Ryzen Owners Zen Garden

[tabascosauz]

Thanks, yeah that was an early run, it will run 286. Ok so I took it off the 1 speed 1 voltage thingy. And I think I may have found a secret button..

tRC, not tRFC not that it should make nearly as much difference as tRFC, but 75 is pretty loose

Nice review. Good to know I'm not the only one who doesn't like the way Asus does LLC on this platform. I'm probably giving it too much and driving my temps higher than they need to be. I'll be trying to nail the vdroop for each level, maybe this weekend.

Think you might have made a little mistake on entering the timings on the Rev.E - Rev.E doesn't scale tRP so you're going to have to run it at 18, not 16. You can definitely do tRCDWR same as tCL at 16 though, just not tRCDRD and tRP because it's not B-die. Chances are you'll probably manage to make it stable at 4000 16-18-18. i'm clearly retarded, goes to show that coffee =! sleep

 

[PaulieG]

Think you might have made a little mistake on entering the timings on the Rev.E - Rev.E doesn't scale tRP so you're going to have to run it at 18, not 16. You can definitely do tRCDWR same as tCL at 16 though, just not tRCDRD and tRP because it's not B-die. Chances are you'll probably manage to make it stable at 4000 16-18-18. i'm clearly retarded, goes to show that coffee =! sleep

 

[Space Lynx]

tRC, not tRFC not that it should make nearly as much difference as tRFC, but 75 is pretty loose



Think you might have made a little mistake on entering the timings on the Rev.E - Rev.E doesn't scale tRP so you're going to have to run it at 18, not 16. You can definitely do tRCDWR same as tCL at 16 though, just not tRCDRD and tRP because it's not B-die. Chances are you'll probably manage to make it stable at 4000 16-18-18. i'm clearly retarded, goes to show that coffee =! sleep

any recommendations for me tightening my b-die ram? 2x16gb 3200 cas 14-14-14 OC'd to 4000 2000 fclk 16-16-16

what values should i try to mess with a bit to tighten it a little?

 

[Mussels]

It just doesn't make sense to me. I pass everything, and I can go hours and hours of daily tasks, streaming, everything. No signs of instability whatsoever. It isn't until I turn the computer off and leave it off for several hours that this occurs.

As far as the manual OC goes, I posted a couple of weeks ago about the big jump in voltage needed for anything over 4.6ghz. Needed just 1.14v for 4.4,ghz 1.16v for 4.5ghz, 1.24v for 4.6ghz and 1.29v for 4.7ghz. My line in the sand on these chips is 1.3v, so I never really messed with 4.8ghz.

wait... bad CMOS battery?

I'm pretty sure the throttle limit is 90c or 95c. Can anyone confirm with a reliable link? I've set the manual throttle limit to 85c anyways. Honestly when I go back and look at screenshots, under the EK AIO my highest temp is 76c on OCCT and 74 on Realbench. The only time I reach 80c was with PBO enabled on a Cinebench R23 run where the vcore jumps up to 1.35v. Never over 74c on any bench run with all core manual overclocking under 4.7ghz.

i've seen 90C on my 5800x, throttle is 90

 

[PaulieG]

wait... bad CMOS battery?

Bad battery? Didn't think of that. Would be odd on a new board, but if the increasing the SOC voltage doesn't help, I'll look into that next.

 

[harm9963]

Haven't bench in a while, for my new ASUS DARK HERO.

 

[thesmokingman]

Bad battery? Didn't think of that. Would be odd on a new board, but if the increasing the SOC voltage doesn't help, I'll look into that next.

Do a full shutdown, wait a minute then boot up.

If it does it again then it might be a memory training issue. It doesn't look like AMD supports memory training disabling like on Intel back in the days, so it might be unavoidable.

 

[freeagent]

I loaded my ram for the night, I though it was ok before, still no errors so I can I can move forward

 

[INSTG8R]

I loaded my ram for the night, I though it was ok before, still no errors so I can I can move forward

View attachment 191537

Congrats on hitting 2000! I am just happy my 3600 runs at 3800 for 1900 as close I am gonna get to the max without better RAM...

 

[PaulieG]

I loaded my ram for the night, I though it was ok before, still no errors so I can I can move forward

View attachment 191537

Humor me. Run OCCT for an hour both on cpu and memory test and see if you get any WHEA errors. I tested over 4000/2000 just like you did here running for hours successfully but failed OCCT with around 100 WHEA errors after 45 minutes on both tests.

 

[freeagent]

Congrats on hitting 2000! I am just happy my 3600 runs at 3800 for 1900 as close I am gonna get to the max without better RAM...

Thanks man! I still have my 3600XT and she was good at 1900/1900 too, and with 4 sticks! I havent been able to do that with this one yet.. but like I said before.. once I got to 2000 I gave up on the idea of four sticks, because I had already made of the gain that you would get from dual rank

Humor me. Run OCCT for an hour both on cpu and memory test and see if you get any WHEA errors. I tested over 4000 just like you did here running for hours successfully but failed OCCT with around 100 WHEA errors after 45 minutes on both tests.

I ran Linpack Xtreme, seems to get warmer then OCCT and is free..

 

[PaulieG]

I ran Linpack Xtreme, seems to get warmer then OCCT and is free..

Hotter, yes but not nearly as good at picking up WHEA errors. Trust me.

 

[freeagent]

Hotter, yes but not nearly as good at picking up WHEA errors. Trust me.

Spotted errors!

Dammit.

So I have it at 2000 and I will let it go to see what happens, so far so good but I am only 2 minutes into the run. I also switched to a fixed frequency for testing.

 

[INSTG8R]

Spotted errors!

Dammit.

Doh! Thankfully I’ve managed to avoid them BUT I will admit to not doing any real thorough stress tests like that to actually break things...I may have to actually run OCCT and join the club. I usually use AIDA64 which apparently pretty heavy and any attempts at messing with timings or speeds above 1900 have just instantly errored out.

 

[PaulieG]

Spotted errors!

Dammit.

So I have it at 2000 and I will let it go to see what happens, so far so good but I am only 2 minutes into the run. I also switched to a fixed frequency for testing.

Really hated to rain on your parade. No matter what I did, I couldn't get any of my kits stable on higher than 4000/2000, even up to 1.45v on either of my B550 boards with OCCT. I spent hours and hours trying. I think it is the best program available for finding WHEA errors. Aida ran fine, Linpack and Realbench both passed for an hour. Not OCCT. Once I dropped to 4000/2000 or below, I passed no problem.

 

[freeagent]

No its ok man. If you didn't do it I would have found out at some point. Really weird. Everything I ran said it was ok, and then this new program comes in to burst my bubble.

I am now starting all over. Maybe I had something too tight somewhere.. ugh. I briefly considered putting my other pair in.. but cant because there is no offset to my cooler, so I lose a dimm.

Might have to put LGMRT back in

I wont give up yet. If this is a taste of what AM5 will be like sign me up. This beta is pretty sweet this time.

This program is pretty good. I had to start all over again, right from scratch. My 4600 at 1.225 was no good with the small data set and extreme setting. Bummer. I have the auto oc going for 4850, at the program settings I have listed cpu is running at 4550. So everything I thought I knew is wrong. Thanks Paul!

Looks like 1966 it is..



Bummer do0d.

 

[thesmokingman]

I wont give up yet. If this is a taste of what AM5 will be like sign me up. This beta is pretty sweet this time.

DDR5... yea no this isn't anything like that.

 

[freeagent]

DDR5... yea no this isn't anything like that.

I wasn't talking about the memory so much, the whole thing is pretty great. I like this little system.

 

[PaulieG]

No its ok man. If you didn't do it I would have found out at some point. Really weird. Everything I ran said it was ok, and then this new program comes in to burst my bubble.

I am now starting all over. Maybe I had something too tight somewhere.. ugh. I briefly considered putting my other pair in.. but cant because there is no offset to my cooler, so I lose a dimm.

Might have to put LGMRT back in

I wont give up yet. If this is a taste of what AM5 will be like sign me up. This beta is pretty sweet this time.

This program is pretty good. I had to start all over again, right from scratch. My 4600 at 1.225 was no good with the small data set and extreme setting. Bummer. I have the auto oc going for 4850, at the program settings I have listed cpu is running at 4550. So everything I thought I knew is wrong. Thanks Paul!

Looks like 1966 it is..

View attachment 191593

Bummer do0d.

OCCT has always been my "go to" stability test ever since it first came out a decade ago. I just think it's more comprehensive than anything else and it's programmed to pick up the subtle errors that prime can miss. The only thing I don't like now is how they do the support pop up, and you can no longer buy the premium version outright like you could before. now it's a yearly subscription.

Realbench is pretty good too, just not quite as thorough. I prefer either of them to prime because they effectively test for stability without excess heat.

 

[dont whant to set it"']

This chip is kind of growing on me, on day one I arbitrarily set 45.75X multi for all-core oc and it turns out its the maximum I can get out of it under a decent ~25£ tower air cooler.

Sure , with a super premium air cooler or 240mm(or above) AIO I see it possible to cool it enough for 46X multiplier and maybe a little bit more , but not much more without a infinite heatsink.

Yesterday it passed a 30 minutes loop at 38x multiplier at 975mV VID reaching 52 degrees C.

Now: 38x multi , 950mV VID , LLC 1 ,CPU vrm switching frequency 200KHz 250KHz.

 

[PaulieG]

This chip is kind of growing on me, on day one I arbitrarily set 45.75X multi for all-core oc and it turns out its the maximum I can get out of it under a decent ~25£ tower air cooler.

Sure , with a super premium air cooler or 240mm(or above) AIO I see it possible to cool it enough for 46X multiplier and maybe a little bit more , but not much more without a infinite heatsink.

Yesterday it passed a 30 minutes loop at 38x multiplier at 975mV VID reaching 52 degrees C.

Now: 38x multi , 950mV VID , LLC 1 ,CPU vrm switching frequency 200KHz 250KHz.

Try running an hour or 2 of OCCT or Realbench to see if she is really stable. Neither will overheat the chip. They will just find errors.

 

[freeagent]

That program is funny. Not haha funny but.. funny. Every stress test program I have run said my settings were good. Windows had no error reports tucked away, games ran great, benches were record breaking for me lol. And then this app that constantly lets you know it wants to be paid says my oc is no good. Just kinda funny.. I dialed it back for now.. but this makes me want to investigate further.

 

[PaulieG]

That program is funny. Not haha funny but.. funny. Every stress test program I have run said my settings were good. Windows had no error reports tucked away, games ran great, benches were record breaking for me lol. And then this app that constantly lets you know it wants to be paid says my oc is no good. Just kinda funny.. I dialed it back for now.. but this makes me want to investigate further.

I understand what you mean. If I hadn't used OCCT for years with the same type of characteristics BEFORE the support ad pop ups, I'd be suspicious too. I just think it's better at finding almost imperceptible errors than other programs. I mean, it gives me stable results just under what I thought was stable. It's not telling me outrageous info, like anything above stock is unstable. Try running Realbench which doesn't ask for a cent. It finds the same types of errors as OCCT about 80% of the time.

For what it's worth, I'm sure your OC was "everyday stable". Probably more stable than most people who claim high overclock stability. That type of stability isn't enough for me and my OCD. If it fails anything, that little voice in my head won't let me leave it alone.

 

[freeagent]

I know what you mean buddy, we have the same internal monologue lol. I know the errors are real because they disappear when I slow things down.. I was hoping I can trace it back to some subs but I have my doubts because I made things sloppy inside and it did help marginally but not enough to call it a success. I also tried to tweak llc a little to no avail. I thing If I want to go further I have to learn how Ryzen really works and get into the meat and potaters of it all and work some sploits.

 

[Mussels]

This is exactly why many of us dont believe overclocks we see scattered around the web, because people go HURR ITS STABLE after 15 minutes of running MS paint (or whatever their favourite quick run is) and then the damn thing crashes and causes problems because its not truly stable. PCMR facebook group is non stop technical issues posts of people who think that because it turned on, its stable.

OCCT may just be the best way to really stress out zen chips, and should probably become our official tester