Each 16GB stick is fine, no errors in HCI Memtest or TestMem5 - but soon as I try both sticks the errors fly all over. I have actually tried 32GB before - with Patriot RAM and the HyperX I have - but none of them work. Not at 4000MHz. I know the IMC is on the CPU, and I tried increasing the voltage of System agent and IO, but that didn't help either. I'm bewildered - MSI state clearly this RAM works at 4GHz at 32GB. But for love nor money I can't get it to run without memory test errors.
I was thinking about the IMC - wouldn't that affect even one stick of RAM at 4GHz? I have zero issues with any single stick, and zero issues with two sticks of 8 @ 4GHz. And one stick of 16GB is fine @ 4000mhz. Just not 32GB.
One ramstick is a lot easier to drive. A single ramstick is often used to get a system running to change settings in the CMOS because it is much more likely to function when there is a bad memory problem. I recommend trying 1.25V VCCIO and SA with a slight boost to CPU core voltage, and Ram voltage .05V higher than its rated voltage. (Try manually setting the ram timings if you get no luck) Lower VCCIO and SA to 1.22V if you still get no luck, sometimes these voltages can be fussy when too high as well. Once you get it working, reduce voltages as appropriate. ps are you sure you have the correct revision of your ram? I havent checked your mobos QVL, sometimes there are different revisions/versions, these matter.
Thanks for the reply. Yes, the correct revision is on the QVL. A little confused by your VCCIO and SA recommendations - they're reported as about 1.35v at present, you're actually saying lower them to 1.25v?
From newegg --- Pros: It comes with a 14/3200 ratio in XMP mode but was able to BIOS it from 9/2400 all the way to 15/4000 for gaming. Great flexibility. Best results at 12/3400...1.6V ('gaming' stable) Cons: After a bunch of tweaking and overclocking one of the two sticks stooped registering. swapped the sticks between each other and back to normal. no real problems for 6 months now. Overall Review: April 25, 2019 update: Running this kit fully stable at 4000 Mhz (16-17-17-37) 1.42 V (MSI MEG Z390 ACE mobo) in Memetest86! Also fully stable at 3500 Mhz (14-14-14-34) 1.42 V (ASUS Prime Z370 & MSI MEG Z390 ACE) Will try 4400 Mhz (19-19-19-39) and is stable will tighten the timings. This is what Samsung B die can get you! April 26, 2019 update: Fully stable at 4400 Mhz (19-19-19-39) 1.42 V (MSI MEG Z390 ACE & 8700K OC to 5.2 Ghz) P.S. My FBS = 200 Mhz May 3rd, 2019 update: Fully stable at 4400 Mhz (18-18-18-38) 1.42 V (MSI MEG Z390 ACE & 8700K OC to 5.2 Ghz) Fully stable at 4000 Mhz (16-16-16-36) 1.42 V (MSI MEG Z390 ACE & 8700K OC to 5.2 Ghz)
The CPUs IMC is fed VCCIO/SA voltage and itself has a maximum rated tolerance before it will start to age fast which can result in less stable ram at higher speed after a while and ultimately can kill the CPU if run too hot as well. That is 1.25V. (If the CPU is ran very cool it can be increased but that can make it more stable anyway so may not need an increase) Yet many motherboards feed higher than this which 'can' present problems when ram is on the edge of stability. You should use the minimum voltage possible once you have it running. I would set them to a max of 1.25V (which will likely end up slightly higher in reality). It might help or it might make things worse, but ultimately you have to live with it. Work from there. Also note: If your CPU is overclocked to its max, that can prevent high speed memory from functioning at rated speed. Set the CPU to stock speed until you get the ram sorted, then you can find your max CPU overclock again. If the CPU ends up running hot, that can cause ram instability so better cooling may be necessary. For example: My 2nd machine has a 6700K with a slow silent custom water cooler, a while back it was running ram at 3733c17 (overclocked from 3000c15, D Die). But to get that speed ram I had to reduce my CPU overclock 100MHz from 4.6GHz to 4.5GHz. If I ran the ram at 3600MHz the CPU ran fine at 4.6GHz. So I could only max one or the other. Temps were good, around 70C max. I decided to try a modern AIO cooler with 420mm radiator (much better cooling) and now the CPU will run at 4.7GHz and the ram at 3866MHz! I cant remember the temps and cant check because that AIO cooler is in my main rig now, its back to slower speeds. So, cooling the CPU better really does help max memory speed. ps the max VCCIO/SA it needed was 1.22V, It has never needed higher. This is dependent on your ram speed, the CPUs need and CPU temperature, it can vary a lot, especially if there has been heat damage or high voltage damage to the CPUs IMC.
Unfortunately I imagine this is single stick. I'm no closer to success with my dual stick but the info is useful - best results at 12/3400 hmmm... tbh I've been looking at performance at different speeds and amounts, and the differences are absolutely marginal.
tbh in Windows and in game I've had no issues - it's just failing testmem (and HCI Memtest) every time. Some points to clear up - not overclocking the CPU at all - it's at stock. Never even tried an OC with that. If you can indulge me a little, what do I prioritise? SA voltage? IO voltage? RAM voltage? CPU voltage? I'm hoping for a 'get this RAM stable at 4000Mhz on this board' for dummies... guide here. If I put the ram to XMP 4000Mhz, it runs in windows but fails testmem. What things do I adjust first to try to pass this test and get all-round stability? Thank you.
I gave the starting points for each, hopefully you can get it stable at its rated speed. Once you have it stable start reducing voltages to find the best settings. Its as simple as that, a long session of trial and error. If you cannot get it fully stable at rated speed, slow it down a little (ie use looser timing(s) or reduce the ram speed). Because you arent overclocking your CPU its voltage may not be that high. I suspect it will help to use a bit more VCore as a minimum starting point. But get VCCIO/SA voltages down to 1.25V asap unless you are water cooling the CPU. High voltage with high temp causes damage much faster.
Not sure off hand - I grabbed it from your exact kits reviews off newegg. ever run XMP with stock BIOS settings as to establish a baseline?
Thank you again - just to be 100% absolutely clear before I forage ahead with your advice - I have NO issue with 4000MHz with 16GB - hell, it happily goes up to 4400 with this G.skill AND I can reduce the timings as well with that RAM. The issues are ALWAYS with 32GB. It's not the speed, or the sockets - it's purely the amount - whether it's 8x4 or 16x2 this board says no to that at anything above 3700 despite the RAM being rated for 4GHz. I've tried 3 different brands of RAM and all of them were joyful with 16GB @ 4GHz - just not the full 32. Does your advice still apply?
Yes, it will only run as fast as it can with what you have. A higher amount of memory in a module is harder to get running fast. This is why there are plenty of very quick 16GB kits and not many 32GB kits that run at 4000MHz+. And the cost is often very high, better quality engineering and silicon are needed. Even then it can be troublesome as you are finding. Changing sub timings of the ram might help but its not a route I will go down, I dont have much experience beyond the first 4 basic timings + Ref Cycle and Command Rate. (life is too short) I will say this, if Command Rate isnt set to 2T automatically (also written as 2N) , set it to that. You can try increasing CAS by 1. Its the last thing you want to change (along with the MHz) but it depends whether max performance matters to you. You shouldnt have to mess with these when using QVL approved kit but it can help where problems exist. Its possible your ram or CPU is sub standard and a replacement will work better. But theres also a chance your motherboard is the issue. Its hard to know without identical replacement parts. Something to be sure of: Check your ram is being used in dual channel mode otherwise you can lose a fair bit of performance because it will halve the max bandwidth if not. ie make sure they are in a single pair of slots, not 1 module in each slot pair. Check the motherboard manual to find the slot pairs. They are usually 1 apart, ie not next to each other. You can also try putting the ram in the other pair of memory slots, this might give less issues.
Yeah so apply defaults first then go change to UEFI/CSM (which allows for legacy) plus just the basics you need to get your OS bootable then apply the XMP in tandem with your 32 GB kit. Also not to be a nitpick but for your motherboard according to your product manual verify you have your sticks installed in "DIMMB2 & DIMMA2" bout all I can think of for now... I'd proceed as so and try running memtest86+ in DOS (Not Windows) I repeat not windows. - If it's fine in DOS disregard errors in Windows as false positives. If you begin to OC disregard what I just said, you're in a whole different ballgame my friend and it gets all the more complicated. If it holds - great, start changing stuff little by little from there...
Hi, what makes the windows tests less valid? And also DOS? It's recovery environment or command prompt these days.
Sure no worries, the test in DOS is really simple as opposed to the possibility of when in Windows a crash could cause data loss.