Post your Zen 2/2+ memory speeds!

Possibly the fastest 32gb memory setups here. Beast of a machine you have there.


He isn't the slowest, but there are a few of us running 3600c14 and quite a few running 4700-3733 c16-18/19.

jhatfie, dud you set the recommended ProcsODT and CAD BuS (and other CSD settings)? What voltage are you running for VDDR/VCCP voltage? I needed 1.45V to run my dual rank B die sticks at 3600c14.

The fact that you have SR DIMMs should not hinder your ram speed even though you have 4 physical DIMMs. Since you are running SR sticks that is still only 2 banks per channel which means you should get the minimum 3200 speed and should be able to run 34/3533/3600 with the right settings.

Do you know if you know uses T topology or Daisy chained memory setup by Chance? I think I remember reading most x570 were Daisy chained but I might be wrong.


Also, what is your SOC Voltage? Do you have SoV current set to 110/120%? I had to run 1.1V and 120% for my speed.
 
4 sticks of B-Die, 32gb, 3733, ASUS C6H, 3900x

2019-08-21.png
 
Nice! What are your setting/voltage? I'm pretty sure you 3900X guys have an overall bandwidth advantage.
1.4v boot also 1.4v for the memory and the following settings, I am slowly tweaking the sub timings. I have a slightly slower setting which I tested with Memtest and Prime95. I probably do more tweaks and then retest.


MemSettings.jpg
 
G.Skill F4-4000C19-16GTZR.JPG
3733 mem 1866 fclk DRAM Ryzen Calculator 1.6.0.3 Safe Preset 3733 RAM 1.485v CPU -0.1 Offset.png
3733 DRAM Calc 1.6.0.3 3733 Safe Preset at 1.47v CPU offset -.1.JPG


Fclk will go no higher than 1866, but it will work there with voltage at Auto. I'm running the RAM at 3733 using the Safe settings from the Ryzen DRAM Calculator for 3733, but I need way more voltage for stability. I have DRAM voltage at 1.485v and SOC voltage at 1.1v. CPU is -0.1 offset on voltage. BTW, it seems the current calculator is flawed, Safe settings are tighter and recommended voltage is higher than Fast, like they've been reversed.

Now, to my issue, something just ain't right. Based on my timings and RAM speed, my latency should definitely be lower. I ran the Memtest feature on the DRAM calculator and it started throwing up errors fairly quickly, so I guess that's it. I rebooted at default optimized settings with the RAM at 2133 and ran Memtest for about a half hour and got no errors. I don't know if it's the motherboard (Asus x570 Hero Crosshair VIII) or the RAM sticks themselves, but I guess I'll try loosening timings a bit and see what happens before I drop down to 3600. I really don't want to raise the RAM voltage anymore.
 

Attachments

  • 3733 mem 1866 fclk DRAM Ryzen Calculator 1.6.0.3 Safe Preset 3733 RAM 1.485v CPU -0.1 Offset.png
    3733 mem 1866 fclk DRAM Ryzen Calculator 1.6.0.3 Safe Preset 3733 RAM 1.485v CPU -0.1 Offset.png
    92.3 KB · Views: 0
  • G.Skill F4-4000C19-16GTZR.JPG
    G.Skill F4-4000C19-16GTZR.JPG
    136.2 KB · Views: 0
Last edited:
dud you set the recommended ProcsODT and CAD BuS (and other CSD settings)? What voltage are you running for VDDR/VCCP voltage? I needed 1.45V to run my dual rank B die sticks at 3600c14.

The fact that you have SR DIMMs should not hinder your ram speed even though you have 4 physical DIMMs. Since you are running SR sticks that is still only 2 banks per channel which means you should get the minimum 3200 speed and should be able to run 34/3533/3600 with the right settings.

Do you know if you know uses T topology or Daisy chained memory setup by Chance? I think I remember reading most x570 were Daisy chained but I might be wrong.


Also, what is your SOC Voltage? Do you have SoV current set to 110/120%? I had to run 1.1V and 120% for my speed.

I did put in the recommended voltages, even tried loosening the timings a bit. But honestly I did not play with it for a long time. I'll give it another go to see if I can improve it further, but sort of waiting for the next bios update before fiddling more. Overall I am pretty happy with the performance.
 
View attachment 182198 View attachment 182199 View attachment 182200

Fclk will go no higher than 1866, but it will work there with voltage at Auto. I'm running the RAM at 3733 using the Safe settings from the Ryzen DRAM Calculator for 3733, but I need way more voltage for stability. I have DRAM voltage at 1.485v and SOC voltage at 1.1v. CPU is -0.1 offset on voltage. BTW, it seems the current calculator is flawed, Safe settings are tighter and recommended voltage is higher than Fast, like they've been reversed.

Now, to my issue, something just ain't right. Based on my timings and RAM speed, my latency should definitely be lower. I ran the Memtest feature on the DRAM calculator and it started throwing up errors fairly quickly, so I guess that's it. I rebooted at default optimized settings with the RAM at 2133 and ran Memtest for about a half hour and got no errors. I don't know if it's the motherboard (Asus x570 Hero Crosshair VIII) or the RAM sticks themselves, but I guess I'll try loosening timings a bit and see what happens before I drop down to 3600. I really don't want to raise the RAM voltage anymore.

I guess I'm going to have to do some extended RAM testing this weekend when I can squeeze it in. I've noticed that Ryzen Master doesn't report my MEM VDDIO or MEM VTT, it just shows 0. Now, I have it set for the FAST timings from the DRAM calculator which is 16-16-16-32 and I have voltage at 1.44 which is the max recommended. The SAFE settings seem out of reach for my B Dies. It seems my sticks require more voltage for any kind of stability. I have noticed that BIOS reports slightly lower voltage than what they are set to. Now, with the FAST settings, I'm getting Latency of 65.1ns, which is great. When I run the memtest in the DRAM calculator, I don't get any errors anymore, but my computer will just reboot after about 20 to 30 minutes with no warning. The computer doesn't shut completely off either, all the fans and everything keep spinning like there is no power down. Windows logs just show the generic critical error for an unexpected shutdown. I don't know if this is an OS issue because I borked it before with RAM errors and had to rebuild the networking stack or if its a hardware problem. Do you guys have any ideas? It does seem that my RAM is power hungry. I should be able to boot to memtest tonight and let it run to hopefully isolate the hardware from Windows.
 
I haven't done anything other than enable XMP but I am getting the advertised speeds of the kit I bought, 32GB (4x8) Trident Z Neo 3600 CL16.16.16.36.
 
I haven't done anything other than enable XMP but I am getting the advertised speeds of the kit I bought, 32GB (4x8) Trident Z Neo 3600 CL16.16.16.36.

I have a very similar system to yours. I would love to see screenshots of all your settings in BIOS. I'm now able to do 3733 at 16-16-16-30, but with high voltages on everything. I would love to pare it down some. I'm scoring below 7200 in cinebench, so I think I'm hitting TDP limits a little quickly.
 
I have a very similar system to yours. I would love to see screenshots of all your settings in BIOS. I'm now able to do 3733 at 16-16-16-30, but with high voltages on everything. I would love to pare it down some. I'm scoring below 7200 in cinebench, so I think I'm hitting TDP limits a little quickly.
Absolutely, I'll grab a screenshot tomorrow, I didn't do any manual tweaking though, just set D.O.C.P. to profile 3 (OC) and it started running the RAM at it's advertised speeds.
 
Absolutely, I'll grab a screenshot tomorrow, I didn't do any manual tweaking though, just set D.O.C.P. to profile 3 (OC) and it started running the RAM at it's advertised speeds.
Thanks, I'm interested in voltage settings, Digi Power settings, etc. My RAM will not boot the system at BIOS defaults with XMP enabled (4000 at 19-19-19-39 1.35v). Of course, I want to run at 3600 or 3733, so it's irrelevant, but I would think it should work. I've had to tweak every setting in the BIOS to get where I am now.
 
Decided to try some Gskill Trident RGB DDR4-3200 C14 kit (2x16GB) when it was on sale at NewEgg the other day. Right away am having better luck than with my 4 stick Corsair setup. Need to run memtest for an extended period, but I am liking what I am seeing.

gskill_3600_c16_3.PNG
 
Last edited:
Decided to try some Gskill Trident RGB DDR4-3200 C14 kit (2x16GB) when it was on sale at NewEgg the other day. Right away am having better luck than with my 4 stick Corsair setup. Need to run memtest for an extended period, but I am liking what I am seeing.

View attachment 183933

Looks good! I'm running at G.Skill 2x16 kit at 3733 at 16-16-16-32 and I'm getting 65.2 ns latency. I'm using the DRAM calculator 3733 Fast settings. What are you using for timings?
 
Using a 2x16 3200mhz crucial e-die kit. Enabled xmp and set it to 3600 with no problems. Will be tweaking when i get some time
 
I've got Corsair Dominator 3600 c18 B die

Running 3600 c14

Stupid quick ram. I'm gonna OC it to 4200 c 17 or 18 soon.

Isn't that pointless though where you have to up the divider? which will reduce performance. Sweet spot it 3733-3800 but IF clock is not guaranteed after 1800 though. I can do 3733 fine but at 3800 after 2 hours of OCCT I had a shutdown for some reason. But 3733 c16 I can run OCCT for hours. Still trying to figure out why it shut down as them temps were fine.
 
Isn't that pointless though where you have to up the divider? which will reduce performance. Sweet spot it 3733-3800 but IF clock is not guaranteed after 1800 though. I can do 3733 fine but at 3800 after 2 hours of OCCT I had a shutdown for some reason. But 3733 c16 I can run OCCT for hours. Still trying to figure out why it shut down as them temps were fine.

Not pointless. I want to see what this ram can do. Dont care about amd performance. I'll go back to 3600 14 for that.
 
I've gotten 3533c14 completely stable over the last few days. I am staying under 3600 right now to ensure that boost isn't affected and I am tired of tweaking ram lol.
 
I've gotten 3533c14 completely stable over the last few days. I am staying under 3600 right now to ensure that boost isn't affected and I am tired of tweaking ram lol.

I am running my 32gb B-die gskill 3200mhz c14 at 3733 16-16-16-32 and subtiming tweaks. 64ns in Aida64 and down to 62ns in other programs.

All core OC 4350 at 1.31v. Pretty happy with it.
 
Thanks, I'm interested in voltage settings, Digi Power settings, etc. My RAM will not boot the system at BIOS defaults with XMP enabled (4000 at 19-19-19-39 1.35v). Of course, I want to run at 3600 or 3733, so it's irrelevant, but I would think it should work. I've had to tweak every setting in the BIOS to get where I am now.
Sorry about the delay but here is what I have going on, xmp profile 3.


Screenshot 2019-09-06 13.01.57.png
 
Thanks man, going to give that a shot and run memtest tonight. Looks like low voltage on DRAM and VDDP compared to me.

I just re-installed Windows today and ran some benchmarks on a clean system, no HW monitor. Thanks to some settings recommended here, I'm running 3733 at 16-15-15-32. I still need 1.43v to work. Aida latency is 64.2. Cinebench multi was 7232.
 
Upped memory speed to 3800, all I had to do was increase SOC voltage to 1.15v - all timings stayed pretty much the same. Memory voltage still at 1.4v. MemTestPro passed without issue while watching video's and other stuff for at least 400% - looks good so far.

3800memoryTest.jpg
 
Upped memory speed to 3800, all I had to do was increase SOC voltage to 1.15v - all timings stayed pretty much the same. Memory voltage still at 1.4v. MemTestPro passed without issue while watching video's and other stuff for at least 400% - looks good so far.


Cool! I never dared to push my SOC voltage that high, but I could never get fclk 1900 to even come close to booting. Did yours fail to boot at 1900 until you pushed up the voltage or did it just boot, but was unstable at lower SOC voltages?
 
Cool! I never dared to push my SOC voltage that high, but I could never get fclk 1900 to even come close to booting. Did yours fail to boot at 1900 until you pushed up the voltage or did it just boot, but was unstable at lower SOC voltages?
SOC voltage in Auto was around the 1.09v, and 1900 would fail in memtest. With increase SOC voltage that is no longer the case and is working good so far.

I have 4 sticks of ram so it makes sense I would need some more SOC voltage as in more than others.
 
SOC voltage in Auto was around the 1.09v, and 1900 would fail in memtest. With increase SOC voltage that is no longer the case and is working good so far.

I have 4 sticks of ram so it makes sense I would need some more SOC voltage as in more than others.

Thanks for the info. Sounds like you've got good infinity fabric on that chip. Mine will not boot on any safe SOC voltage at 1900, so I think I'll just stick with 1866.
 
So was toying around with ram and let me tell you what I noticed.

So apparently CB r15 and 20 are not latency intensive as far as ram.

I got a 3100'ish score with my ram at 1000'ish mhz whatever base speed of the module is at XMP but set multi to auto. On Gbyte bios auto means lowest module speed per jdec. But I left the Inf Fab at 1800mhz.

Then all I did was change multi to 36 not touching the XMP profile and got a whopping like 3150 or a 50 point gain. Which is nothing quantifiable.

However I loaded titanfall 2 before and after. Before at slow ram speed on my 2080ti I got avg 110 to 120fps 1080p.

Then at 3600mhz ram it jumped to 144mhz pegged with zero drop. It was just maxed period.

So clearly ram speed has little to no effect on cinebench. But gaming makes a huge difference. Well I tested one old 2016 title so.

I found this by sheer accident. Is this remarkable information? Well no, but maybe it's something that is not well known.

Then all I did was change my multiplier to 36.
 
So clearly ram speed has little to no effect on cinebench. But gaming makes a huge difference. Well I tested one old 2016 title so.

I'm actually a bit surprised -- the workload that Cinebench represents should be bandwidth but not latency sensitive, but it seems that it is neither, at least at the levels of bandwidth you tested. Perhaps, and I'm not asking you to do this unless you just want to try it, if RAM speed were set to say 2400 there would be a more marked difference?


With respect to Titanfall 2, I'd consider it modern enough. It likely lacks the most modern eye candy, but as a 'workload', it's representative enough of modern gaming in my opinion. But I'd also ask: what did the longest frametimes do with the change in RAM speed?
 
I'm actually a bit surprised -- the workload that Cinebench represents should be bandwidth but not latency sensitive, but it seems that it is neither, at least at the levels of bandwidth you tested. Perhaps, and I'm not asking you to do this unless you just want to try it, if RAM speed were set to say 2400 there would be a more marked difference?


With respect to Titanfall 2, I'd consider it modern enough. It likely lacks the most modern eye candy, but as a 'workload', it's representative enough of modern gaming in my opinion. But I'd also ask: what did the longest frametimes do with the change in RAM speed?

I'd have to test to answer those specific questions
 
So was toying around with ram and let me tell you what I noticed.

So apparently CB r15 and 20 are not latency intensive as far as ram.

I got a 3100'ish score with my ram at 1000'ish mhz whatever base speed of the module is at XMP but set multi to auto. On Gbyte bios auto means lowest module speed per jdec. But I left the Inf Fab at 1800mhz.

Then all I did was change multi to 36 not touching the XMP profile and got a whopping like 3150 or a 50 point gain. Which is nothing quantifiable.

However I loaded titanfall 2 before and after. Before at slow ram speed on my 2080ti I got avg 110 to 120fps 1080p.

Then at 3600mhz ram it jumped to 144mhz pegged with zero drop. It was just maxed period.

So clearly ram speed has little to no effect on cinebench. But gaming makes a huge difference. Well I tested one old 2016 title so.

I found this by sheer accident. Is this remarkable information? Well no, but maybe it's something that is not well known.

Then all I did was change my multiplier to 36.

It really depends on the game. I know Hardware Unboxed, and Gamers Nexus did testing and found huge increases in some games/apps when manually tweaking memory. It pays to get really good memory speeds/timings now!

 
CB 15 and 20 pretty much works in the CPU cache’s so memory speeds have little impact. This goes for most rendering programs.
 
CB 15 and 20 pretty much works in the CPU cache’s so memory speeds have little impact. This goes for most rendering programs.

yeah its clear to see that from my post above where I discovered this by sheer accident lol ... its good to see confirmation of this from others like you

I dont run rendering software so I am not up to par in knowledge of exactly what and how they leverage resources
 
V-Color SKYWALKER PRISM RGB DDR4 3200 (PC4 25600)
Timing 16-18-18-38

..not spectacular .. but I'm running 4 x 8GB modules. I couldn't do that with my 2700x

 
Back
Top