The AMD Ryzen All In One Thread /Overclocking/Memory Speeds & Timings/Tweaking/Cooling Part 2

Do not know how or why but they are working on this particular rig.

 

Might have to do with my memory. I have duel rank 32GB(16x2) kit.

 

5GHz
As of ZEN2 we can have similar Lows 0.1% and 0.01%.
New games writen with Zen in mind (game cache) will run better.

 

Here my todays Test
AC:O mixed settings (towards Ultra)
Vega @ 1658MHz 1.081v 1120HBM2 918mV (CAP to 70FPS)

CPU is in ECO Mode (45tW)
I find it better for everyday Gaming, not a single FPS loss, same 81+ FPS in 3Dmark FS
It needs less V, generates less heat and is more stable at Manual OC.

4300MHz is enough for AC:O

Spoiler: CPU settings

Spoiler: The in-Game bench

Update with new Win version of 1909.
A notch better, now same pts. w/4300Mhz as before at 4383MHz !

Spoiler: The in-Game bench WinX v.1909

 

That looks fine, but what about frame pacing, it was all over the place with RX5700.
I never get it smooth no matter what I tried as you know.
Even when zooming in with a bow equipped and zooming back made fps drop from 60 to 50fps for a second, everytime.

 

More for me, became curious how stable the ram settings were. Ran MemTest over night. CPU was: offset -0.08125 | SOC offset -0.03125 | DRAM Voltage 1.3900v | CLDO VDDG .9250v

Spoiler

 

Nice!! I can get 3733mhz memtest stable but I have booting issues with it intermittent. When I go to power up the system it will power on then power off three time before fail booting into F1 BIOS. Other times it will boot fine. If I drop to 3666mhz or lower no boot issue. But like I said Memtest stable so I'm little lost as to whats causing it. On other note. I take it you running the 3700X at 4.4ghz? No issues? how did you test it? Gets hot with Aida or Prime

 

No all core oc. Just running the cpu as is. Only cpu performance item enabled is an option ASUS labels Performance Enhancer. There are three levels. The level that it is set at is labeled 3(oc). All of PBO is left at default settings. Ryzen Master shows that that setting sets 4096 for both PPT and TDC along with 145 for EDC. Mem Test gets the boost to peak at 4.425 and averages 4.400 overall. Test Mem5 sees the same boost results. Gaming though does see a peak of 4.500 from time to time but mainly at 4.45 to 4.475. 4.425 usually sees a voltage of 1.387 with 1.385 rarely being seen for same speed. Still playing around with the cpu offset setting. Looking to see what it would take to keep core VID voltage from ever hitting 1.500. Mem Test and Test Mem5 never see that but gaming does. Temp wise, have not seen it hit above 65 while gaming. This is the first cpu that has never felt Prime95 or any cpu stress test. No BSOD or any hiccup related to cpu stability. RAM has been the only hardware stressed this go around. Feel the 3000 series does a good enough job policing itself even when allowed to exceed specifications.

There is a tradeoff with lowering the cpu voltage setting. The lower one sets it, the lower the boost becomes and core VID voltage increases. Letting the cpu voltage setting stay on Auto, have not seen core VID voltage hit 1.500v but the reading for the cpu voltage setting does hit 1.500. This is the balancing act I am currently playing with.

Not every cpu is going to be the same. I feel that CAD Bus Drive Strength settings tend to have an influence on the internal policing communications. The cpu did not show much activity along those policing lines until the Drive Strength settings were set at 24-24-20-20. It could be said that the cpu actually woke up or no longer felt like it was being held back. The odds are the setting that does not hinder the internal policing communications will be different for others.

 

-> https://twitter.com/1usmus/status/1194264084230918144?s=20

 

I'll try and see how those setting work on my CPU.

 

How long do you guys test memory stabilty for?? Reason I ask is I run Memtest for 1000% per thread. Figured that would be good enough. But I left it on past that and ran into 2 errors at 1344% and 1564%. Make me wonder if I should go longer and if so how long?

 

When Memtest was first run, ram voltage was at 1.400v. Ran into 1 error on a single thread at 18XX%, cannot remember exactly but it was in the 1800s. Raised the voltage to 1.4100 and an error popped up at 553%. Then lowered the voltage to 1.3900 and let it run overnight. When woke up the percentage was where it was. Was initially thinking that 2500% was good enough. So running it while asleep was the goal.

 

Tried offsetting voltage.... BSOD. raised CPU offset alittle CPU was: offset -0.03125 | SOC offset -0.03125 | DRAM Voltage 1.3900v | CLDO VDDG .9250v but any offseting get me less performance..(full load CB15 offset 4.075-4.10ghz vs PBO 4.10-4.125ghz) and in games I'm pulling 4.325-4.375ghz....so odd same board, same CPU but your getting better Boost

 

Know it does not help with boosting, but was that 100MHz bus clock fix profile applied? Loaded the profile from a USB stick, booted, powered down, booted into bios and set everything as normal. As can be seen the bus clock no longer reads 99.8. It was a small fix but it did not hinder performance any, it actually improved it. CPU and SOC LLC is set as Auto | CPU is set at 130% and Ultra Fast Power phase response. SOC is set at 120% and optimized response. The VRM frequency for both were left on Auto. In the CBS section C states was enabled, IOMMU-Disabled, and ECC-Disabled. The rest were left on Auto. Also, using the Ryzen Balanced power plan. Not sure if any of this will help but these are the only settings have not mentioned.

 

My DIMMs work better if I set a fixed voltage and set SOC LLC to 2 with capability to 120%, all phases active (called Extreme on my board). Especially at higher frequencies my RAM didn't boot into Windows when voltage was set by offset, resulting in a sudden reboot.
Regarding "better values" you may reduce latencies a lot by getting the optimal voltage for CPU + RAM + SOC.
AFAIK you can check for the correct voltage of SOC by running AIDA cache benchmark. If the bandwidth of the L3 cache differs a lot (say more than 50GB/s) between READ and COPY your SOC voltage is too high or too low. Getting this almost even gave me 2ns less latency (69.x instead of 71.x).
My difference between READ and COPY was more than 150GB/s when I first checked.

I am on X370 btw. , but the procedure and outcome may be the same (less latency, how much exactly will differ per setup).

 

Best desktop CPU?

-> https://www.pcworld.com/article/3453320/amd-ryzen-9-3950x-review.html

 

It looks really impressive. I especially love the low power numbers.
It's using less power than the 3900X !!

 

CPU wise this has helped alittle. Still only getting random peaks of 4.425ghz... In CB15/20 all core is now 4.125ghz stable and peaking 4.150ghz Single core is 4.4ghz. Gaming seems stuck at 4.350-4.375ghz but that in the Division2 will check other titles. Havent check Memory stablitiy yet. Still tweaking.
Update: UV got me a few more points in CB... but Boost only hits 4.4ghz now instead of 4.425ghz

 

386SX linked this for me: https://www.reddit.com/r/ryzen/comments/ahtubu/xpost_from_roverclocking_demystifying_memory/

A lot of the esoteric settings like all the CAD stuff, was nice to finally get some decent information that was understandable. Now I know why setting 1s in all the CAD Bus settings provided stability. It was at the cost of additional latency; which explains why 0s are preferred instead of taking what the DRAM Calculator spits out as gospel.