Overclock your AMD Epyc

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

RageBone

Active Member
Jul 11, 2017
617
159
43
@nero243 the "Gamer commands" are able to set Loadline values, but that is the IR BS Feature not all controllers have etc.

@Quade if you are a gambler, good luck. Have heard the seller is pretty good with hardware.
And those cpus don't look good at all.
 
  • Like
Reactions: nero243

nero243

Active Member
Oct 28, 2018
108
78
28
Holy cow, I just stumbled upon this thread, very interesting.
There are SM boards up for grabs locally for like 400 bucks new bulk packaging, but finding a CPU is an absolute headache.

Seeing this japanese dude has been putting batches of questionable SKUs for sale, compatibility is dodgy, and these might have been fried already by seeing it doesn't boot on their testing SM board.
Sell As-Is! AMD EPYC 7551 ES 2S1451A4VIHE4 1.44GHz 32C Socket SP3 14nm | eBay
Sell As-Is! AMD EPYC 7551 ES 2S1905A4VIHF4 1.9GHz 32C Socket SP3 | eBay

I wonder has anybody tried their luck with those?
This seller is actually really nice, i ordered a few xeon ES from him and also my first epyc. But he stepped up his epyc prices after finding this thread.

I'd say if he can't get a CPU to work you're probably out of luck without some knowledge. If a CPU will post to bios but fail to boot windows i'm pretty sure it's a agesa or mc issue. Like i wrote in the start post, i tested different Threadripper microcodes and agesa patches on my ES and it would always post to bios (even without any microcode) but would fail to boot windows or randomly bluescreen.

@nero243 the "Gamer commands" are able to set Loadline values, but that is the IR BS Feature not all controllers have etc.

@Quade if you are a gambler, good luck. Have heard the seller is pretty good with hardware.
And those cpus don't look good at all.
I haven't played around with the USB005 for a long time, can't comment on this. You probably have more experience with this stuff than me by now.

I just bought a dead vega 56 and brought it back to life only to test my usb005 on it lol.



Has anyone been able to OC a Rome cpu yet?
Stumbled across this yesterday, so Wendell from level1techs probably found the secret sauce MSR. They cover OC at 23:50

 

Quade

New Member
Nov 13, 2019
6
0
1
@nero243, very much appreciated for your educated input, you are MVP ;)
What is your opinion on the 1S1601A4VIHF4 chip? Is it worth grabbing for attempts at getting the chip pushed up to 3.6 all cores or close on an H11SSL-i v1 board? Judging by the thermals, I expect it to have an equal TDP of the 2990WX, more or less depending on binning. Of course it would take some effort to chill the board's VRM down so it won't be toasted in a long run, to some extent at least.


Has anyone booted up eight 2Rx4 2133 16gb modules in octa-channel?
 
Last edited:

MrCake117

Member
Feb 28, 2019
32
24
8
25
Japan
@nero243, very much appreciated for your educated input, you are MVP ;)
What is your opinion on the 1S1601A4VIHF4 chip? Is it worth grabbing for attempts at getting the chip pushed up to 3.6 all cores or close on an H11SSL-i v1 board? Judging by the thermals, I expect it to have an equal TDP of the 2990WX, more or less depending on binning. Of course it would take some effort to chill the board's VRM down so it won't be toasted in a long run, to some extent at least.


Has anyone booted up eight 2Rx4 2133 16gb modules in octa-channel?
I'm running 10 2Rx4 2133 16gb modules and 6 2Rx4 2133 8gb modules on a H11dsi board
 
  • Like
Reactions: Quade

epycforlife

New Member
Sep 21, 2019
3
0
1
I don't think so. There is no MSR documentation and no one seems to have unlocked the CBS menu yet. Maybe we'll have to wait for the new threadrippers, hope that these use the same MSR and some insider makes a tool for it. Maybe someone can monitor the MSR with RW-everything on a ryzen 3000 platform while ocing with the ryzenmaster software. I think the MSR has changed for the new generation Zen because they introduced the CCX overclocking options, which is substantially diffrent from Zen and Zen+.



No, there are no LLC options in the bios and i don't remember to have found them in the IR controller software.

Thanks for clarifying that, hopefully it gets figured out. Its hard to find info on overclocking epyc processors, so it wasn't clear.
 

Quade

New Member
Nov 13, 2019
6
0
1
On a side note, I'm very much positive a voltmod or pencil mod (does not void the warranty on your board) is possible, in fact there is always a functioning resistor placed down somewhere on the VRM PWM chockablock. From my experience with Intel X48 and X58 boards, one resistor can affect both Vdroop and Vdrop simultaneously, controller dependent in most of the use cases. I highly discourage you people doing that on clocked systems beyond 1.3v, as we have yet to find out whether the VRM subsystem on the H11 board is efficient enough to react to duty cycles faster than massive, silicon degrading overshoot takes place. All of the Threadripper boards seem to have beefy and quite responsive VRMs, good enough to probe CPU changes of its power states right before detrimental voltage spikes happen. Nonetheless, LLC could be very beneficial with fixed power states, such as running one power state regardless of the workload, as there virtually no need to adjust for overshoot. Example: 3ghz at 1.2v, fixed, no power save features enabled.

Moderate changes to LLC are fine, in fact we can shove off 50mV of Vdroop easily at lower Vcore without having to worry about catastrophic voltage spikes. This will ensure stability, in contrary to increasing Vcore even further to combat Vdroop, which is, in turn, silly and would degrade the chip even faster than intermediary overshooting.

Some glossary: https://en.wikichip.org/wiki/load-line_calibration

It all comes down to one important question: is the VRM's quality up to the task, and whether we can find the appropriate resistor(s) on the board or not.

@MrCake117 I've read that (possibly older firmware) couldn't have booted up with 2133 sticks occupying either of the slots, you had to put in one 2400 on each die to fool it into thinking it's the correct memory, and then let the IMC drop down freqs of all sticks to the most underperforming module. Is it still the case?
 

MrCake117

Member
Feb 28, 2019
32
24
8
25
Japan
On a side note, I'm very much positive a voltmod or pencil mod (does not void the warranty on your board) is possible, in fact there is always a functioning resistor placed down somewhere on the VRM PWM chockablock. From my experience with Intel X48 and X58 boards, one resistor can affect both Vdroop and Vdrop simultaneously, controller dependent in most of the use cases. I highly discourage you people doing that on clocked systems beyond 1.3v, as we have yet to find out whether the VRM subsystem on the H11 board is efficient enough to react to duty cycles faster than massive, silicon degrading overshoot takes place. All of the Threadripper boards seem to have beefy and quite responsive VRMs, good enough to probe CPU changes of its power states right before detrimental voltage spikes happen. Nonetheless, LLC could be very beneficial with fixed power states, such as running one power state regardless of the workload, as there virtually no need to adjust for overshoot. Example: 3ghz at 1.2v, fixed, no power save features enabled.

Moderate changes to LLC are fine, in fact we can shove off 50mV of Vdroop easily at lower Vcore without having to worry about catastrophic voltage spikes. This will ensure stability, in contrary to increasing Vcore even further to combat Vdroop, which is, in turn, silly and would degrade the chip even faster than intermediary overshooting.

Some glossary: https://en.wikichip.org/wiki/load-line_calibration

It all comes down to one important question: is the VRM's quality up to the task, and whether we can find the appropriate resistor(s) on the board or not.

@MrCake117 I've read that (possibly older firmware) couldn't have booted up with 2133 sticks occupying either of the slots, you had to put in one 2400 on each die to fool it into thinking it's the correct memory, and then let the IMC drop down freqs of all sticks to the most underperforming module. Is it still the case?
I've encounter no problem, all of my ram are running 2133mhz on bios 1.0c and BMC firmware 1.46

Also i would like to know if there's a possibility to run Epyc Rome on any H11 rev 1.01 motherboard, I called supermicro they told me to buy a new rev 2.0 mb...
 
  • Like
Reactions: Quade

nero243

Active Member
Oct 28, 2018
108
78
28
On a side note, I'm very much positive a voltmod or pencil mod (does not void the warranty on your board) is possible, in fact there is always a functioning resistor placed down somewhere on the VRM PWM chockablock. From my experience with Intel X48 and X58 boards, one resistor can affect both Vdroop and Vdrop simultaneously, controller dependent in most of the use cases. I highly discourage you people doing that on clocked systems beyond 1.3v, as we have yet to find out whether the VRM subsystem on the H11 board is efficient enough to react to duty cycles faster than massive, silicon degrading overshoot takes place. All of the Threadripper boards seem to have beefy and quite responsive VRMs, good enough to probe CPU changes of its power states right before detrimental voltage spikes happen. Nonetheless, LLC could be very beneficial with fixed power states, such as running one power state regardless of the workload, as there virtually no need to adjust for overshoot. Example: 3ghz at 1.2v, fixed, no power save features enabled.

Moderate changes to LLC are fine, in fact we can shove off 50mV of Vdroop easily at lower Vcore without having to worry about catastrophic voltage spikes. This will ensure stability, in contrary to increasing Vcore even further to combat Vdroop, which is, in turn, silly and would degrade the chip even faster than intermediary overshooting.

Some glossary: https://en.wikichip.org/wiki/load-line_calibration

It all comes down to one important question: is the VRM's quality up to the task, and whether we can find the appropriate resistor(s) on the board or not.

@MrCake117 I've read that (possibly older firmware) couldn't have booted up with 2133 sticks occupying either of the slots, you had to put in one 2400 on each die to fool it into thinking it's the correct memory, and then let the IMC drop down freqs of all sticks to the most underperforming module. Is it still the case?
I can only partially follow you, but before you start figuring out a voltmod maybe check the limits as is beforehand ;)

And no not all x399 boards have a beefy powerstage, it's mostly standard 8 Phase 60A. All Sp3 Boards use 6 Phase 70A.
I've encounter no problem, all of my ram are running 2133mhz on bios 1.0c and BMC firmware 1.46

Also i would like to know if there's a possibility to run Epyc Rome on any H11 rev 1.01 motherboard, I called supermicro they told me to buy a new rev 2.0 mb...
I sersiously think the only change on every SP3 Board for Rome is just a 32mb bios which requires a new eeprom chip. Supermicro probably just put a bigger chip and a few stickers on the rev. 1 boards. Asrack just straight up writes "32mb bios chip required" in their bios download section.
 
  • Like
Reactions: Quade and RageBone

Quade

New Member
Nov 13, 2019
6
0
1
Time to stock up on 32mb EEPROMs, then :D Fairly possible to tear the UEFI firmware apart and shuffle CPU IDs, and flash back with a programmator. Although, I don't think anyone's going to have a second thought of replacing their 64C Epyc back to the older counterparts.
Not unless there were some discreet modifications to the board, just like vendors did with rev1 to rev2 X99 boards having better RDIMM RAM support and what not.
Either way, it would require a surgical alteration. I don't think people feel comfortable enough to shell out another $400 on, essentially, the same board.
 

nero243

Active Member
Oct 28, 2018
108
78
28
Time to stock up on 32mb EEPROMs, then :D Fairly possible to tear the UEFI firmware apart and shuffle CPU IDs, and flash back with a programmator. Although, I don't think anyone's going to have a second thought of replacing their 64C Epyc back to the older counterparts.
Not unless there were some discreet modifications to the board, just like vendors did with rev1 to rev2 X99 boards having better RDIMM RAM support and what not.
Either way, it would require a surgical alteration. I don't think people feel comfortable enough to shell out another $400 on, essentially, the same board.
I already checked the Bios, there is alot of new stuff in there. The microcode oddly seems to be the same, at least i couldn't find a new one right away. Modding a bios for a 16mb chip isn't worth the huge amount of work in my opinion, also i don't know anyone who would be able to pull this of. Board manufacturers should be able to release a 16mb bios for Rome which would break compatability with Napels, i don't know why no one is doing this, or maybe i missed something.

Bios modding for intel was super easy compared to what i experienced with my bios. You have to do alot by hand/hex edit since UBU sometimes corrupts the rom.

I think gigabyte boards have a socketed bios chip, maybe someone can try it there first with 0 risk involved.

Edit.:

So i got back at at trying to unlock the CBS menu. I'm still working on supermicro, but i think i unlocked the custom core P-states and memory OC for the Gigabyte MZ31-AR0.

Can't guarantee it'll work or even post, so testers are welcome.
MZ31-AR0-F13-CBS
 
Last edited:

NORM123

New Member
Nov 19, 2019
16
0
1
Hi, when I use dual epyc7551 es with 3.5GHz, I only got 7000score in benchmark r20, It looks much lower than one epyc7551, is this special situation?
 

alex_stief

Well-Known Member
May 31, 2016
884
312
63
38
My two retail 7551@stock get 15381 in CB R20. Either there is something wrong with your engineering samples, or the motherboard can not handle the 3.5GHz clock speed. Maybe run at a more reasonable CPU frequency to check?
 

nero243

Active Member
Oct 28, 2018
108
78
28
Hi, when I use dual epyc7551 es with 3.5GHz, I only got 7000score in benchmark r20, It looks much lower than one epyc7551, is this special situation?
Are you using octa-channel on both CPUs? For me the performance was really unreliable with a 1 DIMM per Die setup. Maybe also try playing around with coreprio, this tool could get rid of a few performance issues under windows for me.
 

NORM123

New Member
Nov 19, 2019
16
0
1
Are you using octa-channel on both CPUs? For me the performance was really unreliable with a 1 DIMM per Die setup. Maybe also try playing around with coreprio, this tool could get rid of a few performance issues under windows for me.
I use the 1*32 DIMM for the test. so the RAM maybe one reason,but dual 7551 just got 7400 in R20. And other prblem is i cant use my 2080ti in this computer, when i boots with 2080ti my monitor turn on, but no signal, but use BMC and VGA eveything is fine. is there same problem ?
 

Attachments

NORM123

New Member
Nov 19, 2019
16
0
1
My two retail 7551@stock get 15381 in CB R20. Either there is something wrong with your engineering samples, or the motherboard can not handle the 3.5GHz clock speed. Maybe run at a more reasonable CPU frequency to check?
thx i will try lower frequeency
 

NORM123

New Member
Nov 19, 2019
16
0
1
My two retail 7551@stock get 15381 in CB R20. Either there is something wrong with your engineering samples, or the motherboard can not handle the 3.5GHz clock speed. Maybe run at a more reasonable CPU frequency to check?
upload_2019-11-19_19-53-56.pngafter i try lower frequency @3.0Ghz still got lower score
 

MrCake117

Member
Feb 28, 2019
32
24
8
25
Japan
I use the 1*32 DIMM for the test. so the RAM maybe one reason,but dual 7551 just got 7400 in R20. And other prblem is i cant use my 2080ti in this computer, when i boots with 2080ti my monitor turn on, but no signal, but use BMC and VGA eveything is fine. is there same problem ?
Try to at least run with 1 dimm per die and for your gpu you need to update the Bios, unfortunately the latest bios aren't stable for what I experienced.
If you disable cpu core boost performance with the latest bios you won't be able to post.
 

am45931472

Member
Feb 26, 2019
87
17
8
I've seen alot of good benchmarks on here of overclock performance however havnt seen any for single core performance. Im looking for a workstation CPU with good clock speed, but would also like RDIMMs. so what kind of single core performance an an AMD epyc give
 

NORM123

New Member
Nov 19, 2019
16
0
1
Try to at least run with 1 dimm per die and for your gpu you need to update the Bios, unfortunately the latest bios aren't stable for what I experienced.
If you disable cpu core boost performance with the latest bios you won't be able to post.
THX,when I got the h11dsi, it was bios2.0( produced 2019/9/17), so I need to downgrade my BIOS? or enable some option in BIOS?
 

MrCake117

Member
Feb 28, 2019
32
24
8
25
Japan
THX,when I got the h11dsi, it was bios2.0( produced 2019/9/17), so I need to downgrade my BIOS? or enable some option in BIOS?
Well it seems that you're already running with the latest bios... did you set the VGA priority option to "Offboard" in the pcie submenu ?
 
Last edited: