gorpo wrote:
@gupsterg Thanks! It's great how well you document your tests.
Comparing your 3200MHz MLC runs with mine (), my latency is lower and yet so is the bandwidth. For the Loaded Latencies section, my results start with higher latency and lower bandwidth than yours, but end with lower latency and similar bandwidth. Any ideas on how to interpret these results?
@Mordorr
That is weird... Yeah, try another USB drive.
What file did you save on it in UEFI? Could Win10 be hiding it due to file extensions? Maybe check it in linux?
@arvinz Glad you got it working. Formatting and USB size/type seem to matter.
Thx, I'm still trying to optimize as best as I can. I feel like running the CPU in auto for the most parts is yielding the best performance/cooling than doing manual OC. In terms of ram though, I feel like I could get my ram higher but just not sure on the settings. I am able to also run The Stilt's 3333 Fast setting but can't for the life of me do his 3466 setting, again, there might not be much performance increase to that.
By the way, how do you link a text file? I tried to upload the bios setting text file here but it wouldn't accept it.
gorpo wrote:
@gupsterg Thanks! It's great how well you document your tests.
Comparing your 3200MHz MLC runs with mine (), my latency is lower and yet so is the bandwidth. For the Loaded Latencies section, my results start with higher latency and lower bandwidth than yours, but end with lower latency and similar bandwidth. Any ideas on how to interpret these results?
Sorry I do not know how to interpret results in IMLC. I have not found the time to understand IMLC results. IIRC the PDF with app may have some info.
I ran these today.
Currently I have lowered tRFC from 333 to 297, tWR from 12 to 10, without issues; tRTP from 8 to 5 or 6 is causing issues. Once I've seen if I can gain any others lower I'll rerun AIDA64/IMLC.
Cache runs at same speed as CPU MHz on Ryzen, AIDA64 will reflect this in relevant test boxes.
Ever so mildly I do believe CPU speed affects the RAM related results in AIDA64, biggest gains I get are from RAM MHz and timings though.
arvinz wrote:
Hey gupsterg-finally managed to save my bios settings to USB. The culprit was my 128GB USB 3.0 USB stick formatted to exFat. I don't think the BIOS likes that. I used an older Fat32 formated 8GB stick and it worked.
I've attached my latest settings, currently using The Stilt's 3200Mhz Fast setting but added a bit more DRAM Voltage to 1.355. Please have a look and let me know if you see anything that can be improved to boost performance. Also I haven't been able to get to a stable 3466Mhz but not sure if it's going to be much better than the 3200 Fast setting.
Here's my BIOS settings text file:
https://drive.google.com/file/d/1mW2gZbpagukttvn_mPjXlXE7Ghp5e4tC/view?usp=sharing
~+0.005V extra to gain The Stilt 3200MHz Fast setup seems about right. Even though the dimms I use currently can do 3533MHz using The Stilt 3466MHz timings at 1.35V and command rate 1T, they can not do 3200MHz using The Stilt 3200MHz Fast timings without extra juice.
Firstly if trying to gain RAM OC I'd roll down Performance Enhancer from [Level 3] to [Default]. I'd also set Ai Overclock Tuner to [Manual].
As you have not elaborated on how 3466MHz fails I'll just state some general points based on what I'd experience on my HW.
If the board POST and the POST code gets stuck at some code and I see no display on screen I know I'm wildly out. I've also had situations that I'll see a message stating UEFI is corrupt and restoring "blah blah blah", usually if I hit reset at these messages board will recover. I may need reasonable jumps in SOC/VDIMM to get board to POST at targeted settings in these situations.
If after those jumps in SOC/VDIMM I get Q-Code: F9 at POST it could be that ProcODT and or RTT settings or something else is hindering training.
I have experienced Q-Code: F9 on initiating restart in OS after having done say a stability test which passed. When this occurs on current HW/targetted settings, ProcODT was incorrect. For example whilst on 3533MHz I set ProcODT 48, based on previous experiences this had worked well. Everything functioned as it should, until I did the restart in OS, bam I had Q-Code: F9, I set ProcODT 53 and all was well.
Thing I have noted before and recently, sometimes increases in voltage yield stability, but it maybe masking a lack of tuning of other settings. For example when trying to gain 3600MHz I set RTTnom to 34 (RZQ/7), I leave this at [Auto] for 3533MHz, which defaults to [Disabled]. When I applied RTTnom 34 to 3533MHz profile I was able to lower SOC from 1.025V to 1.012V.
Recently when I gained reasonable stability in limited testing for 3666MHz I decided to apply the RTTpark of 80 (RZQ/4) to my 3600MHz profile. Even though it has not allowed to me to gain lowered SOC/VDIMM, I believe it has allowed to me to gain some tighter timings. Below is current profiles I have saved in my UEFI, it excludes 3600MHz with improved timings as I'm still working on it. 3666MHz profile still needs some work.
Another thing I have noted is as I reach the max this CPU is doing RAM wise (ie 3600MHz/3666MHz) I'm finding I may sometimes have to load the "Base Profile" and then apply 3600MHz Profile to gain back stability when I did something wrong.
For example today when I lowered tRTP to 5 I had a PASS in RAM Test, I then shutdown PC and retested and had failed RAM Test. I then set tRTP as 6, test failed again, I reverted back to 8 (which is what it was prior to lowering) and it failed again. So then I applied "Base Profile", then 3600MHz profile and it has passed using tRTP 8 repeatedly.
gorpo wrote:
@gupsterg Thanks! It's great how well you document your tests.
Comparing your 3200MHz MLC runs with mine (), my latency is lower and yet so is the bandwidth. For the Loaded Latencies section, my results start with higher latency and lower bandwidth than yours, but end with lower latency and similar bandwidth. Any ideas on how to interpret these results?
Sorry I do not know how to interpret results in IMLC. I have not found the time to understand IMLC results. IIRC the PDF with app may have some info.
I ran these today.
Currently I have lowered tRFC from 333 to 297 (165ns), tWR from 12 to 10, without issues; tRTP from 8 to 5 or 6 is causing issues. Once I've seen if I can gain any others lower I'll rerun AIDA64/IMLC.
Cache runs at same speed as CPU MHz on Ryzen, AIDA64 will reflect this in relevant test boxes.
Ever so mildly I do believe CPU speed affects the RAM related results in AIDA64, biggest gains I get are from RAM MHz and timings though.
Mordorr wrote:
Well, we are always learning...
I found the problem: forget give a name to the file! lol
Here are my actual config:
Glad to read you got there. I'd zip txt and attach to post, otherwise post becomes large :eek: ....
gupsterg wrote:
Sorry I do not know how to interpret results in IMLC. I have not found the time to understand IMLC results. IIRC the PDF with app may have some info.
I ran these today.
Currently I have lowered tRFC from 333 to 297 (165ns), tWR from 12 to 10, without issues; tRTP from 8 to 5 or 6 is causing issues. Once I've seen if I can gain any others lower I'll rerun AIDA64/IMLC.
Cache runs at same speed as CPU MHz on Ryzen, AIDA64 will reflect this in relevant test boxes.
Ever so mildly I do believe CPU speed affects the RAM related results in AIDA64, biggest gains I get are from RAM MHz and timings though.
Yes, I took a look through the PDF documentation for MLC, but I can't say I understood too much.
So AIDA64 and MLC give somewhat different throughput and latency results. I am just using 2 x 16GB @ 3200 MHz DOCP and haven't tuned timings at all. Maybe at some point I'll try the Ryzen Calculator suggestions out of curiosity, but I don't know anything about timing parameters, want to keep my system stable, and don't want to spend too much time on tuning.
