Finally: Overclocking EPYC Rome ES

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

madbrad

New Member
Dec 10, 2020
15
0
1
sounds like a PSU issue honestly. try a different PSU.

were you using the wrong pins for the power button before?
Yes, I was using the wrong pins.
UPDATE 5: I got it to work, thanks to all of you!!! I tested the mother board out of the case and it works. Checked the case, found a stray support screw that was causing the short, removed it and booted windows 10. My cooler is revving every 10 seconds or so, I'll check that later. Haven't tried any OC yet, will update after I add the peripherals in a couple hours.
 

madbrad

New Member
Dec 10, 2020
15
0
1
Yes, I was using the wrong pins.
UPDATE 5: I got it to work, thanks to all of you!!! I tested the mother board out of the case and it works. Checked the case, found a stray support screw that was causing the short, removed it and booted windows 10. My cooler is revving every 10 seconds or so, I'll check that later. Haven't tried any OC yet, will update after I add the peripherals in a couple hours.
UPDATE 6: Didn't get to the OC part yet, the performance is shitty. To get rid of the fan revving I ran ipmitool sensor thresh FAN1 lower. That solved the fan isue but my CPU will now be even more sluggish and very rarely go over 400 MHz (its minimum frequency), even if I try to run benchmark. I reset the IPMI/BMC to factory defaults and the fan started revving again, but the performance is still bad (not always 0.4 GHz but almost all the time). Man, these things are extremely hard to get right. Any other ipmitool commands I can run to get it to go at full frequency before I start to OC?
How do I get the system to understand that FAN1 is the CPU fan?
 

lixinran0809

New Member
Nov 26, 2019
6
2
3
I am a little confused about the relationship between changing all-core frequency, CPU voltage, and lock frequency. The CPU I'm using is ZS1711 and the mobo is H11SSL. When I change the all-core frequency only, the VID seems to rise or fall with it as well. To make sure that I was reading the correct voltage, I used the PowIRCenter to directly monitor the VRM output voltage. For example, at default, the VID shown in PowIRCenter is ~0.9V, and when I set the frequency to 3.4Ghz, the VID will also change and become 1.2~1.25V automatically. This seems relatively high for a zen 2 chip, and I naturally thought I can change the voltage setting to lower it. But once I set the voltage to a lower level, such as 1.0V, the frequency also drops, in this case, it dropped to 3.0Ghz. At this point, I thought checking the "Lock Frequency" box should help, but once I apply lock frequency, the voltage adjustment no longer functions.

Is this the work of the SMU trying to maintain the set frequency? When I set the frequency to all core 4.0Ghz just to test and see, the VID was changed to over 1.4V!:eek: I briefly looked into the PowIRCenter and I couldn't find a way to add a negative offset to the output. @ExecutableFix Is this an intended behavior? If so, is there a way to lower the voltage to prevent fast degradation?
 

madbrad

New Member
Dec 10, 2020
15
0
1
I am a little confused about the relationship between changing all-core frequency, CPU voltage, and lock frequency. The CPU I'm using is ZS1711 and the mobo is H11SSL. When I change the all-core frequency only, the VID seems to rise or fall with it as well. To make sure that I was reading the correct voltage, I used the PowIRCenter to directly monitor the VRM output voltage. For example, at default, the VID shown in PowIRCenter is ~0.9V, and when I set the frequency to 3.4Ghz, the VID will also change and become 1.2~1.25V automatically. This seems relatively high for a zen 2 chip, and I naturally thought I can change the voltage setting to lower it. But once I set the voltage to a lower level, such as 1.0V, the frequency also drops, in this case, it dropped to 3.0Ghz. At this point, I thought checking the "Lock Frequency" box should help, but once I apply lock frequency, the voltage adjustment no longer functions.

Is this the work of the SMU trying to maintain the set frequency? When I set the frequency to all core 4.0Ghz just to test and see, the VID was changed to over 1.4V!:eek: I briefly looked into the PowIRCenter and I couldn't find a way to add a negative offset to the output. @ExecutableFix Is this an intended behavior? If so, is there a way to lower the voltage to prevent fast degradation?
In my case, using any of the presets in @ExecutableFix's awesome tool gets my voltage to 1.45-1.5V and it's very, very scary.
To top that, on my 32-core 2S1705E3VIVG5 only the High Multicore option doubles performance, the others halve it, and if I press the "Revert" button the frequency gets stuck at 0.4 GHz and I have to restart.
On top of that, sometimes the frequency gets stuck at 0.4 GHz for no reason at all and only restarting and setting full speed fan mode takes it out of its slumber.
I am also interested in a way to lower the voltage to prevent fast degradation, thanks for mentioning this @lixinran0809 And also an easier way to apply overclock when starting windows (like user presets or OC straight from BIOS?)
 
Last edited:

lixinran0809

New Member
Nov 26, 2019
6
2
3
In my case, using any of the presets in @ExecutableFix's awesome tool gets my voltage to 1.45-1.5V and it's very, very scary.
To top that, on my 32-core 2S1705E3VIVG5 only the High Multicore option doubles performance, the others halve it, and if I press the "Revert" button the frequency gets stuck at 0.4 GHz and I have to restart.
On top of that, sometimes the frequency gets stuck at 0.4 GHz for no reason at all and only restarting and setting full speed fan mode takes it out of its slumber.
I am also interested in a way to lower the voltage to prevent fast degradation, thanks for mentioning this @lixinran0809 And also an easier way to apply overclock when starting windows (like user presets or OC straight from BIOS?)
I asked the seller who sold me the ZS1711, he was experiencing similar issues with 2S1705 as well. Where the frequency is often only at 0.4Ghz. But I don't know much more about the situation. 1.5V on a 2S zen 2 chip sounds very scary indeed!
I thought of running a batch file at startup to open up the overclocking tool, but it would be too much effort to figure out a way to automatically apply the settings. I think someone already asked @ExecutableFix before to add this in the future update.
 

lixinran0809

New Member
Nov 26, 2019
6
2
3
I have some AIDS64 results that I find a bit strange. Here is a pile of them:

But these two in specific are interesting and represent a consistent trend I'm finding.

They are from the same bios settings and bootup, they are separated only by applying the Overclock tool.

I'll summarize:

All latencies go down with overclock... Great!
Ram bandwidth stays Flat (as expected)
L1 and L2 cache bandwidth go way up (proportional to clock speed?)
L3 bandwidth plummets!!
I'm finding this in many different configurations. For some reason applying an overclock cuts the L3 bandwidth significantly. Any thoughts as to why?

P.S. testing my CFD workloads has shown that 2400 MHz with IF synced is better than 3200/1600 mode (and uses cheaper RAM). Like the other member "Epyc" I am unable to get more than 2400 MHz to work on a "ZS" model.


View attachment 14397View attachment 14398
Have you been able to figure out what was causing the performance drop for the L3 cache when overclock was applied? I am running into the same exact situation right now.
 

Spartus

Active Member
Mar 28, 2012
323
121
43
Toronto, Canada
Have you been able to figure out what was causing the performance drop for the L3 cache when overclock was applied? I am running into the same exact situation right now.

No, ultimately I moved away from the synthetic tests after they helped me figure out the differences between the ZS and -04 models. I was able to get >90% of retail performance in my workload on the ZS with overclocks, and over 100% of retail performance on the -04. it was concerning, but ultimately did not hugely hinder my usecase, despite it being bandwidth intensive.

I did find i got better results at 2400 mhz with IF synced than at 3200/1600 Memory/IF async.
 
  • Like
Reactions: lixinran0809

madbrad

New Member
Dec 10, 2020
15
0
1
Guys, I am having huge thermal issues here on my 32-core, and I doublechecked the cooler and paste, it's just the huge voltage demands and the weird Supermicro board which starts shrieking like the Naz'gul in a couple minutes of full load at low clocks.
I am using the 2S 32-core CPU and a Noctua NH-U14s with Thermal Grizzly liquid metal paste which I thoroughly applied (twice, to be sure that wasn't the issue) The radiator is snug and tight.
If I clock it at 2.8 GHz the Supermicro H11SSL-i starts the thermal alarm in about 5 minutes of Cinebench and I have to stop it. HWinfo says the CPU and die are both under 90 degrees. The voltage caps at 1.495 V for this frequency, which is hideously high.
Also, the board doesn't seem to be able to control the fan by itself so I have to manually set the speed, which is double annoying and counterproductive. If a light baking task takes longer than a few minutes I have to cut the power supply to the PC.
Again, any tips for lowering the voltage would be amazing. Also, do any of AMD's overclocking tools work on this CPU? I'd love to see single-core boost like on the Threadrippers.
 

yesoos

Member
Mar 10, 2020
40
4
8
PL
1. Set FANs in IPMI to "heavy I/O" (speeds will stablilize and increase RPMs under load)
2. Where Do you read the voltages? Read it from IPMI sensors under P1_VDDCR / P2_VDDCR
3. You can control voltages - use ExecutableFix's app from first post / do not use presets / you need to figure out best settings / 2S CPUs are known to be worse at OC than ZS samples
 

madbrad

New Member
Dec 10, 2020
15
0
1
1. Set FANs in IPMI to "heavy I/O" (speeds will stablilize and increase RPMs under load)
2. Where Do you read the voltages? Read it from IPMI sensors under P1_VDDCR / P2_VDDCR
3. You can control voltages - use ExecutableFix's app from first post / do not use presets / you need to figure out best settings / 2S CPUs are known to be worse at OC than ZS samples
Thanks, yesoos. I put it on heavy I/O but I can't yet tell if it works, will report back later.
But I did set the voltages like you mentioned. My 2S CPU is a terrible overclocker indeed. It can reach 2.93GHz single core and 2.4 GHZ all cores full load at 1.35v and the performance is close to a 16-core threadripper, very slow, 20k multi core / 800 single core Cinebench 23.
Setting the voltage lower decreases the frequency. Setting it higher increases it. But anything higher and I am afraid of electron migration and frying it. With the EDT bug I can reach 3.0 on all cores at 1.5V but I get the result mentioned earlier (thermal alarm in a couple mins). Also 1.5V yikes.
I guess this is what I get with this CPU. Does any1 know what the difference is between ZS and 2S? What does the naming scheme mean? Which are the best chips for overclocking?
 

yesoos

Member
Mar 10, 2020
40
4
8
PL
AMD Ryzen Decoder:

For 64 core model I think this is the best ES (2GHz stock and overclockable): 100-000000053-04_32/20_N
Personally I tested those, OC is very similar on 1 socket , for dual sockets I have a lot of trouble with those:
ZS1406E2VJUG5_22/14_N
ZS1407E2VJUG6_22/14_N
 
  • Like
Reactions: Sparky'sAdventure

madbrad

New Member
Dec 10, 2020
15
0
1
AMD Ryzen Decoder:

For 64 core model I think this is the best ES (2GHz stock and overclockable): 100-000000053-04_32/20_N
Personally I tested those, OC is very similar on 1 socket , for dual sockets I have a lot of trouble with those:
ZS1406E2VJUG5_22/14_N
ZS1407E2VJUG6_22/14_N
Setting fan mode to Heavy I/O didn't work. Everything blows at full speed 2 hours after the setup even if my cpu temp reads 37 C in IPMO
What kind of single core/multi core frequencies did you see on 100-000000053-04_32/20_N , and what where you using to cool them?
 
Last edited:

alex_stief

Well-Known Member
May 31, 2016
884
312
63
38
Liquid metal is not necessarily the best option for the interface between heatspreader and coldplate.
More often than not, you end up without proper contact, rendering it way worse than regular thermal paste. Even when ignoring the potential long-term and short-term risks, I would not recommend it for this application.
 

yesoos

Member
Mar 10, 2020
40
4
8
PL
As far I know
VDDCR - it's Vcore cpu voltage
SOCDUAL/SOCRUN - are related to other System-on-Chip logics voltages (as northbridge is integrated)
 

irgen

Member
Jan 14, 2021
33
5
8
First to thank you ExecutableFix for this beautiful software )
Guys, anyone have Asrock EPYCD8 running with ES? I plan to buy one however not sure what BIOS i should use.
I see first they added support for AMD EPYC 7002 series processor in 2.10. But what AGESA version in that 2.10? Is is ok like <1.0.0.3 or newer.
Would much appreciate if someone clarify.
 

handyserv

New Member
Jan 18, 2021
9
1
3
Hello, thank you all for the tools and info in this thread. Im running 2x ZS1711E3VIVG5 on a H11dsi with modded bios (from OP) and currently NO overclock. According to ipmi health monitoring P1_SOCRUN is at 1.174 volts "upper critical". Others in this thread have asked if this is something to be concerned about but got no responses. I would like to start overclocking but Im not sure if current P1_SOCRUN voltage reading is acceptable and if theres anything I can do to lower it...? Thanks again, this forum is awesome!
 

maxermaxer

Active Member
Oct 28, 2016
289
48
28
48
Hello, thank you all for the tools and info in this thread. Im running 2x ZS1711E3VIVG5 on a H11dsi with modded bios (from OP) and currently NO overclock. According to ipmi health monitoring P1_SOCRUN is at 1.174 volts "upper critical". Others in this thread have asked if this is something to be concerned about but got no responses. I would like to start overclocking but Im not sure if current P1_SOCRUN voltage reading is acceptable and if theres anything I can do to lower it...? Thanks again, this forum is awesome!
Hi Handyserv, I am also considering the 2ZS series EPYC but I am concerned if SM motherboard have overheating issue. When you are not overlocking how do the temperature look like? Especially the temp of VRM. Have you tried making the CPUs run 100% for a period time and do you hear the alarm? 2ZS is a lot cheaper than -04 series. I am doing 3D production rendering work so OC is not something I am keen to do.
 

handyserv

New Member
Jan 18, 2021
9
1
3
Hi Handyserv, I am also considering the 2ZS series EPYC but I am concerned if SM motherboard have overheating issue. When you are not overlocking how do the temperature look like? Especially the temp of VRM. Have you tried making the CPUs run 100% for a period time and do you hear the alarm? 2ZS is a lot cheaper than -04 series. I am doing 3D production rendering work so OC is not something I am keen to do.
Hello, running xmrig not overclocked with all threads at 1.6 ghz (cpu 100%). Ive run this for a few days non stop without any alarms or overheat. Only issue is the P1_SOCRUN voltage.

Edit to add screencap from zenmonitior
 

Attachments

Last edited:
  • Like
Reactions: maxermaxer