After a long needed hibernation

[chew*]

Above mem clock/timings failed prime. Not surprising. It was from ryzen dram calculator which never worked on taichi x370 nor x399.
tested repeatedly same 16k iteration same exact cores.



Works on giga and more than likely asus since I think thats the brand the author uses.

Taichis always been odd man out and the natural response is to run looser but...you would be wrong. Hashing out manual profiles now for the board. Might be usefull to some endusers. Seems all the memory profiles that i built on x370/x399 port right over so that should save a crapload of time.

[chew*]

Trend continues cross vendor and all the way down to slacked 3200 mem speeds. Now ive swapped memory to gskill royals. if same outcome I think the chip may be dieing/degrading.

If that's the case that's a first for me in 10 years, yet to damage a chip testing on air/water especially in a case where I have hardly even oced it.

[chew*]

Finalized settings on the giga Aorus master. Included a screenshot with voltages and timings. I also took out my 15-15-15-35 3600 gskill kit tossed in a set of gskill royal 16-16-16-36 3600 in used the same identical settings and its currently 15 hours in passing prime so it appears bdie is bdie regardless of bin if settings are tuned well. The only other setting you might want to play with is ODT but its working so I have left it alone.

I should also point out that I ran some aida and noticed interesting results when comparing both sets of memory. Seems we still don't have some access to some hidden timings that can effect performance as everything else was identical. I'll post those results up later.

Also should point out in the 24/7 stability department on r1 I found less is more when it pertains to vdimm and the trend continues on r3. I wonder if this is what pilsy was experiencing on his aorus as 1.45vdimm passed with fan on memory and failed without. 1.4v passed without a fan.

[chew*]

Ported settings to gskill royal = pass.

Also locked frequency and retested aida and the anomaly went away so must have been a fluke with fluctuating cpu speed or maybe origin was doing something in background.

[chew*]

probably overkill but I ran hci 2000% coverage as well.

[chew*]

I explored a few things today as far as PBO goes and people claiming it never hits peak rated PBO speeds without some modifications.

So first up we tested AMD overdrive creator mode auto overclocking.

For all testing affinity was set to core 8. c05 = core 8 so this is not an idle core but a loaded core.

32m pi 4.6 exceeded.

cinebench could not exceed 4.6





After finding an app (32m pi) that could exceed 4.6 I tested without AMD overdrive and just my memory tune.



Armed with this info and just simply knowing AMD all to well from my experience of working with them for years I ran one last test with defaults loaded and no memory/NB clock.

Nailed it. Just as they most likely tested and in a lightly threaded app as they claimed. Highest boost recordable is achieved at completely stock and I saw higher than I was able to screenshot.

[chew*]

Taichi settings. 1.4vddr leave vtt auto.

Ran overkill stability tests as usual.

testing gskill royal 3600 to make sure settings port from 1 set to another now, post that up when its done.

[chew*]

new driver out for AMD vgas.

old driver


new driver

[charged3800z24]

I had the exact same finding with the 3700x and 3900x on the turbo. Also used Super Pi to validate my findings.

[charged3800z24]

Also, nice run.

Here is my 3900x - Radeon VII Stock run on new driver.

[chew*]

Forgot to post these up for the royals. settings ported fine.

I went to my shop and stole my other set of 3600 out to test 4 dimm matched sticks operation.

Haven't decided which is the culprit yet but so far no luck running even remotely tuned on taichi.

[chew*]

Also, nice run.

Here is my 3900x - Radeon VII Stock run on new driver.

Were you in game mode or legacy mode or manual OC? @13xxx cpu score you would slaughter my 5700xt stock run.

Speaking of which "game" legacy mode does not boost as high for whatever reason. Possibly 1004 agesa solved native default boost operation but has not addressed it in all modes of operation as of yet.

[charged3800z24]

Was not using anything but defaults. Clean loaded win 10 1909 and enabled AMP profile for RAM. BIOS vF10 for x570 Aorus Elite.

[chew*]

Was not using anything but defaults. Clean loaded win 10 1909 and enabled AMP profile for RAM. BIOS vF10 for x570 Aorus Elite.

Yah something is off and robbing you. I doubt its the memory/tune its not worth that much. My guess would be either were warm and throttled cpu or out of curiosity are you running a ryzen high performance power plan?

Here just for an example. 3800x creator mode auto OC which puts the clock speeds almost on par with 3900x base/boost. It puts this 8c chip way to close to your 12c cpu score. Im guessing a 13k cpu score run puts you 10,2-10,3 in timespy.

[charged3800z24]

I do have Ryzen High Perf. selected.

[chew*]

well that seems odd then. This isn't like a tweaked OS setup for benching. It's also pretty dirty from driver board cpu swaps. Something appears to be hurting the cpu score.

[charged3800z24]

CPU-z and Cinebench both are good and Turbo is working as expected as well... I re-ran the run a few times even with some tweaks and the score is now 1000 point lower every time.

[chew*]

That is really odd.

I ran prime for 22 hours with the auto OC plus the memory profile. Seems it can swap between boards cpus and high quality bdie and just work so I'm calling that a wrap.

Shoved my 3600 c15 4x8gb back in my intel rig ill just be using the royals from this point forward as it does not seem to matter which sets of bdie I use. I'm far from impressed with the 4x8gb operation in the AMD boards still so i'll pick up a set of 2x16gb as DR always seemed to work better on r1 as well vs 4x8 setups.

[charged3800z24]

Chew*

I monitored the CPU in Task Manger during the Time Spy Benchmark. The CPU test hit 90% at max and was for barely a second shortly before end of the test. It sat near 70-80% the rest of the test.

[bluestang]

Yeah, I'm having difficulty getting my TridentZ b-die 3600C15 4x8GB stick to actually run at 15-15-15-35 stock settings. No matter what I try its at 16-15-15-35. Geardown disabled and system won't boot lol. 3900X on a Crosshair 7.

Time to stick in my 2x16GB Neo 16-19-19-39 and call it a day

[charged3800z24]

Latest AMD driver, 2050 core / 1200 mem on the Radeon VII still stock CPU. Look like my CPU is functioning properly now

[Alex-Ro]

Looks like he is again stuck in the basement for 6 weeks. Was curious if he tested the C8 impact and if he did find any timings difference at speeds he tests versus normal boards.

[MrIcee]

Hiya Boys

Icee has been bored so changed out my i7-5960X rig for my beloved AMD, specifically a 3900x on a custom watercooled loop utilizing EKWB and Optimus parts. I didn't end up winning the silicon lottery, much like the old days as this puppy takes a lot of volts to run 24 threads, and similar to run up to 4550 on 12 cores/12 threads. Regardless, not too shabby bandwidth with this puppy at this point It's stable enough to run 3D, and I have a 5700 XT on board watercooled with an NZXT 280mm AIO for the time being. Have been planting it pretty deep in Intel territory in some of the 5700 XT 3Dmark benches, some 1st's, top tens and some not high enough yet, but having fun and that's what counts. Welp, here's a screenie for ya's

Randi

[FlanK3r]

it is very nice Ryzen 9 3900X, if is it stable!

[OPPAINTER]

MrIcee, Good to see you my friend
I remember I owe 5 bucks Randi, as soon as I get my $1200, life sucks and then you die check I'll pay paypall you.
Lets do some Overclocking!!