Post your Ryzen memory speeds!

[Formula.350]

ahah O_O Ho-ly shit.....

OK so I gave up on 3533 for now. It'd cause everything to flake out when trying to boot Windows. Opted to try out 3466 instead and seeing where I can take it for tighter timings.

Was trudged through some of the Overclockers.net Ryzen DRAM thread and came across freestaler's post...
I opted to keep my known working primary timings of 16-18-18-38 just for now. I noticed his tRC is super low and decided I'll keep mine at 56... after plugging a couple more in I realized his sub-timings in general are insanely low, at which point I look over at what RAM he has and his is a 4133 CL19 kit. Well bluh... for giggles I try them anyways, finishing typing them in. All in all I kept the majority of them, only compromising on a couple others. I forgot to set the same DRAM and SOC voltages though, which I had set to 1.40V and 1.1V.

MUCH to my surprise it actually POSTed! Alas, it would basically insta-reboot when it tried to load Windows. So I now set my DRAM to 1.42V (Titanium doesn't accept 1.415V) with the appropriate VREF of 0.710, and the SoC to 1.0875V.

It POSTed.
It starts loading Windows.
It loads Windows.
It lets me load RyzenTimingsChecker... AND take a Snip of the screen!
It loads AIDA64.... o_0
It completes the entire Cache and Memory benchmark O_O

With my CPU ON AUTO, it pulls in a score of:
Read: 54771 MB/s
Write: 53490 MB/s
Copy: 50409 MB/s
Ltncy: 68.5 ns . . . . . . . EDIT: I've re-ran it a few times now and the Copy is lower, but I've had a 67.5ns and just now a 63.3ns turnout!

Like, some of these are insanely lower than even my 3200 subs were! I'm not getting my hopes up that this is going to be stable, but goddamn, if it somehow is, that'll be amazing...

EDIT: Well smack my ass and call me Shirley! 16-16-16-30 50 works. Yet to try his 16-16-16-21 37, or specifically, test to see if this is stable. Still not holding my breath on that, though lol It isn't No surprise.

I'm hating BIOS 1.F0's POST Code LED changes, and it not reading temps anymore/properly (hopefully just a bug), as well as the way the BIOS is handling temps now as well (my old fan settings don't work well now, and my tweak to it ramps up/down too much), but... I can't deny that it does a lot better with RAM! I'm really going to miss being able to look at the LED for my Core Temp if that doesn't get fixed

 

[Ebernanut]

I'm currently running 3466 16-15-15-35(GDM=Enabled) on 2700x with a 2x8GB 3600cl15 kit. This is just what I settled on for now until I upgrade the CPU cooler so I can get everything dialed in together, it will give me some time to learn more about memory on Ryzen too.

PUBLIC SERVICE ANNOUNCEMENT:
While this may not be possible for every motherboard, but for anyone who isn't aware of this already (to spare you having to read through this entire thread), if you set GearDownMode to DISABLED you should then be able to run Odd numbered tCL (CAS) Latencies. This way you will not have to go from, say, 14-14-14-34 at 3200, to 16-16-16-36 just to get 3466 stable. With GearDownMode set to Disabled, you'll be able to use 15-15-15-35 and it won't be forced to 16-15-15-35. Just be sure to set the Command Rate (tCR) to 1T as on Auto with GDM Disabled it may switch to 2T and as such lower your performance (specifically, it'll increase latency).

I feel it's worth noting that enabling geardownmode can sometimes help quite a bit with memory stability and I've seen conflicting info on the performance penalty. How much it helps seems to depend on how good the IMC on a CPU is more than anything and I imagine the performance penalty depends on the application.

So far I've had much better luck with it enabled and the performance penalty has been almost nonexistent in the benchmarks I've run. The settings I'm running right now wouldn't even boot with it disabled but passed an initial test of over 400% coverage in HCI memtest with it enabled, I was able to boot with I believe 3333 or maybe 1400 with 15-15-15-35 timings but the cinebench score was noticeably worse and one of the HCI instances threw an error at a bit under 200% coverage.

I'm still learning so maybe I'll figure out how to run things fast and tight with it disabled but for right now it seems to be helping a lot more than it's hurting. I'd be open to any tips more tips you have though since you obviously have a head start on the learning curve.

 

[os2wiz]

Ryzen 2700X at 4.2 GHZ on all cores manual overclock.
MSI X470 Gaming M7 AC
Bios 1.1 Agesa code 1.02a
G.Skill Trident Z DDR4-4266 CL19-19-19-39 1.40 volts 16GB kit F-4-4266C19D-16GTZKW
Memory sticks are in A2/B2 slots .


I was able to post as high as 3866MHZ but wholly unstable. I had to do a retrograde decrease in speed until I reached ultimate stability at 3466mhz at timings of 15-15-15-15-32-64 at 1.39 volts.
I passed YCrunch at these settings could not pass YCRUNCH at higher speeds.

 

[gigaxtreme1]

You have gear down disabled?

 

[os2wiz]

ahah O_O Ho-ly shit.....

OK so I gave up on 3533 for now. It'd cause everything to flake out when trying to boot Windows. Opted to try out 3466 instead and seeing where I can take it for tighter timings.

Was trudged through some of the Overclockers.net Ryzen DRAM thread and came across freestaler's post...
I opted to keep my known working primary timings of 16-18-18-38 just for now. I noticed his tRC is super low and decided I'll keep mine at 56... after plugging a couple more in I realized his sub-timings in general are insanely low, at which point I look over at what RAM he has and his is a 4133 CL19 kit. Well bluh... for giggles I try them anyways, finishing typing them in. All in all I kept the majority of them, only compromising on a couple others. I forgot to set the same DRAM and SOC voltages though, which I had set to 1.40V and 1.1V.

MUCH to my surprise it actually POSTed! Alas, it would basically insta-reboot when it tried to load Windows. So I now set my DRAM to 1.42V (Titanium doesn't accept 1.415V) with the appropriate VREF of 0.710, and the SoC to 1.0875V.

It POSTed.
It starts loading Windows.
It loads Windows.
It lets me load RyzenTimingsChecker... AND take a Snip of the screen!
It loads AIDA64.... o_0
It completes the entire Cache and Memory benchmark O_O

With my CPU ON AUTO, it pulls in a score of:
Read: 54771 MB/s
Write: 53490 MB/s
Copy: 50409 MB/s
Ltncy: 68.5 ns . . . . . . . EDIT: I've re-ran it a few times now and the Copy is lower, but I've had a 67.5ns and just now a 63.3ns turnout!

Like, some of these are insanely lower than even my 3200 subs were! I'm not getting my hopes up that this is going to be stable, but goddamn, if it somehow is, that'll be amazing...

EDIT: Well smack my ass and call me Shirley! 16-16-16-30 50 works. Yet to try his 16-16-16-21 37, or specifically, test to see if this is stable. Still not holding my breath on that, though lol It isn't No surprise.

I'm hating BIOS 1.F0's POST Code LED changes, and it not reading temps anymore/properly (hopefully just a bug), as well as the way the BIOS is handling temps now as well (my old fan settings don't work well now, and my tweak to it ramps up/down too much), but... I can't deny that it does a lot better with RAM! I'm really going to miss being able to look at the LED for my Core Temp if that doesn't get fixed

I have the MSI X470 Gaming M7 AC. Could you please email your screen shots of your memory timings and voltages and other appropriate settinhs like ProcODT etc. I would like to optimize my timings better. Thank you.

 

[Formula.350]

I'm currently running 3466 16-15-15-35(GDM=Enabled) on 2700x with a 2x8GB 3600cl15 kit. This is just what I settled on for now until I upgrade the CPU cooler so I can get everything dialed in together, it will give me some time to learn more about memory on Ryzen too.



I feel it's worth noting that enabling geardownmode can sometimes help quite a bit with memory stability and I've seen conflicting info on the performance penalty. How much it helps seems to depend on how good the IMC on a CPU is more than anything and I imagine the performance penalty depends on the application.

So far I've had much better luck with it enabled and the performance penalty has been almost nonexistent in the benchmarks I've run. The settings I'm running right now wouldn't even boot with it disabled but passed an initial test of over 400% coverage in HCI memtest with it enabled, I was able to boot with I believe 3333 or maybe 1400 with 15-15-15-35 timings but the cinebench score was noticeably worse and one of the HCI instances threw an error at a bit under 200% coverage.

I'm still learning so maybe I'll figure out how to run things fast and tight with it disabled but for right now it seems to be helping a lot more than it's hurting. I'd be open to any tips more tips you have though since you obviously have a head start on the learning curve.

It probably can help with stability, yea. I haven't had any impact one way or another that I could tell yet; however, I also haven't tried it out yet with this latest BIOS, which give I've had some struggle with 3533 and getting 3466 (albeit with these ultra-tight sub timings), it might be worth me enabling to see if it'll help in my case

I'm still learning as well! I wouldn't position myself any higher than "Informed Novice" lol Even in the DDR3 era, the board I had didn't offer but a small fraction of what's available here, and what it DID offer with drive strengths, I never understood. But I'm all for sharing info, and as you can see I post any tiny tidbit I come across, as well as the tiniest shred of info from my own experiences

I have the MSI X470 Gaming M7 AC. Could you please email your screen shots of your memory timings and voltages and other appropriate settinhs like ProcODT etc. I would like to optimize my timings better. Thank you.

More than happy to, though sadly, mine are not going to be of much help I don't think, as I left all those advanced settings on Auto since that's where my limitation of knowledge comes in
(For everyone else, since I mentioned this to him in his Conversation message, that this is from MSI Titanium BIOS v1.74. On top of that, it's my modded BIOS with unhidden settings, so not all of these may be available for you. Primarily it's BClk, the (BIOS) & (Training) DRAM Voltage [which I don't think actually work, but I dunno], and the tRFC1, tRFC2, tRFC4 sub timings. BankGroupSwapAlt was removed in my 1.F0 for my first gen Ryzen, though I suspect Ryzen+ will still have access. Also the two red-lined options in DigitALL on the Titanium were removed entirely in like v1.9, and to what impact I don't know.)

BankGroupSwapAlt is I believe Disabled by default there. Sorry for the older RyzenTimingsChecker image, but besides BGSAlt, all of the other actually HELPFUL settings cannot be read by the newer RTC anyways due to BIOS v1.74 having too old of an AGESA (1.0.0.4a).

For what it's worth, from BIOS v1.F0, I suspect they would result in the same for 3200 as they did for 3466, since I run Auto on them anyways. So this is what Auto at 3466 produced:
AddCmdSetup . . . 0/0
CsOdtSetup . . . . .0/0
CkeSetup . . . . . . .0/0
RttNom . . . . . . . . Disabled
RttWr . . . . . . . . . .Disabled
RttPark . . . . . . . . .48 ohm
CLKDrvStr . . . ... . 24 ohm -------------v
AddrCmdDrvStr . .. 30 ohm --------------> (CAD Bus Settings: One of Stilt's recommendations was to try higher Ohms for speeds above 3200MHz; try 40-40-40-40?)
CsOdtDrvStr . ... . 24 ohm -------------^
CKEDrvStr . . . . . . 24 ohm ------------^
(I'm an idiot and should've justused the [ code ] function instead lol)

 

[Formula.350]

Well hmpf.

After spending the last 9hrs (almost actively) trying to get 3466 at those tight timings stable from a few posts back, my only progress has been backwards progress. I'm now not even able to get into Windows heh

I know it sounds absurd to spend so long on that, but man... the speeeeeeeeed! Was too tempting lol

I think to wind down the night I'm going to toss in his timings except instead of 16's on primary, I'll use my 14's, but leave the rest of the Subs, and then drop Speed back to 3200. I'm hoping THAT will be stable at the very least. So fingers crossed...

EDIT: *whistles* Bummer lol I was hoping maybe SOME of that speed and latency was thanks to the subtimings but noooope Granted, I oopsed and the tCWL is still 16... huh I guess I did set it? Or the BIOS somehow did, as it is showing up in TimingsChecker as being 14. Anyways, at 3200 14-14-14-26 37, it's all gone away lol
Read: 50530, Write: 49555, Copy: 47132, Latency: 70.2ns.
(No idea if this is stable yet, too late to bother with serious testing)

I'm also curiously noting that the L1, L2, L3 latency is higher when I'm manually clocked at 3.7GHz all cores, vs default. Even at 3466, when I changed the multi in Windows, they'd rise. Was usually 1.0, 4.4 and 11.6ns respectively when default (thus, CPB functioning). At 3.7GHz it's 1.1, 4.6 and 12.1ns. Only thing I can attribute it to is the 3.9 boost being to thank.

Tomorrow will be 3466 straight 16s, with auto subs, to see if even that is stable. If not, then I shall pout and try 3333 at 15's with my stable 3200 subs. If THAT isn't stable, then I'm just going to flat out cry and throw a tantrum!! lol Cuz it's no fair that 3600 will post, 3533 will make it to Windows Load, 3466 able to pass benches, and yet I'd still be stuck with 3200!


EDIT2: Came across this which may help quite a few people out in getting different DRAM speeds working and/or stable for you. It details CLDO_VDDP a bit and explains "Memory Holes" a bit. Though... given he is not a native English speaker means it's a bit tougher to follow. Like for me, while I'm still not certain what a Memory Hole is on the technical side, I took it as being a "hole" where speeds did NOT work. However, how he explains it sounds like it's the exact opposite and the "hole" is where speeds DO work... *shrug* Either way, it still is a worthy ready and along with the Google Doc, may assist some of you in stabilizing your RAM at certain speeds, or even getting them to work when they previously didn't. (BTW, this is by the same guy that makes the Ryzen DRAM Calculator)
http://www.overclock.net/forum/26457559-post292.html

 

[Below Ambient]

What are the original timings on your kit? And when did you buy it?
Also, would you be willing to post a RyzenTimingsChecker screenshot and share what you're running for all the voltages (if you've manually set any)? Pretty pretty please? lol

I've been fighting with getting my system running at 3533,

im running stock timings i just changed the TrFC, TrFc2, TrFc4 from whatever it was and im using 1.5v... everything else is default... no LLC calibration changes...

 

[Below Ambient]

heres aida

 

[Below Ambient]

heres 3607 1.7v, crashes in games after an hr

 

[Formula.350]

im running stock timings i just changed the TrFC, TrFc2, TrFc4 from whatever it was and im using 1.5v... everything else is default... no LLC calibration changes...

1.5V and 1.7V on your RAM? You're going to cook those poor things dude!

And considering you increase the DRAM speed AND lowered the timings, I'm surprised it kept stability in game for even that long! lol

But besides the DRAM voltage, everything else is set to Auto, like the NB/SoC, DRAM VTT (or VREF, depending on your motherboard), CLDO_VDDP, etc?

 

[MMitch]

WOW I was going to push my results (assembled, booted, set to default 3200MHz and forgot about it) but it seems .... Zen has come a long way Glad to see that and I will not enter into details...
Really wow, nice work guys and thanks for sharing, I'll read the thread ^^

 

[Formula.350]

WOW I was going to push my results (assembled, booted, set to default 3200MHz and forgot about it) but it seems .... Zen has come a long way Glad to see that and I will not enter into details...
Really wow, nice work guys and thanks for sharing, I'll read the thread ^^

The thread was basically dead lol I necroed it off of like page 7 of the forums. So if anything, I'd start at page 5 of this thread for more recent and relevant posts.

I'm like you. I'm one of what seems like a very very very small group of people who were fortunate enough to have their Ryzen running with 3200 memory at launch (in fairness, it took me two weeks past launch just to get my board since Amazon screwed the pooch on pre-orders.)

 

[Below Ambient]

1.5V and 1.7V on your RAM? You're going to cook those poor things dude!

And considering you increase the DRAM speed AND lowered the timings, I'm surprised it kept stability in game for even that long! lol

But besides the DRAM voltage, everything else is set to Auto, like the NB/SoC, DRAM VTT (or VREF, depending on your motherboard), CLDO_VDDP, etc?

The Asus overclock seems to maybe overdue things? When I use it it doesnt want to do anything but put my CPU at 4180 1.55v which makes me nervous.. it also messes with my LLC which I havent found necessary... seems to run more stable without certain setting messed with that you would think need ot be adjusted...maybe if i tried harder lol... i set my SOC manually to 1.18, vcore manual as well... everything else is default or auto... running a strix F after I blew the K3 up running 1.7+

 

[Formula.350]

The Asus overclock seems to maybe overdue things? When I use it it doesnt want to do anything but put my CPU at 4180 1.55v which makes me nervous.. it also messes with my LLC which I havent found necessary... seems to run more stable without certain setting messed with that you would think need ot be adjusted...maybe if i tried harder lol... i set my SOC manually to 1.18, vcore manual as well... everything else is default or auto... running a strix F after I blew the K3 up running 1.7+

Yea 1.55V on the core is way excessive for that speed. Though as a bit of reassurance, according The Stilt, the SMU on Ryzen will not allow it to draw above its safe voltage. That being said, there is a caveat: it cannot sense if you're using "Voltage Offset", which I know for sure ASUS has, pretty sure ASRock as well. Because using that executes at the VRM, and so the CPU only can detect what the initial "VID" is set at in the BIOS.

You also should see about using less SOC. Most people seem to get away with 1.05V to 1.15V. I think general consensus is keep it below 1.15V for longevity, and 1.2V for max. Obviously if you're running heavy duty sub-ambient cooing (cascade or a Peltier setup), you could probably get away with more and be safe, but even at low temps, too high of a voltage can still degrade electrical pathways.

Anyways, while you're indeed running way higher DRAM speed, something to consider is my 1700X, running my DRAM at 3200 I only need 0.975V on SOC. One of the guys on the Overclockers.net thread, albeit with a 2700X and a Crosshair VII Hero, was using a similarly low voltage but he was managing RAM speeds of 3600.

More isn't always necessary on SoC and can hinder stability as well, as it'll produce considerably more heat. Don't get me wrong though, if you've spent a bunch of time dialing that voltage in, then I've basically wasted my breath (at least, in terms of helping you, others can still benefit heh)
Either way, kudos on the RAM speed

 

[Below Ambient]

Yeah took me months to find my soc...bought this 1600x two days after release it loves voltage...had it way past 1.5v a few times, killed a board with it...xfr it likes 1.55, clocks to 4180... i just set it to 1.35v and run 4ghz manually...just sucks it hits a wall at 3600 if I use tight timings

 

[drescherjm]

Bookmarked to post my ECC results. That is if I can get a 2700 tomorrow @ MC and some unbuffered ECC next week.

 

[Formula.350]

Bookmarked to post my ECC results. That is if I can get a 2700 tomorrow @ MC and some unbuffered ECC next week.

Your findings are very relevant to Peat Moss' interests! Though, he'll be more interested in the modules used and the motherboard model, since he'd be using a Ryzen 5 1600X.

 

[Formula.350]

So instead of working backwards, trying to get the highest overclock possible to be stable, I've reverted back to the sane way of going about it: baby steps. IOW 3200 is so next... well technically next would be 3266 but we're going with 3333.

Initially I just go with AIDA64's Stress Test, and tick "Cache" and then "Memory", but when starting out I'll just use one or the other. Cache seems to still somehow be tied to Memory stability, and I"m not sure if it's because of the Infinity Fabric, or what. Either way, that's how it's been in my experience with Ryzen.

Now, curiously, I gave that a go and after about 30 seconds it failed. For shits and giggles I gave HCI MemTest a go and unlike usually when RAM was unstable where it'd throw error after error... there was none. So I loaded up 8 more instances each with 1700MB and much to my surprise, after 90% Coverage there are no err.... Welp, here I had to go and jinx myself.

Interestingly it was the last instance that I clicked Start on. And another error, same instance, at 95%.

So I guess my "Initial" method of verifying stability IS valid! And also way faster at establishing results! lol


EDIT: Another day, another round of attempting to get thing stable and... Plausible Progress!!
I change the "RttNom" from Auto (Disabled) to 34 Ohm based on a guy's 3466 stable configuration, considering we have 99% the same timings (again thouggh, I'm currently at 3333).

The AIDA Memory Stress ran for 8 minutes before failing, versus the usual 30 seconds which I think is a considerable improvement. That 30 seconds has been roughly the normal duration, so I'm hoping that this isn't some sort of fluke. While the Cache Stress only ran for 4 minutes before failing, that again is a very large improvement and could indicate I need to look elsewhere.

OR it means that the components have reached a temperature that has caused its resistance to change and so maybe I need to raise RttNom to 40 Ohm... I think that's what I'll try first since I appear to be on the right path.
----------
And the results of RttNom @ 40 Ohm is... well, at the risk of jinxing it by typi... wow, really???? This is getting to be too much of a coincidence, honestly. I am 3 for 3 of beginning to type my findings and it becoming unstable at that exact moment (I'm on a laptop BTW, so this isn't being typed on that system).

Anyways, the Memory Stress results at 40 Ohm: 13.5 minutes!
Impact on bandwidth or latency.... None and Better???
I don't really feel confident claiming this is the cause of the improvement, but I haven't been changing anything else.
When I made some voltage bumps but nothing else it was 68.1ns (RttNom Disabled).
With it at 34 Ohm it was 66.xns.
With it at 40Ohm it's now 64.8ns
(CPU is manually clocked at 3.7GHz, so it's not because of Turbo)

Anyways, time to try RttNom of 48 Ohm!
-----------------
RttNom @ 48 Ohm *drumroll*... 10 seconds Womp-womp! Well you know what they say, there's no such thing as a failure, only the discovery of one way how NOT to do something! So while this was not the result I wanted, it's a helpful result nonetheless, because now I know I need to set it to 40 Ohm. I'm not sure if 60 Ohm (I think that's the next choice) is going to help given how fast this failed.

Of course, I wanted to follow up on the curiosity of Latency and.... well now it's 63.7ns
I mean sure, I'm literally only running the test one time, so an accurate scientific test this is not, but the fact it's consistently going down as I increased the resistance is definitely interesting! Worth noting, is the stand-alone Latency test in the "Benchmark" category tells another tale, and is producing 67.8ns, same as at 40 Ohm

That all being said, time to move on to something else. Maybe RttWr which is also Disabled, but not sure.
In full disclosure though, here's where I'm at currently with partial stability:
CPU: 3.7GHz
DDR: 3333MHz
CPU Voltage: 1.2375V (LLC Mode 1)
CPU NB/SoC: 0.9875V (LLC Mode 3)
DDR Voltage: 1.37V (Phase Freq 600Hz)
Timings and other jazz: (red dots = Auto configured)



EDIT 2: I went with dude's CsOdtSetup and CkeSetup adjustments and was re-testing. 12mins in all was well, then Windows decided nah and threw up a "MEMORY MANAGEMENT" BSOD. I believe when you are hit with that error it means you need a bit more CPU NB (SoC on some boards) Voltage, so I have it bumped up to 1V now.

However, when I open Timings Checker, I see that his "1/31" is unfortunately NOT just him having entered a value of "1" into the BIOS. When I did that, mine reads "0/1" :\

Seems I'll have to do some digging, and/or trial and error to determine how to get a similar value....
It may be relevant that his AddrCmdSetup box is completely blank... so perhaps that's where the "1" comes from, and I need to set the others to "31" in order to attain "1/31". Nope, definitely NOT. It F9's and refuses to post with it all set to that! Even with those two set to 31 it throws an F9 code.
TO THE BAT-CAVE BIOS-MENU! *spinning MSI logo*

 

[Nightfire]

Here is a 2400g with 3200 c14 FlareX:

There is no way for me to go above 3333 mhz even if I crank the voltages and drop the timings. Overall, plenty happy.

 

[Below Ambient]

Here is a 2400g with 3200 c14 FlareX:
View attachment 75091

There is no way for me to go above 3333 mhz even if I crank the voltages and drop the timings. Overall, plenty happy.

Some of your timings are scewed is why... Trfc needs to be loose like 500+ yours is at 24?? I would say start over and hit the speed you want then tighten... Just saying

 

[Nightfire]

Some of your timings are screwed is why... Trfc needs to be loose like 500+ yours is at 24?? I would say start over and hit the speed you want then tighten... Just saying

My trfc is 256 which makes a big difference in latency. 65 ns is pretty awesome. My bandwidth is limited by the memory controller. Trust me, there is absolutely NOTHING I can do to increase the speed or overall bandwidth.

 

[Formula.350]

Some of your timings are scewed is why... Trfc needs to be loose like 500+ yours is at 24?? I would say start over and hit the speed you want then tighten... Just saying

Yea, 500s is a "Safe" amount, but definitely far from a Fast setting.

My trfc is 256 which makes a big difference in latency. 65 ns is pretty awesome. My bandwidth is limited by the memory controller. Trust me, there is absolutely NOTHING I can do to increase the speed or overall bandwidth.

Are you able to lower your tWR at all? Granted, I've not seen many results from the APUs, but from all the Ryzens, everyone is able to run that around at least 1/2 of where you're at, and it'll help bandwidth a little bit.
tFAW is also rather high, but it's been a long time since I adjusted that to remember if there was a bandwidth impact.

Also, depending on that things purpose, if you're gaming you'll want GearDownMode Disabled (if you're stable with it off) and BankGroupSwap Enabled. That will increase FPS; however, your AIDA bandwidth will go down. Even if you can't disable GDM, definitely try out enabling BGS.
If you're doing office related or rendering work then having them set how you do is better.

=-=-=-=-=-=-=-=

As an update on my own situation, I'm still fighting with 3333. GDM on or off hasn't seemed to make an impact on stability. Granted, I'm still running at my tightened 3200 timings, so I may simply need to accept it doesn't want to run at those timings and start loosening (it IS a 3200 15-15-15-35 kit after all, and I'm asking it to do 3333 @ 14-14-14-36). Though, I have a feeling the Air Conditioner may be hurting my Stability Testing... It was running fine for 13 minutes and when the HVAC blower kicked on it failed. I do live at the end of the power lines and when the AC and HVAC kick on there IS a power dip, so I really can't rule it out.

Though it was interesting that the fail happened exactly at 13 minutes Also kind of curious that it seems to be the rough max length of time it will run before crapping out. I can't help but think something else may be at play if not the AC. This is where having more knowledge with DRAM would come in handy :\

Alright AC just turned off so lets restart and see what happens... Aaand that was anti-climatic. Ended after ~50 seconds

Ya know. I'm not sure how many people try this and if it even plausibly helps, but I'm going to swap my DIMMs around, placing them in the other's slot. *shrug* EDIT: Well... that may or may not be telling, but it BSODed shortly after starting the stress test, which is kind of uncommon. We'll give it another go, see what happens....

 

[Nightfire]

Yea, 500s is a "Safe" amount, but definitely far from a Fast setting.

Are you able to lower your tWR at all? Granted, I've not seen many results from the APUs, but from all the Ryzens, everyone is able to run that around at least 1/2 of where you're at, and it'll help bandwidth a little bit.
tFAW is also rather high, but it's been a long time since I adjusted that to remember if there was a bandwidth impact.

Also, depending on that things purpose, if you're gaming you'll want GearDownMode Disabled (if you're stable with it off) and BankGroupSwap Enabled. That will increase FPS; however, your AIDA bandwidth will go down. Even if you can't disable GDM, definitely try out enabling BGS.
If you're doing office related or rendering work then having them set how you do is better.

=-=-=-=-=-=-=-=

I lowered tWR and tFAW, but my bandwidth stayed just under 50k and my latency stayed at 65 ns. Some here may be confused about the lower cache bandwidth, but remember that RR has less cache.

The only way for me to get better performance is to further lower the primary timings or lower trfc. At 1.4v, 256 is as low as trfc will go. I might try 1.45v...

 

[Formula.350]

I lowered tWR and tFAW, but my bandwidth stayed just under 50k and my latency stayed at 65 ns. Some here may be confused about the lower cache bandwidth, but remember that RR has less cache.

The only way for me to get better performance is to further lower the primary timings or lower trfc. At 1.4v, 256 is as low as trfc will go. I might try 1.45v...

As long as it's stable, tighten away! heh

I've also noticed people able to get away with lower tRAS, despite what the "rule of thumb" would have it be set at. The majority seem to stick with tRP+tRAS=tRC though, with some providing an additional bit of headroom with usually 2 to 6 aditional ticks.
Here are two examples


Again, non-APU, so I can't say if and/or how big of an impact tighter timings have on stability for those, since there's a GPU involved. Plus there's always the different board, different memory, different capabilities.

I don't have any details on the second's configuration, but on the first he's running his DRAM at 1.4325V

 

[Nightfire]

Hmm I will play some more this weekend but my initial test of lowering trc saw no change to bandwidth or latency.
Bottom line, 50k is about as good as I have seen with any 3333 mhz ram, and 65ns is right there with the best. period.

I will try 1.45v this weekend and see if I can lower trfc or primary timings. Honestly, it is the only way I see dropping below 65ns. If not, I will still be more than happy!

 

[Vader1975]

G.SKILL Ripjaws V Series 64GB (4 x 16GB) 288-Pin DDR4 SDRAM DDR4 3200 (PC4 25600) Intel Z170 Platform Desktop Memory Model F4-3200C16Q-64GVK

• Timing 16-18-18-38 1T
• at 3000mhz
• 1.375V

 

[os2wiz]

You have gear down disabled?

I never disable geardown. Screws up performance by 4 or 5 fps.

 

[m3ta1head]

Managed to break the 60ns barrier.

 

[Nightfire]

Very Impressive! I will have to lower my tras to see if I can match. That looks to be the only setting of yours that is lower than mine. The 66 mhz advantage shouldn't make the 5 ns differece I have. What voltage dis you need?

Awesome work, man!

 

[m3ta1head]

Very Impressive! I will have to lower my tras to see if I can match. That looks to be the only setting of yours that is lower than mine. The 66 mhz advantage shouldn't make the 5 ns differece I have. What voltage dis you need?

Awesome work, man!

I used 1.48vdimm and 1.075vsoc. Can probably come down on vdimm, going to take some further tweaking. Still testing for full stability at the current settings.

 

[Nightfire]

I have tried once more with timings more in line with m3ta1head:

Perhaps I have hit a limit of what Raven Ridge can do with 65ns. This was at 1.47v (1.45 crashed). I will loosen things back up and set closer to 1.4v.

 

[Nightfire]

So it looks like formula.350 ran into the same latency limits with his Ryzen as I did with my Raven Ridge.

m3ta1head has some amazing cache latency numbers as well with his Ryzen+.

m3ta1head, are you able to get trfc any lower? I managed to hit 248. It might be worth an extra nanosecond or two.

 

[Formula.350]

So it looks like formula.350 ran into the same latency limits with his Ryzen as I did with my Raven Ridge.

m3ta1head has some amazing cache latency numbers as well with his Ryzen+.

m3ta1head, are you able to get trfc any lower? I managed to hit 248. It might be worth an extra nanosecond or two.

Are you able to drop your tWRWRSD/DD from 7 to 6, and tRDRDSD/DD from 6 to 5? I'm noticing that both myself and m3ta1head are able to run it with that. (If you do, I'd just change them in their pairs, so both the SD and DD in WRWR, then test to make sure stability, before going back and changing the RDRDSD and DD. Then only once they're both considered stable, pass judgement on whether they're worth keeping there, since it could be one of those timings that on their own will not show much gain, but when both are adjusted you see the benefit...)
Further comparison between our three timings, I'm seeing that your tWTRS is at 6, where as our's is at 3. Alas, I don't know what impact that has on bandwidth, latency, or stability... But if you can achieve it at a voltage you're comfortable at (which B-Dies apparently are safe up to 1.5V), then what the hell right? lol I also see lots of people run the tCKE at 1 instead of 6, so is something else to look into trying also.

I've seen some people, albeit with insane modules to begin with such as 4133 kits, able to get their tRFC down into the 220s even at high speeds. I don't recall which kit he has so it may or may not be attainable for him. Though, he's also running 133Mhz faster than you are, too (3466 according to his AIDA, so I ignored the 3400 in his timings shot, chalking it up to having snapped it prior and not changed any timings).

I'm still out of town until around the 31st, but I'm raring to get back to tweaking lol Been at the GF's where my old Phenom II 1090T rig is, and so I've been scratching my itch of "MOAR PERFORMANCE!" by tuning that things lol Such a shame that we don't have access to the Infinity Fabric speed like what we did with the CPU-NB on S939 through AM3 days... I forgot how much memory bandwidth there was to uncork by increasing that [/off topic]

 

[m3ta1head]

So it looks like formula.350 ran into the same latency limits with his Ryzen as I did with my Raven Ridge.

m3ta1head has some amazing cache latency numbers as well with his Ryzen+.

m3ta1head, are you able to get trfc any lower? I managed to hit 248. It might be worth an extra nanosecond or two.

I tried going down to 236 and was getting hard crashes. Haven't tried anything between that and 256 - worth a shot though, I'll try dropping down and seeing if it makes a difference. Formula.350 I am now running a 3200C14 kit, and your assumption was correct regarding the discrepancy in timings/speed in the screenshot - I snapped that screenshot in RTM earlier, but kept the same timings going from 3400 to 3466 - only thing I changed was a slight bump in voltage.

 

[Nightfire]

Hmm, seems like I killed my B350 TUF. I dont think I used any crazy voltages, but it can get confusing at times with the offset.

It seems to have the same issue as my MSI Mortar B350 which arrived DOA. That board would be completely dead with everything hooked up, but the motherboard would light up fine with the cpu power unplugged.

The TUF B350 seems to light up ok with the cpu power disconnected, then does this weird slow blink with the cpu power connected.

Yes, I have reset cmos.

There seems to be an issue with power on the motherboard enroute to the cpu on both of these B350s now.

I have an a12-9800 that was fine, but also fails to boot now, so it is not the cpu.

 

[Nobu]

Hmm, seems like I killed my B350 TUF. I dont think I used any crazy voltages, but it can get confusing at times with the offset.

It seems to have the same issue as my MSI Mortar B350 which arrived DOA. That board would be completely dead with everything hooked up, but the motherboard would light up fine with the cpu power unplugged.

The TUF B350 seems to light up ok with the cpu power disconnected, then does this weird slow blink with the cpu power connected.

Yes, I have reset cmos.

There seems to be an issue with power on the motherboard enroute to the cpu on both of these B350s now.

I have an a12-9800 that was fine, but also fails to boot now, so it is not the cpu.

The usual softbrick fixes (unplug 24-pin, discharge capacitors, clear cmos, etc?) don't help? Seems like softbrick, but could just as well be dead—the symptoms are similar.

 

[Nightfire]

I cleared the cmos but I never discharged the capacitors. Not really sure how to do that. the motherboard has been unplugged from the pc and taken apart for a couple of days now.

 

[Nobu]

I cleared the cmos but I never discharged the capacitors. Not really sure how to do that. the motherboard has been unplugged from the pc and taken apart for a couple of days now.

See this post. Basically, remove power (disconnect/turn off psu), then press the power button and hold it a couple times. I think the gigabyte reps suggested disconnecting the 24 pin and/or removing the cmos battery too, but I don't know if it's required. It's a shot in the dark, as I think it was rare on non-gigabyte boards in the first place, but worth a shot.

They speculate that it works because there are some variables stored in memory on the cpu itself, and when they get in a bad state, the only way to reset them is to discharge the system completely. Not sure if that was because of the electrical configuration of those boards or some greater architectural flaw.

 

[BurntToast]

2700X
ASRock X470 Taichi
Corsair Vengeance RGB 16GB (2x8GB) DDR4 3466 CMR16GX4M2C3466C16

Using XMP profile 1 at 3200 16-18-18-36 1.35v the system is stable. Using the same profile at 3466 Prime95 will fail relatively fast. This memory is advertised to be compatible with Ryzen. I'm sure that the memory profile isn't setting timings correctly. Does anyone have any suggestions?