Post your Ryzen memory speeds!

IdiotInCharge

[H]ardForum Junkie
Joined
Jun 13, 2003
Messages
13,257
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.
One way to look at it is this: ATX always has power on, even when it's off, if it is connected, and there are a number of capacitors that help with that. Note that if you take the power away while a system is running it takes a bit for board LEDs to fade- there's a lot of power being 'buffered' there!

That also means that there are a lot of places to hold some nominal charge that could keep corrupt data from being cleared.

Now, the way to get the system to try and burn any charge off (as mentioned above) is to disconnect all power including the CMOS battery and push the power button. This opens all of the circuits and attempts to draw a load, hopefully draining any latent charge and clearing any volatile memory.
 

Nightfire

[H]ard|Gawd
Joined
Sep 7, 2017
Messages
2,039
I am now using an ASROCK AB350 mini ITX and get some nice results at 1.45v:

aida.PNG


62ns is pretty awesome for 14nm Ryzen and I may have been stable at higher speeds, but I am not willing to push my dram voltage past 1.45v

I still hate this motherboard as it does not allow iGPU overclocking. Even when using Ryzen master, I get a BSOD when I push even the tiniest overclock.
 

Neapolitan6th

[H]ard|Gawd
Joined
Nov 18, 2016
Messages
1,083
I am now using an ASROCK AB350 mini ITX and get some nice results at 1.45v:

View attachment 90882

62ns is pretty awesome for 14nm Ryzen and I may have been stable at higher speeds, but I am not willing to push my dram voltage past 1.45v

I still hate this motherboard as it does not allow iGPU overclocking. Even when using Ryzen master, I get a BSOD when I push even the tiniest overclock.
Yeah I've tried overclocking a 2200G on that guy. Very temperamental that one when overclocking more than 1 thing.

I do hate that I have to go through Ryzen Master as well. Come on Asrock. It was a friends build so I was hoping to have it OCed out of the box for them.

Were you really able to get a 2400G to 4.2ghz? Or is that just Cinebench being Cinebench? Great latencies btw.
 

Nightfire

[H]ard|Gawd
Joined
Sep 7, 2017
Messages
2,039
Were you really able to get a 2400G to 4.2ghz? Or is that just Cinebench being Cinebench? Great latencies btw.
Yes, although a bit of a suicide run in regards to volts. 4.1 ghz, however, requires just over 1.4v and I can run that 24/7 on quiet mode now that I am delidded.

My 925 is a satisifying score as it is right at half of ehat techshowdown got with a 4.3 ghz 1900x with 1855 points. This was the best score of any Summit Ridge cpus by reviewers.

With 4.1 ghz I get about 900, which is still plenty good.
 
Last edited:

Formula.350

[H]ard|Gawd
Joined
Sep 30, 2011
Messages
1,107
For those of us still rocking original Ryzen gear, I think the new AGESA is helping! (1.0.0.6

In April of 2018 I updated my BIOS on my Titanium to v1.F0 and, well, it borked my ability to achieve 3200 anymore, so I stuck it out at 3000 for quite awhile. After FO76 started having some hitching I decided I'd give a new BIOS a shot and grabbed v1.J0 (non-BETA). I was skeptical but, thankfully, it seems to have resolved the earlier issues and I'm back at 3200.

Context aside I've been trying to see where the limit is on this RAM, CPU, and what help the latest AGESA has given...
So far, I've managed to start loading Windows at DDR4-3600 (16-18-18-40 1.45V auto sub-timings) before instantly getting a BSoD. BUT that's amazing given I could barely get 3466 to POST!
I did get benches at 3466 (same as above but DDR @ 1.4V), but given the combined looser timings and auto sub-timings, it scored quite horribly. Additionally, my go-to stress test failed immediately so... damn.

Granted, MSI did quite a bit of work on the BIOS for my board as well, so it could be tweaks they've made in that area, too. For example they've reworked the voltage options and now let you either do Static voltage or run it with an Offset. I'm leaning more towards the AGESA as the main thing and the BIOS refinements lending a helping hand. :p

PS: I usually run CPU-NB voltage at 1.00V for 3200, but opted to let it run Auto for these max-POSTing speeds, which sets it to 1.2V
 
Joined
Nov 6, 2018
Messages
1
I have the Ryzen TR 2950x

Ram Speeds 3200MHz
Latency 14-14-14-34
I have 2 Kits installed but the Kit Model Number is CMW32GX4M4C3200C14

FYI my kit is EOL so not sure how long it will be around, but corsair has the best warranty and it is lifetime, so you can't beat that. I saw a white version of my kit for 279.99 on Newegg earlier today. I paid 449 so if you see the price i am talking about it is a steal (relatively speaking)
 

Srath

n00b
Joined
May 2, 2018
Messages
6
If youre having trouble with the newer agesa 1006 compared to earlier versions, try increasing tDDRL by one (credit numlock66 @ overclock.net forums)
ex: b die "fast" in dram calculator for ryzen+ recommends tRRDL 6, so mine is set to 7

running my 8x2 b-die at 3533 1.43v
3533 agesa 1006.png
 

Nightfire

[H]ard|Gawd
Joined
Sep 7, 2017
Messages
2,039
If youre having trouble with the newer agesa 1006 compared to earlier versions, try increasing tDDRL by one (credit numlock66 @ overclock.net forums)
ex: b die "fast" in dram calculator for ryzen+ recommends tRRDL 6, so mine is set to 7

running my 8x2 b-die at 3533 1.43v
View attachment 140377
Thanks for the tip. Can you run AIDA64? After all, faster speed means nothing if performance is lower.
 

Nightfire

[H]ard|Gawd
Joined
Sep 7, 2017
Messages
2,039
Here's 3 runs. I dont have anything of mine to compare it to. Cpu at stock boost if it matters.
What voltage are you running on the ram? Your cache speeds are insane compared to even m3ta1head who is running Pinnacle as well. It would be nice if you could get the latency down a bit. If there is no wiggle room on the primary timings (1.45v is perfectly safe), maybe lower your tRas and trc to something like 24 and 38 respectively.
 

Srath

n00b
Joined
May 2, 2018
Messages
6
What voltage are you running on the ram? Your cache speeds are insane compared to even m3ta1head who is running Pinnacle as well. It would be nice if you could get the latency down a bit. If there is no wiggle room on the primary timings (1.45v is perfectly safe), maybe lower your tRas and trc to something like 24 and 38 respectively.
1.43v on ram in bios
mostly just trying to follow the dram calculator (by 1usmus) @ 3533 fast settings. not really trying to make high benches or burn out my memory sticks

I have not messed with the ohm settings and dont really know much about them
 

Formula.350

[H]ard|Gawd
Joined
Sep 30, 2011
Messages
1,107
Trying running cl14 or lower daily with that msi
That's what I get away with for my 3200. It's a CL15 kit, but was fine at 14-14-14-30

It's unlikely I'd get CL14 at 3466 if that's what you're suggesting lol
I don't know enough to get that stable (ie the other voltages, drive strengths, or specific subtimings), otherwise 16-16-16-36 would be :p The AIDA Stability test on Cache fails instantly, and on RAM fails after 3 seconds.
I'll work on it over the weekend though, since I ran out of bandwidth for the month and it doesn't reset till Sunday :(


What voltage are you running on the ram? Your cache speeds are insane compared to even m3ta1head who is running Pinnacle as well.
While I don't have a Ryzen+ chip, I've had my L1 over 1000 before (wasn't able to get that stable though). I believe it was when I had once gotten 3466 into Windows with my 3200 timings Unfortunately I run a slightly older version of AIDA (about 1yr old) in order to keep my own benchmarks comparable, so I dunno exactly how much that will impact the ability to compare against Srath's. His latencies are better, but that's expected from a Ryzen+ chip. My RAM is 68-70ns, L1 same, L2 4.4ns, L3 11.xns.


If youre having trouble with the newer agesa 1006 compared to earlier versions, try increasing tDDRL by one (credit numlock66 @ overclock.net forums)
ex: b die "fast" in dram calculator for ryzen+ recommends tRRDL 6, so mine is set to 7

running my 8x2 b-die at 3533 1.43v]
What kit DO you have?
(I'm sure you're replying to Brandon, but am hopeful it might be applicable to my case, too heh)

I know most people who have luck with higher speeds have the better binned 3200 kits of CL14, but mine is a CL15 kit, so I'm not sure what sort of headroom I can expect with my 1700X. I did buy this one specifically (2 weeks before Ryzen launched) based on a review of it on an Intel system being able to attain 4000... of course, Intel controllers are wwaaaayyyy better for overclocking RAM as we've all unfortunately found out. :\ However, the RAM itself should, I would think, be capable of at least 3600 with respectable timings. Alas I come from a simpler time of RAM overclocking and thus I lack knowledge of what to adjust things at. I tried to get chew* to tell me his secrets, but respectfully he didn't want to (due to cut-throat extreme OCers, which I understand, given he used to be one).
 

JRZoid

[H]Lite
Joined
Feb 1, 2019
Messages
91
ran the same evga Hynix kit at the same 1.35v...it don't OC for shit...I tried...stuck with 3200. But a stable kit none the less.
 

Srath

n00b
Joined
May 2, 2018
Messages
6
That's what I get away with for my 3200. It's a CL15 kit, but was fine at 14-14-14-30
What kit DO you have?
(I'm sure you're replying to Brandon, but am hopeful it might be applicable to my case, too heh)

I know most people who have luck with higher speeds have the better binned 3200 kits of CL14, but mine is a CL15 kit, so I'm not sure what sort of headroom I can expect with my 1700X. I did buy this one specifically (2 weeks before Ryzen launched) based on a review of it on an Intel system being able to attain 4000... of course, Intel controllers are wwaaaayyyy better for overclocking RAM as we've all unfortunately found out. :\ However, the RAM itself should, I would think, be capable of at least 3600 with respectable timings. Alas I come from a simpler time of RAM overclocking and thus I lack knowledge of what to adjust things at. I tried to get chew* to tell me his secrets, but respectfully he didn't want to (due to cut-throat extreme OCers, which I understand, given he used to be one).
It is (gskill) trident z 3200 c14
F4-3200C14D-16GTZSK bought April 2018

My b-die ram is pretty average compared to other results in the asus c6h overclock thread.
It might run higher than 3533 but I stopped trying when soc was higher than I wanted for 24/7
 

Nightfire

[H]ard|Gawd
Joined
Sep 7, 2017
Messages
2,039
It is (gskill) trident z 3200 c14
F4-3200C14D-16GTZSK bought April 2018

My b-die ram is pretty average compared to other results in the asus c6h overclock thread.
It might run higher than 3533 but I stopped trying when soc was higher than I wanted for 24/7
It is hard to tell if it is the memory controller, CPU, or motherboard. I can only get 3266 mhz max on my current mb when I could manage 3466 with my old board. I am running 1.45 vdimm, 1.175 soc
 

elite.mafia

Broke Back [H]
Joined
Aug 23, 2004
Messages
12,143
T-Force Delta II RGB 16gb(8x2) ddr4-3000 running at 2933 c16-18-18-38-56 @1.36v so basically stock but 1 step slower on the frequency due to my board.
 

Formula.350

[H]ard|Gawd
Joined
Sep 30, 2011
Messages
1,107
It is (gskill) trident z 3200 c14
F4-3200C14D-16GTZSK bought April 2018

My b-die ram is pretty average compared to other results in the asus c6h overclock thread.
It might run higher than 3533 but I stopped trying when soc was higher than I wanted for 24/7
Thanks.

I just realized that Ryzen Timing Checker is incorrectly reading your setup as being in Single Channel mode heh (completely contradicts your memory bandwidth and what AIDA reports, which is why I know it's wrong)
upload_2019-2-9_15-36-8.png

Update: It's a bug with v1.05 as I just updated and mine now displays the same 1DPC. At least right up until I booted at defaults (2133) and now it's reading 2DPC o_0

Anyways, for kicks, I tossed in your timings and drive strengths into my Titanium at 3466, but at 15-15-15-30 (and tRAS 45) since that's what my kit is. At 1.4v it didn't boot, but I upped it to 1.45v... and well irrelevant now that I got my first ever WerFault error window lol I had a feeling this was going to be way too tight for my kit :p Especially given some of yours were tighter than what I had been running for even 3200!

Interestingly, though, the drive strengths all are reporting as 120Ohm instead of 20Ohm... Update: v1.05 properly reflects these now on mine, as I'm sure it was due to the AGESA being updated.

EDIT: *grumbles* I can't help but wonder if it's just my IMC that is the weak link. Used 16-16-16-32 at loosened 3200 subs and the AIDA CacheMem bench is awesome compared to 3200, but it still fails the Stress Test right away for Cache and System Memory (ran separately). I'll grab the updated DRAM Calc, but it's never provided me any help in the past, so fingers crossed!
 
Last edited:

Srath

n00b
Joined
May 2, 2018
Messages
6
Thanks.

I just realized that Ryzen Timing Checker is incorrectly reading your setup as being in Single Channel mode heh (completely contradicts your memory bandwidth and what AIDA reports, which is why I know it's wrong)
View attachment 140803

Anyways, for kicks, I tossed in your timings and drive strengths into my Titanium at 3466, but at 15-15-15-30 (and tRAS 45) since that's what my kit is. At 1.4v it didn't boot, but I upped it to 1.45v... and well irrelevant now that I got my first ever WerFault error window lol I had a feeling this was going to be way too tight for my kit :p Especially given some of yours were tighter than what I had been running for even 3200!

Interestingly, though, the drive strengths all are reporting as 120Ohm instead of 20Ohm...
1 dimm per channel, single rank; two sticks total

Guess your kit is weird, probably would need a lot of manual tuning and not worth the time. DRAM calculator has a v2 for medium quality ram but it is super loose like c16 probably still not worth it, will take a screenshot.
3533 v2med gen1.jpg

Ohm numbers probably more motherboard related. Mine on auto it set most of mine to 24

I went from a 4790k to ryzen, on an x370 and 2700x.
Had to go asus for the usb bios recovery because others wouldnt have booted. Don't like MSI for amd boards anyways because they usually use low quality vrm
 

Formula.350

[H]ard|Gawd
Joined
Sep 30, 2011
Messages
1,107
1 dimm per channel, single rank; two sticks total

Guess your kit is weird, probably would need a lot of manual tuning and not worth the time. DRAM calculator has a v2 for medium quality ram but it is super loose like c16 probably still not worth it, will take a screenshot.View attachment 140808
Ohm numbers probably more motherboard related. Mine on auto it set most of mine to 24

I went from a 4790k to ryzen, on an x370 and 2700x.
Had to go asus for the usb bios recovery because others wouldnt have booted. Don't like MSI for amd boards anyways because they usually use low quality vrm
Yea I saw that V2 profile and used it for most of my attempts today.

I tried all three CAD_BUS suggestions and none would POST. Mine defaults to 120Ohm. What little I do know, is that indeed those settings, and all timings, are very much board dependent.

I just suspect it's my IMC that isn't the best. I'm sure having a C6H would help some, but I don't know if my RAM is entirely to blame here. It can get some of the subs down fairly low and at 1.36V

If Zen2 comes out this year then we'll see what my RAM and board are capable of, since they'll be reused.
 

crazycrave

Gawd
Joined
Mar 31, 2016
Messages
647
I have my 1600 Fire Strike stable at 3466Mhz but i had to back the timings off to 18 - 20-20-20-38 @1.4v as it blue screened with 16-18-18-18 starting Fire Strike http://www.3dmark.com/fs/18373775

Noting overclocked
Physics Score 15 763 I see about 200 point gain going from 3066 to 3466 in memory speed alone if the 2200G lives on speed maybe it would work for mine if Gen 1 can run it , but it's not far off of a 2600 now and it's not overclocked just feed memory speed ,
 
Last edited:

pitingres

[H]Lite
Joined
Jul 25, 2018
Messages
95
Finally got my 2x16 G.skill TridentZ 3200CL14 to run at 3200 (with a 2700X and Asrock X470 Taichi Ultimate). It's been running stable at 3133 but I could never get it to 3200. Latest BIOS update with AGESA 1.0.0.6, another try at the ryzen dram configurator settings, and a 0.01 uptick in voltage seems to have done the trick. Not that I'll ever notice the added 67 MT/s, but it's nice to have the stuff running at the speed I paid for!
 

Below Ambient

[H]ard|Gawd
Joined
Dec 5, 2001
Messages
1,440
Finally got my 2x16 G.skill TridentZ 3200CL14 to run at 3200 (with a 2700X and Asrock X470 Taichi Ultimate). It's been running stable at 3133 but I could never get it to 3200. Latest BIOS update with AGESA 1.0.0.6, another try at the ryzen dram configurator settings, and a 0.01 uptick in voltage seems to have done the trick. Not that I'll ever notice the added 67 MT/s, but it's nice to have the stuff running at the speed I paid for!
If this xmp why did you wait so long...you just need to add a tad bit of voltage...b die can do 1.6v all day long
 

crazycrave

Gawd
Joined
Mar 31, 2016
Messages
647
I have back down to 3333Mhz at 16-18-18-18 - 36 @ 1.36v on G SKill Rip Jaw Black 3200Mhz CL 16 kit with my Ryzen 1600 as to test the memory speed change I leave cpu stock clocked and that is the highest is has scored in
Physics Score 15 805 in FS http://www.3dmark.com/fs/18404852
 

Bigbacon

[H]ardForum Junkie
Joined
Jul 12, 2007
Messages
16,237
Havent tried to oc it but my 4x8 ballistix sport is runnung fi e at 3200mhz using the xmp.

Msi x470 gaming plus. 2700x
 

pitingres

[H]Lite
Joined
Jul 25, 2018
Messages
95
If this xmp why did you wait so long...you just need to add a tad bit of voltage...b die can do 1.6v all day long
Because it's a production machine (i.e. pays the mortgage, it's not a gamer) and until very recently I haven't had the time or inclination to fool around with it.
 

Below Ambient

[H]ard|Gawd
Joined
Dec 5, 2001
Messages
1,440

JRZoid

[H]Lite
Joined
Feb 1, 2019
Messages
91
I'm just at 3200 SoC is stock...and the CL16 Hynix kit CR1 is fine...Like I said it passes benches at 4.2 like 1.387v...it froze past that anything past it...could not get stable...they sent a doggier chip...guaranteed 4.2....good ones stable at 4.3....

It's negligible though anyways....just run 4.2 and set LLC High/max and yeah should be around 1.37-1.38v. sweet spot.

Trident z you might be able to run 3333-3466....it's very possible. or some other kit.


Still not fully stable in benches half the time...it's broke shit..idk...
Just game or something...that's fine....
 
Last edited:

Bigbacon

[H]ardForum Junkie
Joined
Jul 12, 2007
Messages
16,237
think I killed my windows 10 install trying to muck around with timings. lowed the latency numbers by 1 except tRAS and now windows won't boot. Even going back into the bios and changing it back to all auto stuff doesn't fix it. :(
 

JRZoid

[H]Lite
Joined
Feb 1, 2019
Messages
91
think I killed my windows 10 install trying to muck around with timings. lowed the latency numbers by 1 except tRAS and now windows won't boot. Even going back into the bios and changing it back to all auto stuff doesn't fix it. :(
Clr cmos….that sucks...I'm doing great again re-applied paste/reseated the cooler...its passing everything at 4.2.
 

Formula.350

[H]ard|Gawd
Joined
Sep 30, 2011
Messages
1,107
think I killed my windows 10 install trying to muck around with timings. lowed the latency numbers by 1 except tRAS and now windows won't boot. Even going back into the bios and changing it back to all auto stuff doesn't fix it. :(
That's some seriously bad luck if so... I've crashed this Win10 install I'm on about a hundred times just playing around with timings and clocks. Originally was on a HDD and then I copied it over to an nVME drive.

My guess would be that perhaps the BIOS tried to recover the fault and set some things to defaults (despite displaying what you had originally set) and that changed the Drive Boot Order.
For example on my Titanium when it recovers, it says I still have my CPU clock in place (38.5X) and even if I've been into the BIOS and saved different settings elsewehre (timings, for example), it actually isn't setting the CPU.

So as JRZoid said, clear the CMOS if you haven't. Can be annoying having to set things back up, but it's less annoying than having to reinstall Windows and get it configured again. :)
 

Bigbacon

[H]ardForum Junkie
Joined
Jul 12, 2007
Messages
16,237
I got back in. safe mode. I think I messedup because I used the stupid MSI software to change it.....I need to NOT do that and just use the bios.

Back to stock for now...i don't think it matters that much for me anyways.
 

Formula.350

[H]ard|Gawd
Joined
Sep 30, 2011
Messages
1,107
I got back in. safe mode. I think I messedup because I used the stupid MSI software to change it.....I need to NOT do that and just use the bios.

Back to stock for now...i don't think it matters that much for me anyways.
Yea the MSI motherboard software is shit. Though, in fairness, overclocking from Windows with Ryzen is shit altogether. lol

I miss the days of Athlon 64. "A64Info" was the best tweak program! It did everything. FSB, HTT, Timings, Sub-timings! If memory serves me, the only thing it couldn't do from Windows was change the RAM divisors (clocks), but it may have even been able to do that. That program sure spoiled me... heh
 

ThreeDee

[H]ardForum Junkie
Joined
Sep 5, 2001
Messages
10,727
Ryzen 2700x on Asrock X470 Master SLI/ac w/ 2 x 8Gb V-Color SKYWALKER PRISM RGB 3200 Stock Timings are 16-18-18-38.

I have them running at 3400 15-18-18-38 auto voltage so far. I haven't messed with the Ryzen memory calculator thingy yet. I had them running at same timings at 3466 and everything seemed fine until I was playing Hunt:Showdown and then the game just turned off .. no lock ups or stutters or anything prior, game was running great but then just ...turned off after playing in a match for about 25 minutes and went to desktop. no errors.. don't know if was latest Nvidia drivers I just installed or too fast of memory timings so just dialed back to 3400. I'll run some tests and mess with setting some more when I get more time. But I have been happy with this RGB memory kit.
 

Formula.350

[H]ard|Gawd
Joined
Sep 30, 2011
Messages
1,107
Ryzen 2700x on Asrock X470 Master SLI/ac w/ 2 x 8Gb V-Color SKYWALKER PRISM RGB 3200 Stock Timings are 16-18-18-38.

I have them running at 3400 15-18-18-38 auto voltage so far. I haven't messed with the Ryzen memory calculator thingy yet. I had them running at same timings at 3466 and everything seemed fine until I was playing Hunt:Showdown and then the game just turned off .. no lock ups or stutters or anything prior, game was running great but then just ...turned off after playing in a match for about 25 minutes and went to desktop. no errors.. don't know if was latest Nvidia drivers I just installed or too fast of memory timings so just dialed back to 3400. I'll run some tests and mess with setting some more when I get more time. But I have been happy with this RGB memory kit.
My go-to initial stress testing program is AIDA64 and then in it's Stress Test window, using just the "Cache" test. I've found that on Ryzen, not only does it incur the heaviest load in terms of wattage (FPU being second, edit: FPU generates more heat though), but seems to test the RAM and Memory controller. As such, it will generally either fail right away (and possibly lock the system if things are really unstable) or pass for 5+ minutes. [Note: my accounting here has a sample size of 1 system, so I'm not implying this will factually happen for everyone; however, I do expect similar results. If anyone has similar experiences, it'd be nice to hear, so to increase my sample size... heh]

I think the first course of action would be to bump the RAM voltage a smidge. No clue what it's running at currently, but it's probably at 1.35V, unless what my MSI does while on Auto is actually adhering to AMD guidelines and thus your ASRock will apply the same... Which in that case, it could be set to higher than 1.35V. Not sure if yours are Samsung B-Die or not (the timings to me indicate Hynix, but I know Corsair will stick those offset timings on B-Dies occasionally), so I'd put a max-cap of not going over 1.45V on it just to be safe.

Even if it is B-Die, and no offense to Below Ambient in regards to his comment a number of comments ago, but I'm not certain B-Die is robust enough to handle 1.6V on air. In fairness to him, I've not bothered looking at Ryzen and RAM reports as to how B-Die (or any DDR4) cope with such voltages, so now that both Intel and AMD are running DDR4, there may be more data to pull from. Last I had seen the max for B-Die was being said as 1.5V to 1.525V for air (mind you, I'd have a fan aimed in the RAM's direction regardless, even at 1.35V, just for peace of mind and assured stability ;)).


Second thing I'd do would just be put CAS back to 16T (for 3466 stability), as you're not going to see much impact in 90% of situations, even in memory benchmarks (from my experience). Your biggest improvements to overall latencies and bandwidth will come from tighter subtimings, which if the ASRock X470 Master is anywhere near as good as the Taichi, then I'm sure you should have fairly good success with subs. Mind you, I don't pretend to be an expert, in fact I actually make it a point to say I am quite ignorant with timings, but if you post an image of RyzenTimingsChecker then I or someone else can point out subtimings that could have room for improvement. :)


Last thing would be looking into the CPU-NB or CPU-SOC (sadly the labeling is vendor dependent) which is where the Memory Controller resides. Referencing the auto-voltages again, on my board anything over 2933 applies 1.20V, yet my system is perfectly fine with 1.0V at 3200, meaning if yours is applying more than necessary, it could easily increase the heat and induce instability. Even the Ryzen DRAM Calc suggests a much lower voltage. But again, keep in mind that this is my experience with just my MSI Titanium, and I also know that MSI doesn't put anywhere near the work into their BIOS that ASUS and ASRock do, so your Auto settings may not be lazily generic as mine :p


EDIT: Ok I spoke too soon I guess. On the original BIOS (and a few subsequent releases) I was stable with the NB at 1V... but with this new AGESA it seems it wasn't that thrilled by so little voltage, as I'm failing Cache stress now even though the RAM didn't. I bumped the voltage to 1.0250V just to make sure it wasn't VDroop, but that actually resulted in it failing faster, go figure. At any rate, I decided to just "chuck voltage" at it and increased it substantially to 1.1V which is undergoing testing now, having just passed the 6min mark without failing, so it's looking promising (y) Jinxed myself *sigh* Failed at roughly 7min mark, just after having posted this edit (n). Guess I'll crank it up to 1.15V, and if that fails, 1.2V. Should that fail... it's back to the drawing board. :sour:
UPDATE: Meh I dunno now...lol
I let it idle for an hour or so, still at 1.1V, but before I changed the voltage I re-ran the Cache stress and so far it's been chugging along fine for >32min w/o a faliure. So it probably only needs 1.11V on the CPU-SoC, and that first one was just an almost-fluke.
 
Last edited:

chrcoluk

[H]ard|Gawd
Joined
Jul 7, 2008
Messages
1,214
I have 2 sticks of 8 gig hynix A die, marketed spec is 3000CL16.
AGESA 1.0.0.4 on asrock b450 pro and 2600X cpu.

The ram using ryzen dram calculator tool I have both 3000CL14 and 3200CL16 working fine and stable, 3200CL14 hasnt caused any OS crashes but does fail ram stability testing. I think I will be settling on 3200CL16 providing one last stability test turns out ok.

I have been goign backwards and forwards between choosing the lower latency on 3000 or the throughput on 3200, but then noticed disabling geardown on 3200 was very beneficial and aida64 reports both higher bandwidth and lower latency now on the 3200CL16 config.

On my motherboard the BackGroupSwap setting does absolutely nothing, sadly is a few settings in the advanced AMD setting that seem to be for show only and not coded to do anything. I dont know if its fixed in 1.0.0.6 update as I never tried it
 

chrcoluk

[H]ard|Gawd
Joined
Jul 7, 2008
Messages
1,214
What voltage are you running on the ram? Your cache speeds are insane compared to even m3ta1head who is running Pinnacle as well. It would be nice if you could get the latency down a bit. If there is no wiggle room on the primary timings (1.45v is perfectly safe), maybe lower your tRas and trc to something like 24 and 38 respectively.
Does ram speed or core count affect cache speed? your L1 and L2 cache speed is 200 higher then mine o_O. On my 2600X your clock speed during your test is only 100mhz higher so wouldnt be the reason for the big gap.
 

Formula.350

[H]ard|Gawd
Joined
Sep 30, 2011
Messages
1,107
Does ram speed or core count affect cache speed? your L1 and L2 cache speed is 200 higher then mine o_O. On my 2600X your clock speed during your test is only 100mhz higher so wouldnt be the reason for the big gap.
L1 and L2 run at CPU speed, so that actually IS likely why. At 3.85GHz on my 1700X my L1 is ~950GB/s and my L2 is ~900GB/s, whereas at 3.7GHz it's quite a bit lower (don't recall off hand and don't have my desktop running to pull the notes from).


As far as your 3000 CL14 vs 3200 CL16 debate goes...
Since you have GearDownMode disabled, you should be able to utilize Odd-numbered latencies. For example, 3200 at CL15.
I'm assuming your timings are 16-18-18 @ 3200, and so you're just changing it to 14-18-18 @ 3000? Or are you making it 14-16-16 to match?
If they are of that nature, you could give, say 15-17-17-34 @ 3200 a shot. You'll know if your BIOS is not setup to lock-out Odd CL timings by loading Windows and checking timings with anything (AIDA, CPUz, etc). If it reads CL15 then you're good. If it reads CL16 then it's locked-out.

[Note: I've posted my findings on this in the thread before, but as to which page it resides on now I'm not certain... but I'm 90% sure that GearDown=Disabled means the RAM can "latch" at half-timings, which translated to odd-numbers. Enabled (Default/Auto) then I believe means it only will latch whole numbers, such as 1 or 2... Adding 1+1 gets you 2, 2+2=4, 3+3=6, etc...all evens. Whereas disabled it can do 1.5+1.5=3, 2.5+2.5=5, etc... all odd numbers. It's a crude explanation but is the easiest way to explain it I feel.]
 
Top