After a long needed hibernation

[PiLsY]

Nice quote, thanks for that dude

I got screenshots from support showing where they were measuring vDDR from, however they didnt show which ground point to use. This would indicate they were either using the psu ground or an external ground. When I measured using the memory plane ground I saw +16mv compared to using one of the 24 pin grounds. In my case bios set 1.35v, reported 1.404v idle, sampled 1.399v idle memory plane ground and 1.382/1.384v idle 24pin ground. I have raised this point in my reply to them.

There is no readout for vTT and support have given me nowhere to check it in hardware yet. It is possible that vTT is undervolting and vDDR is overvolting and this could be magnifying the issue. I need to check this next, but will have to ask support for the read point and ground.

Derailing chews thread a bit here, old habits die hard lol. Sorry man

[bullant]

Good to see you around posting Chew

[Nizzen]

Any news about Asrock Taichi ABBA beta bios?

MSI have ABBA betabios out: https://forum-en.msi.com/index.php?topic=323855.0

I'm using x570 Taichi with 3900x.

[systemviper]

wow this is going to take a while to read, got to go find a park bench and enjoy the good weather.



great info

[chew*]

ive been busy and waiting for some memory [ gskill royal 3600? ] im supposed to tune the platform on but since new agesas keep popping out downtime does not matter till agesa updates settle down since you need to toss all settings most of the time and start all over from scratch. My sticks are somewhat flaky/tempermental from ln2 abuse on ryzen not to mention something more current and available fits the criteria for user loadable OC profiles.

As far as the product goes and what me and another were or are working on the expectations were a tad high.

Extreme ocing results on ln2 [ not me ] while wow factor are pointless since rtc bug and no win 7 drivers thus making the results oh that's cool now toss because they don't count.

Expectations of 4.3 to 4.5 on a stock cooler or aio and guides on how to [ me ] are also a tad far fetched goal I was given.

Yes it can do it, can it survive full blown stress tests while staying in optimal thermal ranges? No. My criteria is strict and I wont sign off on something nor supply settings that are a recipe for disaster.

Other than the unrealistic expectations myself and another were given I really have no complaints about the actual product.

It's not much of an overclocker in the sense of shear frequency but it can be tuned quite a bit to gain more cpc performance over stock.

Since gains can be had while leaving cpu at a stock cpu speed I will just focus on settings that enhance the platforms total performance.

[chew*]

Nice quote, thanks for that dude

I got screenshots from support showing where they were measuring vDDR from, however they didnt show which ground point to use. This would indicate they were either using the psu ground or an external ground. When I measured using the memory plane ground I saw +16mv compared to using one of the 24 pin grounds. In my case bios set 1.35v, reported 1.404v idle, sampled 1.399v idle memory plane ground and 1.382/1.384v idle 24pin ground. I have raised this point in my reply to them.

There is no readout for vTT and support have given me nowhere to check it in hardware yet. It is possible that vTT is undervolting and vDDR is overvolting and this could be magnifying the issue. I need to check this next, but will have to ask support for the read point and ground.

Derailing chews thread a bit here, old habits die hard lol. Sorry man

If its anything like ryzen 1 you can also pull vddr and vtt from socket since IMC although I would refrain from taking vddr at socket as an actual voltage. ground planes should be used from the source voltage your measuring so at mem slots and for cpu at socket and i'd measure under dim slots to see what voltage your sticks are seeing as well as measure voltage under socket to see what voltage your cpu is seeing. measuring vrm can be misleading especially in cases where a particular board has excessive vdroop.

Oh and be wary of using profiles at least on asrock. I had to hard clear cmos because loading defaults after loading a manual oc profile were still keeping my manual cpu oc's and really ticking off the system. I flashed a new bios in so hoping its fixed as the asrock was getting increasingly annoying to use with that going on.

[chew*]

shotgun approach abba bios. seems things are improving.

[chew*]

Here is what I have to say about the 5700 cards after using one for a month in apex. What I aim at is what I hit, in my case heads. No leading no predicting needed.

Won't make you a god shot....but if you can aim and keep in mind I still trackball it up....1.9%hs per kill is a nice stat to have.

P.S. that is not sniping, its all face rage.

[bluestang]

Any of you have an idea what a 3900X and either a X470 CH7 or Taichi would be capable of with 4x16GB for 24/7 use? Is 3200C14 even a possibility with b-die on these boards with 4x16GB?

[chew*]

The imc is better imo. Im sure if you enable BSG and maybe settle for 2TCR DR dims should be fine.

[chew*]

Playing with gigabyte master and the 2700x currently. There is some deviation from requested vddr and termination and what you actually get however I compensated and it responded to the compensation exactly as expected. Its not horrible actually its working decent but if you don't compensate I imagine it would be a pain to get stable.

[chew*]

Took a couple hours to figure board out, one thing to note it was picky with my flaky dims best stick needed to be closer to cpu weaker stick furthest from cpu. once that was sorted I could pull this. One thing to note I this chip can't pull 3600 nb as I slacked memory in both taichi and gigabyte and was able to gain no stability so its neither of the boards. This was just on stock cooler and since chip has little to no headroom I did not bother clocking cpu.

[chew*]

Ok so I swapped the 3900x in and tried setting porting. It appears that the chips are so similar that if the settings are stable for the memory and board then just swapping chips the settings will port over and be stable. One thing to note is this board or maybe a change in the new agesa ( will have to go back and test taichi ) does not ride the lightning aka 95c however it comes at a cost as the chip downclocks more. Keep in mind I am on stock cooler but there has been no instance that this chip has been allowed to reach 95c on this board vs the taichi.

P.S. sorry of lack of activity lately but as I originally predicted the boards would undergo a myriad of agesa updates and quite honestly its a waste of time (which I have very little of lately) to keep plucking away and then start from scratch repeatedly.

[chew*]

As I suspected board and memory were not the limitation for pulling 3600 or better with the 2700x. Board repeatedly f9'd me no matter what I tried to exceed 3466 on the 2700x. With the 3900x it booted @ 3800 with very little effort. Ill let this cook overnight to verify stability.

[chew*]

Been putting in work with lifeline in apex with the 5700 vanilla the past couple weeks have run into a few hackers but I think any game has them nowadays.

Gonna start focusing on ranked and get to predator status, we got pushed back to gold and i already pushed to platinum in season 3.

[chew*]

ok 21 hours is gud enuf for me, got a couple baseline stable profiles to work with, now I can fool around a little.

[chew*]

Ok i'm starting to like this board, I think it may clock memory a bit better than taichi, couple things I played with were getting past the IF 3800 wall, not much luck there got around it lost hard drive so no boot device via bclk 100.50.

My 16-16-16 settings booted right up at 4000+ async fine however.

Booting over 3800 IF is a no go, one could only hope lol.

Was not really to fond of the baseline 3800 16-16-16 performance so I dialed in subs @ 3600 passed a couple hours in hci, beat the best 3600 settings run time(206) in dram calculator? I dunno whatever that means I got 202 consistently.

gonna let this cook in prime custom overnight again.

[chew*]

ran this off. xmp vs tuned. nice little gain there considering same speeds. tossed in an untuned 3800 run to. maybe ill go tune that next.

[bluestang]

On that XMP 15-15-15-35 setup...tips to getting that stable? Gear Down enabled or disabled, etc? Isn't there something with all odd (15-15-15) and Ryzen and a setting or 2 that needs changed for it too work?

Reason I'm asking is I'm about to setup a 3900X on a X470 CH7 with 4x8GB GSKILL F4-3600C15D-16GTZ. At least until I get some 2 or 4 x 16GB 3600 16-16-16 RAM. It will be running BOINC projects like World Community Grid 24/7.

[Charles Wirth]

Both of mine are fully populated and currently sitting at:

3900X CH8 4x8GB 3800CL16 Samsung B die 24/7
3700X MEG GODLIKE 4x8GB 3373CL16 Samsung B die 24/7

[chew*]

On that XMP 15-15-15-35 setup...tips to getting that stable? Gear Down enabled or disabled, etc? Isn't there something with all odd (15-15-15) and Ryzen and a setting or 2 that needs changed for it too work?

Reason I'm asking is I'm about to setup a 3900X on a X470 CH7 with 4x8GB GSKILL F4-3600C15D-16GTZ. At least until I get some 2 or 4 x 16GB 3600 16-16-16 RAM. It will be running BOINC projects like World Community Grid 24/7.

gear down is enabled by default. to run spec on 15-15-15-35 3600 one needs to load xmp set IF 3600 set vddr 1.35 set geardown disabled and let her eat. This worked on both taichi and giga since abba bios.

No need to buy other ram these are the 3600 c15 gtz kit. will pull 3600-3800 fine @1.45v with tight timings. no clue which board your using but ill post up the timings for the giga later tonight.

IIRC the 3900x-3950x due to multiple dies and the way it communicates DR dimms perform better all things equal. On x470 it might not hit 3800 all depends on how good the trace signal are from dram to cpu.

one thing of interest to note and I don't personally take much stock in cinebench being the end all solve all defacto benchmark however I saw maybe a 1 point deviation between tuned and xmp @ 3600. I then ran heaven and time spy and corona and the trend continued. gains or loss could be chalked up easily to bench variation. Obviously due to aida there is a difference but in many apps its hard to measure gains. ( etc time put in to tune doesn't bear much fruit ) I found this very odd considering xmp geardown disabled = 1.5T CR performance hit as well.. I'm going to have to dig deeper and try other game benches to see if it has any impact.

I will have a very hard time convincing lazy gamers/avg joes to spend a few days tuning there memory vs running xmp in a matter of seconds for no measurable gains except in aida and geekbench which they could give 2 s less about.

Once I went to 3800 CB gained 40 points so it appears that it may be better to try to set your goals on higher speed vs latency/tight timings when comparing 3600 c14-c15 vs 3800 c16

[chew*]

this went 28 hours so now we can do some comparisons.

[chew*]

I think heat may have been a contributing factor other day. I dropped temps down to 65f in house and the results are much closer now. Really it appears once your in 3600-3800 range bottlenecks are no longer an issue like they were on ryzen 1/TR and gains are minimal @ best. I'm sure I could hand pick a handful of memory biased benches to argue the point that you must have 3800 mem speeds but that would be pretty misleading. A handful of random benches is more realistic. Aida shows decent gains but the rest of the results are minuscule in comparison.

One other point of interest, even on stock cooler on this board and this bios and 65f ambients I was able to actually screenshot AMD hitting peak advertised XFR speeds so I guess they can do advertised...

3600 tuned


3800 tuned


3600 tuned


3800 tuned


3600 tuned


3800 tuned


3600 tuned


3800 tuned


3600 tuned


3800 tuned

[bluestang]

Ok, thanks for the info Charles and chew!

Must force myself to find time next week to get it together.

[chew*]

I went back and retested taichi. If you recall at one point I had a radiator strapped to an ac and had it water cooled and could not do XFR peak all core even on the 3700x. Right now its just 65f ambients and has a noctua NH-d15 on it.

Seems the boards are now where they should have been at launch bios wise.

I feel much less like a beta tester and they just work. Initially it felt akin to R1 launch.

I ran all these off back to back np so its not like I was suiciding. would it be stable here? probably not and I wouldn't run these volts through prime anyway.

I still feel like unless you really get lucky in the silicon lottery you should not expect to hit all core frequencies equivalent to peak XFR for 24/7 stable use but you should be able to run benchmarks all core at peak xfr providing you have sufficient cooling to do so.