In my above mixed module config at XMP setting of the 3000 MHz RAM, it has booted into Windows and appears to be working okay. Need to test it properly. I had thought CPU focused benchmarks should have given more noticeable indications. I shall give games a try. Shadow of Tomb Raider have a nice built-in benchmark and seems to be a favourite among reviewers. From 120 to 130 is a significant gain.
It seems odd that a cpu intensive benchmark like cinebench wouldn't get affected by ram, but it's negligable. games and production tests are more likely to show performance games from ram. pc mark 10 might be of use
It is. Adata Gamix D10 XMP configured to be 3000 MHz 16-20-20-38 1.35v. It's dual rank Samsung chips. New kit is Crucial Balistix rainbow-puke, XMP configured to be 3600 MHz 16-18-18-38 1.35v. It's dual rank Micron rev E chips. I've previously had tested the Adata kit stable at 3400 MHz, but seen some whole system crashes so backed off to stock. But crash still happened I think it's more to do with Adia64 and the Asus motherboard I had, no longer run Adia64 in the background stopped the crashes. No crashes so far with updated Aida64 on Gigabyte mobo. Currently configured as slots 1, 3 - Crucial and slots 2, 4 (the termination point) - Adata. At 3200 MHz rest of timing using Adata XMP. I have successfully completed 1 hour AVX using OCCT and overnight 10 hours TM5 with extreme config. https://github.com/integralfx/MemTestHelper/blob/oc-guide/DDR4 OC Guide.md#recommended Currently testing 3400 MHz same timing. If this passes after a full day of testing, I'll back off to 3333 MHz and it'll be a good 24/7 compromise between speed, stability and capacity. --Edit: TM5 found an error, no crashes, at 3400 MHz. I'm going to run 24 hours test for 3266 MHz in preparation for 3200 daily use. (also need to do more testing Crucial-only at 3600 to see how much performance difference is there left on the table)
Did a bit more testing. I'm torn whether to keep the Adata 3000 kit in the system or not. Note, the 3000 MHz testing was done with 64GB of RAM, whereas previous baseline was done with 32GB. On one hand, the 9% minimum gained from going to 3600 MHz would be worth the upgrade in speed. But on the other hand, I like RAM, can never get enough RAM. 64 GB would allow 42 GB of RAMdisk. Increase from 20 GB RAMdisk currently, not that I will make full use of it........ But I always bought more than I needed, I had 16 GB back in 2011 with i7 2700k Sandybridge. I hope to keep this platform for at least 5 years. Current plan is to keep the slower kit to run 64 GB at 3200 MHz and see if I can get any 3600 CL16 kit for under £100. Can Ryzen 5000 series infinity fabric drive 4 dual rank sticks at 3600 MHz? I know anything above 3200 MHz is considered overclocking.
Does your OS not allocate RAM disk memory as it gets used? I've a RAM disk (several, actually, but only one mounted at /media/RAM Disk) and if it's empty it takes up zero RAM. Well, not zero, but functionally zero given the overhead's measured in kilobytes and I've 32GB of the stuff.
Not a bad increase considering you are still XMP and CL16, regarding your stability issues on quad stick OCing did you check the temps, I found that once I was at 4 sticks I had to put some air over the RAM (1.4v so a smidge higher than the 1.35v required ) as once it got over 50C which whilst not very hot I would see mem fails, so I am using an old Corsair RAM cooler, keeps me stable with an OC, can't be 100% sure the RAM sinks themselves contact the chips well on my tridentz as one just slid off one day, so might be why. With some tinkering you could probably extract another few % if you have the time for endless testing, Crucial ballistix thread on overclockers probably a good resource to get some short cuts to timings and vvoltage to try. Certainly was useful for tuning my daughters APU rig. Can't really comment on Dual Rank OCing but 5xxx is running 4 single sticks at 3800 with no bother here bar the cooling.
It does. For example it's currently sitting ~8 GB used. I just had instances (twice or a few times) where I filled up the 20GB and soon after I also got low RAM warning. I use the RAMdisk limit as soft stop to not worry about RAM usage within reason. Obviously it didn't work because 16GB is normal and 32-20 is only 12Giggles. During RAM testing (it is constantly being hit), I was also lightly loading the GPU because MSFS lacks other download system. Highest temperature reported by Curcial built-in sensors were close to 60c. I did some quick google and these Micron chips appears to be happy with even 70c heated by hairdryer. No idea about temperature tolerances with Samsung chips on Adata though. But I bet the 3400 CL16 error was inside this kit. So won't be increasing the 1.35v DRAM voltage. Any voltage or other tweaks to the CPU side could try to help make 3400 CL 16 more stable? I'm not sure any cooling could actually cool RAM. I've been trying to reduce temperature on the NVMe drive between CPU and GPU for a long time. Without pointing a fan at it, no matter what heatsink I use, it just doesn't make any difference. Once in games or when both are generating heat, the SSD goes above 55c because both components next to it are at or above 60c. I've finally ordered a gen 4 M.2 riser to move it out of the heating zone.
Okay. Done 3 overnight tests (1 TM5, 2 Prime large FFT) and countless OCCT 1hr memory tests throughout the day. It's rock solid at 3333 MHz. This will be my 24/7 going forward. Personally, I deem the jump to 3600 isn't worth it, only giving up to 10% at best, 0% at worst, 0-5% in game benchmarks. So I think anything around 3200 MHz is fast enough to not bottleneck the 5800 CPU. I've ended up with 64GB RAM in different slots to my previous post. Slots: A1 = Adata 3000 A2 = Crucial 3600 B1 = Adata 3000 B2 = Crucial 3600 This configuration took a lot longer to fail tests than Adata at slots marked as 2, at 3400 MHz. Signal termination points are at slot numbered 2, hence 2 sticks should use these slots. Please correct me if I'm wrong, I believe the termination points are most important for signal integrity, so the cleanest signal generating IC's should go at the termination point. For cooling, I've found a tiny 40x10mm Noctua fan, placed it at the end of RAM sticks and placed a motherboard thermal probe on the RAM, make the fan only turn on when RAMs go above 50c. Prime test makes RAM go to 60c, with the fan at its inaudible minimum speed (2200rpm!), RAM only goes up to 55c.
Little fan should do the business, I know it seems like low temps but I certainly have seen instability on my B die ram once it goes over 50 despite what it is specced for, but I am running it with an overlock and tight timings so am no doubt fighting the performance over PVT. Theres only the DRAM voltage and the Soc voltage, beyond that on my ballistix slackening the TRCD seemed to help run a higher clock, and keeping the RFC quite high compared to other sticks, probably have settings somewhere from painstaking weeks of testing because I am that interesting