Finally: Overclocking EPYC Rome ES

vavavava

New Member
Jan 13, 2020
9
1
3
Oh yes that's right. I forgot to mention that there are a couple of bios settings that need to be set on the DSI for the ZS cpu's, it's finicky. So turn on IOMMU and disable Above 4G decoding and it should work
I have a question, I've got DSI motherboard and IOMMU and Above 4G decoding is on. It works. What are the downsides of keeping Above 4G active?
 

levelud

New Member
Oct 16, 2020
8
0
1
You don't actually want to start with max multicore performance, because then your single thread performance drops to something like 1GHz. My ZS1406 boots at a normal 1.4-2.2 GHz, check your system configuration.
I have tested those 3 Settings under Windows 10, the only one working for me is Max Multicore Performance Mode, there I will get about 11'000 in CB20 and all cores at 3450 MHz even in Single Thread, with Max Single Core I stuck about 4500 and Single Thread is about the same.
 

bayleyw

Member
Jan 8, 2014
84
21
8
Does it perform like 3450 MHz? The SMU bug that max multicore uses doesn't change the displayed clock speed.
 

levelud

New Member
Oct 16, 2020
8
0
1
Does it perform like 3450 MHz? The SMU bug that max multicore uses doesn't change the displayed clock speed.
I would say it does, the whole operating system feels smoother and even all the tests I have made do look better.
 

bayleyw

Member
Jan 8, 2014
84
21
8
I would say it does, the whole operating system feels smoother and even all the tests I have made do look better.
Any easy way to figure out is to run the CPU-Z single-threaded benchmark, its fairly known what a 3450 MHz Zen2 will score and I've found it useful in the past for debugging gross errors in turbo configuration and the like (for example, it revealed my Cascade Lake + Xen box was not turbo'ing at all).
 

mirrormax

Member
Apr 10, 2020
73
30
18
BUT there is a catch: setting the voltage over 1.05v or setting the frequency > 3800MHz in the software could actually KILL the CPU instantly (this only applies to models > 32-cores, it’s different on lower core models). I’ve done it twice :(. You’ll know when it goes bad when you get a black screen and can’t power off the system with the On/Off button. So please be careful when using this “bug”, but it should be fine as long as you don’t go over the limits mentioned above.
if i apply high multi first, then go manually change clocks to 3.2 does that undo the edc "bug"?
because i can go up to 1.16v that way, but after that i think it hits the OCP and it shuts down if i load it, i assume i could go over if i could disable ocp. (i forgot this notice about not going over 1.05 with edc while playing around luckily havnt killed anything, yet :p)
 

levelud

New Member
Oct 16, 2020
8
0
1
Any easy way to figure out is to run the CPU-Z single-threaded benchmark, its fairly known what a 3450 MHz Zen2 will score and I've found it useful in the past for debugging gross errors in turbo configuration and the like (for example, it revealed my Cascade Lake + Xen box was not turbo'ing at all).
You're right in Single Core it isn't as fast as I thought.AMD 2.PNG
 

levelud

New Member
Oct 16, 2020
8
0
1
But is there a way to get past 4.00 GHz for Single Task Scenarios? My CPU does 4.00 GHz with ease.

EPYC 4000 MHz.PNG
 

bayleyw

Member
Jan 8, 2014
84
21
8
You can see there that the CPU is beginning to throttle under load - the multithread ratio should be slightly higher than the number of cores because of SMT. Keep an eye on Vcore (SVI TFN2 in HWiNFO64) - I imagine at 4GHz it gets dangerously high, enough so that you will see degradation if you load the chip too much or for too long. The SMU usually works some magic to not kill your CPU, but I don't know how much of that is functional on these pre-release parts or if it still works on OC'ed chips.
 

mirrormax

Member
Apr 10, 2020
73
30
18
You can see there that the CPU is beginning to throttle under load - the multithread ratio should be slightly higher than the number of cores because of SMT. Keep an eye on Vcore (SVI TFN2 in HWiNFO64) - I imagine at 4GHz it gets dangerously high, enough so that you will see degradation if you load the chip too much or for too long. The SMU usually works some magic to not kill your CPU, but I don't know how much of that is functional on these pre-release parts or if it still works on OC'ed chips.
Assume he didn't run the multicore test at 4ghz.
 

levelud

New Member
Oct 16, 2020
8
0
1
Assume he didn't run the multicore test at 4ghz.
Yes, I just bumped the multi to get 4GHz for single threaded tasks. I will have a look at the voltage later. But It's under water so, it should work like a Threadripper I think, until you reach max multi for the CPU and then crash, one step back and you should be good to go.
 

yesoos

Member
Mar 10, 2020
38
3
8
PL
From my observations if I used dual ranked memories (2Rx4) for RandomX there is gain for single CPU (zs1406 at 3GHz with PBO hack, 200W from wall) , 4 channels - from 37k (if I used 1Rx4 , for 1Rx8 even lower than 37k) ->41k
 

mirrormax

Member
Apr 10, 2020
73
30
18
i seem to hit some sort of limit at 255a on the cpu and 280w(monitored with elmor evc), any ideas if its the cpu or motherboard(mb is rated up to 280w cpus so iam guessing theres some limit there)? i can raise OCP limit to 60a per chipwith evc (mb has 6x60a mosfets) but if i try raising voltage over 1.16 in it instantly turns off when putting on load. temps are no issue lots of headroom.
 

Set Iron

New Member
Mar 18, 2020
7
0
1
Hey everyone, first post here. I tried going through the thread but didn't find any specific answers.

Could someone point me to a version/stepping reference for epyc processors? Specifically, QS (stepping 0?) Epyc on ebay sometimes claim a lower frequency than retail, sometimes not. Do they turbo up to acceptable frequencies without tinkering? 64 cores for $2000 sound appealing but I'd rather not rely on software tools to set the frequency.

I have a SM server with an H11DSU-IN rev1, currently running a pair of 7371s. Are there modded bioses with Rome support going around for this board? I've only seen the ones from OP for standard ATX systems.

Thanks!