You have to remember that I was looking at articles about the 7nm node from TSMC before the 3rd Gen Ryzen CPUs were released and I remember that piece of info being in a number of articles.
The thing is that so much time has now gone by that when I try to find those articles again they are buried and I just cannot find them again with the search terms.
They are still there, but I really cannot be bothered going through hundreds of pages of search results..
No worries, we already cleared this up in PMs 'member? Post's just been locked in moderation for that long.
I've actually just been using the CCX settings built into the 3103 UEFI for my ASUS Crosshair VII WIFI. No need to set anything in Ryzen master. Memory timings, and per CCX clock speeds are set in UEFI along with a 1.3 voltage. Results for my R9 3900X are below.
Thanks for the guide nec_v20.
I have one question about 3900x series.
Should i turn off Simultaneous Multithreading or should i only do that for 3950x?
Im mostly using my rig for gaming.
While i like when ppl take the time to type a guide, it should be labeled correctly, as well as state a few things more detailed.
its NOT the definite guide (not with one board/cpu/ram tested), uses settings specific to your board/cpu, nor does it tell ppl to test (ram) for stability, before doing cpu tweaks.
LLC (and other power related settings) arent the same for every cpu/board. what was proper for my one (high) was not enough for another (which needed turbo to stay below 2% droop).
with so many variables on ram (2 vs 4 sticks, SR or DR, mhz/timings etc) that ppl can have, it most likely wont be stable if ppl just "flip the switch" to do 3600@xmp voltage.
especially claiming that "quality" ram can do 1T (above 2133) is not true.
on 5 different boards from 3 brands with multiple kits from gskill/corsair/patriot incl decent micron e and samsung b die,
NOT one kit was stable with firstname.lastname@example.org and using 1T, on the samsung even using +1.4v did not help.
the asus board wasnt even able to do more than 2933 (no matter if 2 or 4 sticks) testing all kits i had.
so "blindly" recommending specific settings, WITHOUT telling ppl to run something like TM5 (memory test),
is one of the worst things to do (in a guide).
example: you say "..it should run at 1T.. if not change back to 2t.."
so how do ppl know? no info what they should do, not even mentioning that being able to run 1T, does NOT mean its stable,
especially if popl use something like memtest86, which doesnt catch 1T or oc errors.
3 rigs that were set to email@example.com/1T finished 4 passes, while TM5 would run errors within 5-10 min.
Hi - I recently reinstalled Ryzen Master now that they have gotten past the Win10 Hypervisor/Sandboxing limitation.
I noticed that there is an Advanced setting in Ryzen Master for Persistent PCD Values that I don't remember seeing in the past.
Does this setting allow @nec_v20's recommended modifications survive a PC reboot without needing to re-run Ryzen Master?
Just wanted to say @nec_v20 I really appreciated the guide. I've had my 3900x running at 4300 at 1.3v and RAM at 3600MHz for a while now. I disabled SMT as I have been mostly gaming on this computer.
Looking forward to the 5000 series CPUs, would I use the same settings in Ryzen Master, or is there some adjustment needed? Will you post a guide for the new CPUs?
With precision boost overdrive 2 and curve optimizer, you can set your all core boost to 1.3V using the PPT/TDC limits in UEFI. No reason to reduce performance on lightly threaded workloads, and you'll wind up with similar results to manual in multithreaded. No reason to load ryzen master every time you boot.
From default UEFI settings first set your RAM speed and timings in "extreme tweaker".
Then go into the Precision Boost Overdrive settings under AMD Overclocking.
Turn on Precision boost overdrive, and manually set the scalar to x1, auto overclocking to +100 and the PPT/TDC/EDC limits to 142W/95A/140A. These are the default settings for a 105W TDP Processor.
Run a multithreaded load like Cinebench R23 and note your temperature and voltage on the multithreaded load (Ryzen Master, HWIFNO).
Also run OCCT small data set/extreme/constant workload (1hr), and large data set/extreme/variable to check for errors. If everything is fine, proceed to the next step.
Now, slowly raise the TDC. Also raise the PPT by 1.5W for every 1A of TDC. So at 105A your PPT would be 157W. EDC is the short term boost amperage allowed over the TDC. I tend to keep this with 20A of the TDC.
Run Cinebench or your favorite benchmark again after each increase. Monitor voltages on multicore load and temps again using Ryzen master or HWINFO. Keep doing that until your voltage is hitting 1.3V on multicore loads or the temp is where you want it to be.
Once that is done, you'll see that you are bound by TDC or temperature in multicore load . You can then reduce the PPT such that both are near 100% usage when running a benchmark.
Run both OCCT sets again to test the stability of those settings. I wound up at 215W/140A/160A. I am at 1.3V multicore load and temp is at 70C. TDC is at 100%, PPT is at 94% when running a multicore load using those settings.
From here, you can move to curve optimizer under AMD overclocking and try to get a bit more out of the processor.