chew*

It creates complications but works. To many what ifs. What if it does not like high pci for example. Trust me if it was fast and reliable myself and Zen would be using it. We have put over 200 hours in learning/tuning Ryzen 32m. Still learning new tricks to this day. Win 7 is actually not all that bad...i think people just clung to xp never put time in to get 7 running 32m right. We are also bench partners 2 minds are greater than one. Boards need an ide mode tbh. A rare qualification Asus Arica has it we know it can be added. They refuse to add it even though i requested it. I wanted to try 3d01. Filled another dewar. Try to redeem my crappy 7:58 sub 5 gig time when i get some free time which is rare lately. Also will clock to whatever max of chip is and let one rip now that i worked around my 50x multi limit in k17tk. I should have been 10 secs faster. Annoyed with myself over that.

Thats probably due to using xp 32. 32 7 is also slow. Xp 64 sucks. I made a custom xp for team xs long ago for 32m. It was fast. Many are slow. Removing right stuff and not removing wrong stuff always matters with 32m. Last run cold i was limited to 50x via software. That issue is being solved next run. Hardware tune > software with ryzen. There is still room for improvements. 8:21 with 2666 mem speeds is nasty and a necessity for this chip pretuning cold. Just need to retest phase make sure it cooperates then LN2 again. The current records a disgrace. Can not allow it.

I will run a 4.0 run off on a 1400 for you. Half the cache takes a performance hit but I discussed that quite awhile ago. crappy run ( can go faster ) but gives you a ballpark target.

Here is an example of semi proper scaling. As you can see my ln2 run was "slow" ( sucked monkey balls ) target was and should be 7:48 give or take a second. If I can get a proper run in and hit the target I see no reason why sub 6 could not be done @ 5.4-5.5. Its not perfect math but scaling should be a about 9.5secs shaved per 100 MHz at this crappy memory speed. Better scaling can be had with higher mem speed.

Annoying board and cpu. gen 2 no coldbug? lol. gen 1 no coldbug? lol. got tired of the shenanigans broke mount on cpu, tossed water in the bottom of pot taaadaaa CB gone...technically speaking it acts like cold slow. No clue exhausted a ton of time on it last night. results suck due to trials and tribulations. It can go faster but whatever sub 7 5gig is done.

Forcing promontory gen 2 was the plan...but in the event of a crash or setting that forces clear cmos....stuck on 54. Simple solution. Bios 2.36 with promontory gen 2 loading @ default/bios clear etc etc..2.34 has some annoying bugs...time consuming wise...waste of ln2. Then humidity can kiss mine...i will just freeze everything. Granted it will not clock as high...i only need 5.0 for what i am doing right now. My profile was only @ 2666. Also 2.4 bios freaked out just in general even -30c...2133 was bricking an OS instantly. Bricked SPD on a stick of ram to lol... 2.34 was booting no issues up to -120c. Oh my cpu has a really bad imc btw so might be a factor.

Cold bug cpu wise @ -120 plus 98% humidity has put a damper on my taichi bench session...impossible to go up down temp wise in this humidity...everything will be soaking wet. Think we know what issue is but need a bios to default @ a certain setting or a hardmod to force the setting. Soon as you crash/clear cmos stuck on 54 colder than -120... Shame the effeciency was on par from 4.0 to 4.5 on phase...just needed colder and 5.0 run.

My board has succumbed to the all vendor issue of losing a dimm channel. Swapped various hardware still single channel....maybe i can set a useless memory oc record in single channel now. Otherwise i have a brick.

I have a theory but msi kind of tosses a curve ball. Asrock/Giga copy paste ref trace layout. Asus got creative. Clocks higher/easier..took a hit in PI for it. Normally board to board 1 sec is deviation across vendor. 10 is way out in left field. Weekends coming so is ln2. Time to make AMD look halfway decent @ PI...5.2 6:54 is making AMD look horrible at it.

Its still slow but faster...or was faster than the most recent that i tested...which i admit not tested in recent weeks. I will just show you this. Scaling or more so the lack there of is spot on and exactly what i would expect from speed bump/timing sacrifice. Hynix DR...nothing could be more head/desk than these. Both runs are at what should be prime stable..but that remains to be seen...know more in a day or so.

You on newer bios? Have you tried reverting back?

Yah they are on taichi to. No impact in times usually drives can help attain higher clocks or stabilize rather unstable memory on the platform like say Hynix DR.. ODT however impacts times. More or less the right setting is normal...wrong setting slower. We both obviously have faster times than what is posted. I have not run 32m in weeks...but plan to this week. Been busy helping the 24/7 crowd.

Hate to say it infra...and i have poked ASUS enough to be annoyed already....but going to need a K7 or Taichi....unless ASUS gets there act together. I got one more idea to push them...no clue if it will work.

Would have thought in 5 years...someone would have dug up some effeciency on llano by now...versus one run showing an example of what llano could do...

I am bored....think zen is to...time to shake things up.... Maybe if someone copies my settings they can go fast... I do not share bench settings of course...so you are on your own for that....but my prime stable settings I do share and those are good enough to beat everyone but zen Just for the hell of it i tested 4.2....people still slow in the higher than 4 gig range....probably more in this but something came up so i had to shut rig down.

Here Flanker......I ran one so you could see actual ram speed..... This is the bar that basically K7 and taichi have set.......this is the standard...anything less is slow and not competitive.... You can ask zen......this is extremely fast for 3200.

Really need to be in 17's to 18s range @ only 3200 memory to be a competitive board now. I am knocking out low 18s on taichi with on 12-11-11 3200 no bclk. Gain from the next memory divider is about 3s...if you can hold timings together and not slack them. 3200->8:17 3333->8:14 3466->8:11. Any board doing less is slow. For cold i just tune @ 2666 now since its guaranteed and work my way up once cold.

Its Black and White, it's Taichi which is a form of Kung fu. I dub thee kung fu panda. 3333 strap ( bad imc )

Except...lol i ran out of rearms in my shots..

ok so where did we leave off........high 18's. Taichi wants in so..... quick and dirty still learning board for 32m......seems a touch slow but we will work with it. Old tweaks are the best tweaks......I love my authentic windows pop up tweak.... anyway give me a week to work through this taichi and get a k7 replacement to fill up my bag of sand again with my crappy imc.....

Hold on hold on taichi wants to enter the arena lol...K7 died. Needs rma oops...

Want to make it more challenging...limit it to 5 gig...imo sub 6:59 is super easy...sub 6:59 is a challenge @ 5 gig sort of...i have calculated 6:46 to be realistic @ 5.0 with crappy mem speeds... A 5.4 chip can do 5:59....now that is a challenge... What your going to get is just some sloppy high clocked innefeccient score that someone wallet wins just to cash in. Make it $50 first to 6:59 @ 5gig $150 first person to 5:59 any speed... Then you will see impressively tuned proper runs

Ahh

Veeeeeeesta for wprime multithreaded. Xp for single core.

I would not worry about it flanker. If someone does not run 32m properly soon you know the old saying.... If you want it done right. Do it yourself. I will run it...but i no longer submit results.