VRM modify ICC_MAX to run high TDC OEM cpu

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

RolloZ170

Well-Known Member
Apr 24, 2016
7,172
2,238
113
I would like to know if the T640 has multiple VRM chips that need to be modified.
After I modified the VRM, the system is able to boot with one CPU, but when I install the second 8259CL, the device fails to boot.
two CPU, two VRM controller, one for each CPU.
 
  • Like
Reactions: Haruna

GraFfiX420

New Member
Jan 11, 2025
3
1
3
Hello, I acquired a pair of Xeon platinum 8175M, and decided to replace my Xeon silvers that I currently had in my server. I run an HP ML350 gen10, and I use it as a desktop. I use a GTX1650 as a graphics card in the machine.

Yesterday, the processors arrived, I followed the instructions contained in this thread, and I was able to unlock TDP. This is where things got odd. My graphics card was no longer being detected. The GTX1650 doesn't draw any power from the supply independently, it draws it's power from the PCI slot. I moved the graphics card down one slot, and it works again. I ran cinebench, it ran fine the first time, the second time, I got a blue screen. After reboot, my principle boot drive wasn't being detected (1tb ssd). I moved the hard drive to another slot, and it works fine now. This morning, I tried to plug in an external USB drive and copy a file, the copy fails after 4mb or so, whether I'm using a powered usb device or one that's powered by the port itself. I have had zero problems with this server until now, and I've been running it for a little over two years.

The server has a pair of 800w power supplies, set to balanced mode. So I have a couple questions.
1. Is this the problem, am I undervolting the system?
2. If that's the case, I'll order a set of 1600W power supplies, am I safe to run my system until then, or should I swap back to my silvers?
3. Has anyone else experienced anything like this after unlocking TDP?
4. If needed, can I revert the changes that were made by the TDP unlock?
 

RolloZ170

Well-Known Member
Apr 24, 2016
7,172
2,238
113
Hello, I acquired a pair of Xeon platinum 8175M, and decided to replace my Xeon silvers that I currently had in my server. I run an HP ML350 gen10, and I use it as a desktop. I use a GTX1650 as a graphics card in the machine.
Yesterday, the processors arrived, I followed the instructions contained in this thread, and I was able to unlock TDP. This is where things got odd
we do not unlock TDP.
make sure your 2x 8175M are ok. cheap used ones are worn out.
processors properly installed ?
1. Is this the problem, am I undervolting the system?
doubt that. TDP is 240W but 8175M they don't draw that much.
2. If that's the case, I'll order a set of 1600W power supplies, am I safe to run my system until then, or should I swap back to my silvers?
HP uses a propriätäry PSU system, with high TDP Prozessors you may forced to upgrade PSU, i don't know, check what you need for 2x platinum 8280.
3. Has anyone else experienced anything like this after unlocking TDP?
me not.
but it sound real ylike bad socket install.
4. If needed, can I revert the changes that were made by the TDP unlock?
yes with special tool. i have it not implemented because you can programm the VRM memory of PXE1610C only 14 times(OPT2)
 
Last edited:

GraFfiX420

New Member
Jan 11, 2025
3
1
3
we do not unlock TDP.
make sure your 2x 8175M are ok. cheap used ones are worn out.
processors properly installed ?

doubt that. TDP is 240W but 8175M they don't draw that much.

HP uses a propriätäry PSU system, with high TDP Prozessors you may forced to upgrade PSU, i don't know, check what you need for 2x platinum 8280.

me not.
but it sound real ylike bad socket install.

yes with special tool. i have it not implemented because you can programm the VRM memory of PXE1610C only 14 times(OPT2)
You're correct, I see some ML350 gen10 with 8280's wiith 500w PSU's, so that's not the issue. These are apparently used processors, although they were not supposed to be, there's some visible etching on the top of the processor from where the thermal compound/heatsink made contact. I did pull the processors and reseat them in the socket, however, the issues persist. I'm in contact with the seller, for now, I'm switching back to my Xeon Silvers. I'll post back the results. I do have a microscope if I need it, but the pins look fine to me.
 

RolloZ170

Well-Known Member
Apr 24, 2016
7,172
2,238
113
I did pull the processors and reseat them in the socket, however, the issues persist. I'm in contact with the seller, for now, I'm switching back to my Xeon Silvers. I'll post back the results. I do have a microscope if I need it, but the pins look fine to me.
use a torque screwsdriver and fasten all screws to 1.2-1.3 Nm('click') the leaf springs(screw 3&4) must touch the coolers surface.
 

GraFfiX420

New Member
Jan 11, 2025
3
1
3
use a torque screwsdriver and fasten all screws to 1.2-1.3 Nm('click') the leaf springs(screw 3&4) must touch the coolers surface.
Thank you for that. You were correct, one of, or both of the CPU's are faulty. I reinstalled my xeon silver's, and all is back to normal.
 
  • Like
Reactions: RolloZ170

RavenReign

New Member
Nov 23, 2024
3
1
3
Thank you for that. You were correct, one of, or both of the CPU's are faulty. I reinstalled my xeon silver's, and all is back to normal.
I was about to post, try just one of those cpus, if it works well, look at its stepping and try the other one alone as well and check that it has the same stepping
 

stefan1337

New Member
Sep 5, 2024
16
3
3
only MP2955A used in
X11SPM-T(P/F)
X11DPi-N(T) rev.2.x and up

and others
you need I2C interface MPS EVKT-USBI2C-02:
@RolloZ170 : thank you for making this thread, it's very valuable and interesting information!

however, I wonder, as someone having no any experience with VRM-(Re)-Programming, why External hardware I2C interface is needed? I mean the VRM is already connected to the Motherboard PMBus, right? so, why do not "talk" to it via the PMBus and (re-)program it that way or it is not allowed when the system is running (i.e. in such case the VRM is in read-only mode)?

there is already tool that can talk directly to MP2955 on Supermicro X11-series motherboards via the PMBus:


which would be trivial to modify to write to MP2955 of course if that is allowed when the system is running and the chip is not read-only in such case. thanks!

P.S. I do not own any of those Supermicro X11-series motherboards I am just wondering in pure theory what is the exact case, i.e. can PMBus be used for VRM-(Re)-Programming (i.e. the VRM is not in read-only mode when the system is running) or that can be done only via external hardware I2C.
 

RolloZ170

Well-Known Member
Apr 24, 2016
7,172
2,238
113
why External hardware I2C interface is needed?
more easy. faster, safer. you need to gain access to the bus for multiple bytes.
if any other dev, on the bus talks with you, trouble is on.
I mean the VRM is already connected to the Motherboard PMBus, right?
to one of them. mostly they are connected to BMC and or PCH.
so, why do not "talk" to it via the PMBus and (re-)program it that way or it is not allowed when the system is running (i.e. in such case the VRM is in read-only mode)?
need unique code for every board. in some systems lately access to VRM was blocked to protect against undervolting hacks.
there is already tool that can talk directly to MP2955 on Supermicro X11-series motherboards via the PMBus:

which would be trivial to modify to write to MP2955 of course if that is allowed when the system is running and the chip is not read-only in such case. thanks!
ok. anybody can fry you CPU. interesting.
normaly there are multiple I2C bus. i could'nt get any VRM controller over the PMBus for the PSU e.g.
 

stefan1337

New Member
Sep 5, 2024
16
3
3
ok. anybody can fry you CPU. interesting.
normaly there are multiple I2C bus. i could'nt get any VRM controller over the PMBus for the PSU e.g.
thanks for your answers, I now found the paper to that code:


what was tried:
Supermicro X11SSL-CF - Vulnerable
• Supermicro X11SPG-TF and X11SSE-F
• VRM reachable with default config, undervolting crashed the server
so, at least on the above 3 Supermicro X11-series motherboard, it seems VRM-Re-Programming is possible via the SMBus and while the system is running....
 

stefan1337

New Member
Sep 5, 2024
16
3
3
i don't like to run foreign code target machines. if you like, have fun.
as I mentioned, I don't own any of those Supermicro X11-motherboard, I was just wondering in theory if the VRM can be reprogrammed via the PMBus. now that is confirmed by that paper for Supermicro (at least as far as I understand it) I am finding that mind-staggering, it's really insane to have such "security hole", that can even blow permanently the CPU if the voltage is increased. I doubt other manufacturers made such mistake and as you said
RolloZ170 said:
in some systems lately access to VRM was blocked to protect against undervolting hacks.
- in fact that is the main thing answering my initial question, i.e. why external I2C interface is needed.
 

RolloZ170

Well-Known Member
Apr 24, 2016
7,172
2,238
113
- in fact that is the main thing answering my initial question, i.e. why external I2C interface is needed.
i thought about in the beginning.
but a single universal solution with I2C dongle and a cli tool looked better for me. some can be programmed without CPU, so yu don't need suported CPU first to program.

- in fact that is the main thing answering my initial question, i.e. why external I2C interface is needed.
it is to slow and byte by byte. if you change Page and want to write a value, another may change the Page to another and your value goes to the wrong place.
 

xp58

New Member
Feb 8, 2025
2
0
1
Hi, it's nice to hear that the 8259CL is working on the R640. I am currently trying to modify the DELL R740. I've applied two commands (60 and 62) on P4. After that, only CPU1 is working fine. However, the system still fails to boot (DS1, 2, 6, 7 blinking).
Where is the other chip that I can modify?