I finished off a Ryzen R7 1800X custom water cooled build just last week. So far it is superb (gaming performance is not an issue in the slightest... and as a workstation for scientific computational work, it is doing quite well).
Specs:
Ryzen R7 1800X Gigabyte GA-AX370-GAMING 5 16GB (2x8GB) G.Skill TridentZ DDR4 3600MHz 16-16-16-36 EVGA FTW GTX 1070 (stable at ~2150MHz core clock on water) Full custom loop using EK waterblocks (none for VRMs yet... hopefully soon!) Fractal Design Define S Samsung NVME 512GB SSD (Windows install) Seasonic x850 PSU Various old HDDs for extra storage Samsung 850 EVO 500GB (CentOS install for work)
I have a pretty solid amount of heat dissipation in my loop, and currently I have the 1800X at 4.025GHz stable with it hitting a max temperature of 58C - 62C @ 1.4V depending on ambient temps. This makes me want to go much further with the OC, but I'm getting nervous about sustained VRM temperatures as in my stress testing, I hit a max VRM temp of 92C after about 45 minutes of AIDA64. Thus, I was hoping to start a discussion here on experience/ideas/skills for overclocking Ryzen (and maybe we can even get @wendell to talk about his OC experiences with p-states too!)
Let's start by addressing the following from as many community members as possible:
People with a Ryzen setup, what are your exact voltage configurations for your given OC? Also, what are your load temperatures like with those settings?
Here is my response (please let me know if I should add/remove any details here):
Components
CPU: Ryzen R7 1800X
RAM: 2x8GB TridentZ DDR4 3600MHz
Motherboard: GA-AX370-GAMING 5
OC Settings
DDR Speed: 3200MHz
DDR Timings: 16-16-16-36
CPU Speed: 4025MHz
BCLK: 100MHz
Multiplier: 40.25
CPU VCore: 1.4V
CPU VCore SOC: 1.15V
DRAM V: 1.35V
CPU VDD18: 1.880V
CPU VDDP: 1.000V
CPU LLC: High
Stress Test Results
Load CPU Temp: ~62C Max
Load VRM Temp: ~92C Max
Load Method: 2 hours of AIDA64 Stress Test
I think 4.1GHz would be stable with some very slight tweaks to my current setup, but as I have limited time between work and having a life, I have not spent a lot of time 'fine tuning'. I am also excited to see how far this will go once I have better cooling on the VRM modules, since that is really the only thing that is reaching the point of dangerous heat at the moment. Please respond with thoughts, suggestions, or ideas as you see fit!
Thanks for reading (I'll post pictures of my setup as soon as gigabyte fixes some bugs with the LED lighting on this board).
I will be doing this once the "Gaming K7" motherboard is available, unfortunately the "Gaming 5" does not have an external clock generator so bclock tuning isn't an option yet for me.
It sounds like this will be the only way to get my RAM speed up to 3600MHz, if that even is possible. If it does work I think performance will be rather impressive. At 3200MHz DDR4 and 4.025GHz CPU, this platform's performance is great! Especially compared to the stock settings, I saw huge improvements just bumping the RAM speed up to the 3200MHz profile (which worked without any fiddling on the newest BIOS).
I know G.skill has some kits validated at 3466 using bclock OC, but all the ones hitting higher than 3k are samsung b-die. I'd check the nands on your sticks to see if they're that type before holding out hope.
No idea, still haven*t seen a tool for monitoring that on linux. My 1700 is running 3.8GHz on all cores with everything else on auto, RAM 2x8GB Ripjaws at 2666 on Asus Prime X370 Pro. Cooling is done with a Noctua NH-U14S.
I cant say I've had the pleasure of playing with the new Ryzen chips just yet, but I do have some advice for the VRM cooling situation. Absolutely make sure you have some kind of air blowing directly across them. This is something that can easily be overlooked with water cooling. Either rotate the rear fan to blow into the case across the VRMs, or look into either hanging a fan over the VRM, getting a mounting arm, something like that to direct airflow onto the VRM. Also, keep in mind these components are designed to run much hotter than any CPU or GPU, around 120c is the max I would ever want to see my VRM temp, with a average load temp much lower. Something else that surprisingly helped my OC on my 8350 was setting up a fan to blow on the backside of the motherboard behind the socket. This was back with my Antec eleven hundred, and at first I thought of it as more of a marketing thing but it did help a few degrees with temps. If you feel up to it a dremel and a thin fan like a sycthe slip-stream could help, but again new platform, socket, cpu, who knows.
I would be curious to know what you are using to determine stability. I know my system will post and do most tasks at 4GHz with ~1.37V on the vcore, but if I try to load the CPU heavily it crashes quickly (which for my workloads is rather important). This is why I mentioned wanting to hear people's specific settings. Things you could share that would help us get an idea are SOC voltage, RAM speed, and LLC settings. Then describe what you are doing to determine platform stability (AIDA64? Prime 95? 3D Mark? and for any of those, how long are you testing?).
All that being said, thank you for sharing, I appreciate any and all contributions!
Well first on CPUz stress test for like 10 mins straight and then also did GPUPI 100B on the CPU. Which 1. Takes a LONG time lol. And 2. Maxes out the CPU to 99-100% for the entire run. Iirc took like 4 hours or something to finish that one. I'd have to see. I have it saved ... Somewhere lol.
I agree 100%, this is very good advice and people trying to get into overclocking should take these words seriously. VRM temperatures are very important to keep in check, but as The_Drugs stated, these parts are tested to work at their specified tolerances up to 120C. That is QUITE hot, and most the time your OC won't get close to that point, but it definitely can when you try to push that CPU voltage to its limit (and don't have proper cooling setup).
The reason I brought this up in my original post is because I am nervous to push my voltage much passed 1.4V for daily usage due to the VRMs already getting up to ~92C under load at 1.4V. I am perfectly happy with that as a max temperature, but if I try to get 1.41GHz stable, or 1.42GHz and need 1.45V+, then I need to have something to adequately pull heat away from those VRMs.
I would recommend dropping that SOC voltage. I know it's a different board but there are significant risks of running SOC above 1.2v. I run 64GB's of 3600 at 3200 with an SOC of 1.15v.
Some motherboard have shown components dying due to SOC voltage and in theory 1.25v is shown to be safe but it's at the very limit. Any voltage spike and you could see issues
Thanks for noticing this, perhaps I should edit my original post - I had bumped it up to 1.25V for some short term testing purposes while seeing if I could get 4.1GHz+ stable and didn't set it back to the original setting before posting. My original OC was actually 4.0GHz CPU, 3200MHz RAM with the SOC at 1.15V. However, after a bit more adjusting this week, I found 4.050GHz was stable at the same VCore with SOC at 1.2V. I'm not sure why, but it seems like the SOC needs a lot of headroom on my motherboard. I don't think I want to stick to these settings, but the system is certainly stable.
I also have noticed that it seems the SOC voltage on this board drops by ~.025 V under load. So that might be why I struggle getting it stable as I push further than 4.0GHz on all cores. I definitely have more fine tuning work to do!
For anyone coming here to get information about Ryzen overclocking for your own platform, be careful with the SOC voltage going over 1.2V. But if you are running with RAM speeds above 3000MHz, it does seem like setting the SOC to 1.15V - 1.20V is a big factor in stability.