After a long needed hibernation

[chew*]

here you go. simple proof. take a dud strap radiator to AC now its not a dud and that's amds auto ocing. simply by lowering load temps to 60c in cb.

the problem is 23C ambients are not realistic......and probably need 10c to get it into the sub 50c loaded range.

[chew*]

No magic here just temp drop and now its not a dud. all I'm doing is copying the voltage I see AMD using for peak xfr and running the peak xfr voltage and speed with lower temps.

Willing to hedge bets I get 4.5-4.6 out of the 3900x now under same conditions.

[chew*]

Here's how it works.

your cool cpu you run single core log voltage and peak xfr. the below is just an example. The 4474 run had 4474-4523 single core peak xfr @1.50v I shot for the lower for the all core as it seemed to be more guaranteed.



you apply what your learned account for droop and that is probably your max bench speed possible at X voltage at your current temps.

[chew*]

prime load temps with AC blasting through radiator and that's with cpu throttling back, can't imagine why nobody else is showing actual stability tests......I'd love to see those stock cooler24/7 "stable" prime 95 results

[muzz]

I thought I saw someone mention a 1.70 Bios for that board chew*, I could be wrong.

[chew*]

yah I flashed it, no change in temps not expecting a difference either. Lot of users claiming 85-95c real load temps ( avx512 etc ).

[muzz]

RM shows temps climb up to 85* just running R20 on the stock cooler ~4200 with my 3600.
Definitely not up to long term smashing, I can't imagine that cooler on the 3900X!

[chew*]

RM shows temps climb up to 85* just running R20 on the stock cooler ~4200 with my 3600.
Definitely not up to long term smashing, I can't imagine that cooler on the 3900X!

Yep and the issue is @ 95c it will thermal shutdown making it quite the pain to "stability" test properly.

Theres really only 3 options on the 3900x. Spend bookoo money on high end water. Toss a cheap aio on the ac like I did or settle for stock which will thermal throttle itself.

The 3600-3700x should be able to be kept in check on a dark rock pro 4 or equivalent noctua etc.

My n12 did a halfway decent job of cooling the 3700x in the summer heat no ac on so the ones I'm suggesting should offer overhead for overclocking.

Ive got some timings worked out on asrock if you want to try them muzz. Still dialing it in tighter but I have settings for 14-14-14 @ 3600 stable prime 95 custom with max ram allocated. your welcome to try. currently testing 14-13-13.

[muzz]

Been looking at the DR Pro4 and the NH 14/15(better mounting from what I hear), those things are gigantic, take up >1/3 of the board, and make my fairly Large Scythe Mugen2 look like a shrimp. I am a bit concerned that the 14/15 will hit my VCard, I've seen pics where it's right against it, but I'm not sure how far away my slot is.

[chew*]

Been looking at the DR Pro4 and the NH 14/15(better mounting from what I hear), those things are gigantic, take up >1/3 of the board, and make my fairly Large Scythe Mugen2 look like a shrimp. I am a bit concerned that the 14/15 will hit my VCard, I've seen pics where it's right against it, but I'm not sure how far away my slot is.

Theres a dark rock pro 4 on a x299. Only thing thats annoying is memory but still acessible.

[muzz]

Looks better, and isn't as wide as the D15. It holds it's own in temps as well.
Most likely what I am going to get, wish I just ordered it when I got everything else.

[chew*]

Looks better, and isn't as wide as the D15. It holds it's own in temps as well.
Most likely what I am going to get, wish I just ordered it when I got everything else.

here you go muzz asrock mem settings . 1.452 vddr 1.050 soc LLC 1. its prime hci and just about everything you can overkill test stable......working on faster now.

bank group settings are hidden in cpu features on asrock as well as some of the other options.

Power down disabled
Gear down enabled
Bgs disabled
Bgs alt disabled

[chew*]

now we are getting somewhere. nothing like seeing a little bit of time and effort pay off.

back to stability testing for this new tune.

before


after

[muzz]

Thanks for the settings, there are a bunch of settings now that I never even saw before , so much learning to do( whatever happened to 5-2-2-2 LOL- JK, I was around well after that). I haven't even had time to really mess with it, I just got the ram to run at it's rated speed, and haven't been back in there since, it's been running for a couple days on that boot.
I setup a new steam account, and downloaded some game demos to see how it does in games, and it seems fine at 1080P on my 60" Plasma.
I see 1.70 is much better, 20 seconds due to better timings. Sweet!

[chew*]

Those c14 3200 you bought are most likely bdie. My 3200 c14 set did same as these 3600 c15 so you can probably pull 3600 fine.

The second runs trfc was just a little to hot��. It dropped threads immediately in prime. I dialed it back an little.

[muzz]

AFAIK they are B, I tried to verify that B4 I ordered them, so I'm assuming 3600 shouldn't be much of a problem. I just wanted to do some other stuff on the kit before I started messing too much with it. I'm gonna be away from it this week, so I needed to make sure the VCard was OK in case I had to send it back, and it's been fine gaming, so I think we're good.
I need a new KB though, my K800 was giving me grief, so it didn't make it....
Using a PS2 mechanical right now, and it's horrible!

[PiLsY]

Hi chew*, good to see you posting man

TRRDs from 4 to 5 (and TFAW 25) is the key for getting over 3800 1:1 for me. Im using dual rank and the controller cant go stable over 3800mhz in dual channel+dual rank on either of my cpus without ruining efficiency (by slackening tertiarys too much). 3866 gets to about 600% HCI. If you can pull another 66mhz after TRRDs to 5 (dont forget the TFAW change too) then youre back in the black. I got 266mhz more at the same timings...mostly down to dual rank I expect but it should translate to single somewhat.

[chew*]

I gave it one shot 1/1 @ 3800 and it f9'd me. Right now I am seeing how tight we can go realistic volts( 1.45)/stability. Something in the 14-13-13 tune is not happy. Thought it was trfc i had it pulled in to 248 backed off to 312 then previously known good 345. Still dropping a thread. I think precharge 13 may be a tad to much for these sticks. I beat them up pretty bad on ln2 getting fast 32m times on Ryzen 1.

From what i understand dr is more effecient now on R3 due to way it acesses memory but my DR sticks went with the TR rig to michigan.

Edit: nope i have to pull back to 16-16-16 to boot 3800 or slam voltage into sticks. Neither of which i feel is beneficial for the 24/7 average joe crowd.

its not that I can't do it. timings are fine but needs 1.55vddr and for what i'm supposed to be working on I don't think those kind of voltages are going to fly with the powers that be and i'm ignoring unlinked as I am most likely dealing with the gamer crowd and video stutter is not going to be tolerated.

[PiLsY]

Did you intentionally set VDDP to 1.1? That should stay at or near 0.9v from what I understand and anything 1v or above can be worse unless you bring the cold. VDDG @ 1.1 seems a little high too, though I only have two cpus to compare against (both mine like vSOC @ 1.1v fixed with vDDG at or just under 1v). Also note that VDDG hard max is 40mv less than vSOC. If you have vDDG set higher than vSOC it will only run up to its maximum possible, in your case 1.05v-40mv=1.01v. I'm guessing they're probably auto voltage settings as the vDDP and vDDG at 1.1v thing is classic auto behaviour.

Sample 1 is an MSI B450i R5 3600 - 3866mhz 15-15-16-15-1T needs 1.0875vSOC, 0.9vDDP, 0.985vDDG and 1.41vDDR.
Sample 2 is a GB X570i R7 3800x - 3832mhz (3800 @ 101bclk) 16-16-16-16-1T needs 1.1vSOC, 0.9vDDP, 1.0vDDG and 1.4vDDR.

MSI board clocks noticeably better than the GB board, all 4 sticks are from my best 4 x 16gb B-Die set from x99 (3000mhz c12 1.32vddr 24/7 stable for years). I expected them to do much better than this to be honest.

Anyway, apart from secondary voltages and trrdS 5 + TFAW 25 I'm doing nothing special to get over 3600mhz solid. Maybe that's of some help?

[chew*]

honestly vddp was supposed to be @.92 and reported .92 in bios and in hardware monitors.

I had all the lower page voltages set. now I'm trying top voltages instead.

seems to me asrocks redundancy is getting the best of them. I requested they remove redundancy in past. they actuall have 2 pages for mem timings as well that will conflict.

I usually don't use this page of voltages as the larger array of voltages and llc are elsewhere.


this is what I usually use but seems to not be accurate in ryzen master.


none of that seems to matter and vddr is 1.45 so even higher, prime tells the story in about 10 seconds. dropped threads = imc not happy,

See what happens when i hit the asus board and GB board next. Last round on x370s the taichi was second most effecient giga was fastest asus and msi were slow in 32m

Taichi was not the most stellar @ high mem clocks but its effeciency made up for it.

[chew*]

ok got the taichi sorted somewhat. have to see what I can pull back in. I think cke 1 was to hot a timing for it. Any attempt at 1833 divider 1/1 = post code 07. bclk is fairly broken currently so didn't bother with that.

Most notable of all though..... after probably 8 hours 95C loaded on a AC cooled radiator. lol wtf? maybe ill swap blocks, 2 pumps might be heating up loop to much......

[PiLsY]

Doubling up on settings is not just as Asrock thing, theres a ton of duplicated settings in the Gigabyte x570 bios too, including a full duplicate of memory timings. That oc page does look a mess though.

Theyre all weird in their own ways, my MSI B450i greys out vddp and vddg unless you manually set infinity fabric. Gigabyte dont put vddg on the tweaker page at all, instead its in both the XFR and AMD overclocking sections in bios, and you need to set it in BOTH places for it to work. If either are on auto they override. Also certain manual fclk settings automatically set vDDG to 1100mv fixed in the XFR section and enable SOC OC mode. AMD overclocking section option remains at auto at all times. Oh and DDR power down mode is the same, two places, needs setting in both.

All these 3000 compatible bioses remind me of the kind of stuff I used to get when I was beta testing for Foxconn, functional for the most part but messy and unoptimised.

[chew*]

bios was identical on x370 and x399 redundant up the wazoo. boards pulling same antics to, I referenced my x399 thread on OC.net and sure enough that board also hated cke 1

[chew*]

Pictured next to an nh-u12s just to fathom the size.









Facebook score. Was cheap and bnib but my u12s was missing one pair of brackets i need to use it so ordering an am4 mounting kit.

Im sure it cools well but i dont get all the hype behind this thing. Its about as visually appealing as peanut poop.

I did not feel like dealing with drp4 mounting for multiple board and cpu swaps and since i already have drp4 and i like variety lets see what it can do vs ryzen 3000.

[chew*]

well I have sucked in what could be. it is what it is. feel free to try the settings out on asrock boards. It's certainly stable but not quite sure its as fast as 14 3600 tune.