ES Xeon Discussion

Rand__

Well-Known Member
Mar 6, 2014
6,081
1,458
113
Its always dependent on use case - ie. are those interactive VMs (where you interact a lot with the service/tool running on it) or only background processing? If interactive then i fiound having a higher clock (all core turbo usually, not max boost) to be hugely important for the responiveness of the system.
 

thebluehue

New Member
Oct 7, 2021
3
0
1
I know I don’t run any actual heavy workload but other than people who are practicing some data silence or some AI/ML I am not convinced anybody would need beyond 16cpu, memory maybe more so.
Probably I don’t care as much as some hanging out here because I get to play with the big stuff at work during the day.

Either way probably pick the cheapest or toss a coin. Until 16-core busy I would assume the Intel wins, anything beyond that the AMD will no doubt win.
(Not I didn’t have any other advice since it seems you don’t specifically need lathe cache or AVX512 or anything really interesting)
Thank you, that makes a lot of sense.
 

AveryFreeman

consummate homelabber
Mar 17, 2017
320
38
28
40
Near Seattle
averyfreeman.com
I just want to report that I've been running a single QS E5-2650 v4 on an X10SRL-F for the last 2-3 years without issue, I bought it from the eBay sellers that work out of Japan.

The only thing that's weird about it is the CPUID reports 0000. It has PSODed a couple times, but not any more than a normal CPU AFAICT.

Code:
[root@robotboy:~] vim-cmd hostsvc/hostsummary
(vim.host.Summary) {
   host = 'vim.HostSystem:ha-host',
   hardware = (vim.host.Summary.HardwareSummary) {
      vendor = "Supermicro",
      model = "Super Server",
      uuid = "00000000-0000-0000-0000-ac1f6b48ea2a",
      otherIdentifyingInfo = (vim.host.SystemIdentificationInfo) [
         (vim.host.SystemIdentificationInfo) {
            identifierValue = "Default string",
            identifierType = (vim.ElementDescription) {
               label = "Asset Tag",
               summary = "Asset tag of the system",
               key = "AssetTag"
            }
         },
         (vim.host.SystemIdentificationInfo) {
            identifierValue = "0123456789",
            identifierType = (vim.ElementDescription) {
               label = "Service tag",
               summary = "Service tag of the system",
               key = "ServiceTag"
            }
         },
         (vim.host.SystemIdentificationInfo) {
            identifierValue = "0123456789",
            identifierType = (vim.ElementDescription) {
               label = "Enclosure serial number tag",
               summary = "Enclosure serial number tag of the system",
               key = "EnclosureSerialNumberTag"
            }
         },
         (vim.host.SystemIdentificationInfo) {
            identifierValue = "0123456789",
            identifierType = (vim.ElementDescription) {
               label = "Serial number tag",
               summary = "Serial number tag of the system",
               key = "SerialNumberTag"
            }
         },
         (vim.host.SystemIdentificationInfo) {
            identifierValue = "Intel Haswell/Wellsburg/Grantley                  ",
            identifierType = (vim.ElementDescription) {
               label = "OEM specific string",
               summary = "OEM specific string",
               key = "OemSpecificString"
            }
         },
         (vim.host.SystemIdentificationInfo) {
            identifierValue = "Supermicro motherboard-X10 Series                 ",
            identifierType = (vim.ElementDescription) {
               label = "OEM specific string",
               summary = "OEM specific string",
               key = "OemSpecificString"
            }
         }
      ],
      memorySize = 137313644544,
      cpuModel = "Genuine Intel(R) CPU 0000 @ 2.00GHz",
      cpuMhz = 2000,
      numCpuPkgs = 1,
      numCpuCores = 12,
      numCpuThreads = 24,
      numNics = 4,
      numHBAs = 10
   },
Was $250 in 2018, if I remember correctly, about 15-20% of their going rate at the time.
 
  • Like
Reactions: Vit K

nevee

New Member
Jul 29, 2021
14
9
3
Hi guys,
Sorry for the long time without reply. I was sick.
Going back to the Gigabyte C621-WD12 - I've successfully connected the MPS USB programmer to the SMBus 3-pin connector like on the picture 1.jpg

Here I need to mention that all PWM controllers work on PMBus, but motherboard vendors use PMBus connector as PMBus power supply connector (5-pin),
to connect your power supply for better power management. This can be a little bit confusing, but when you connect USB programmer to this connector, you will not find any
PMBus device address, so simply use SMB connector on the motherboard and run full bus scanning.

Only what you need to do is to set ICC max on rails to 255A.
Save. Done.

If you have any questions, just let me know.
Cheers,
-nevee
 

Attachments

RolloZ170

Well-Known Member
Apr 24, 2016
1,640
424
83
55
Last edited:
  • Like
Reactions: RageBone

RolloZ170

Well-Known Member
Apr 24, 2016
1,640
424
83
55
Only what you need to do is to set ICC max on rails to 255A.
Save. Done.
Can you save this setting permanently ?

i was able to access the MP2955A of an supermicro X11SPM-F (JVR1)
unfortunaly i found no datasheet but with a dump of all registers(elmor labs EVC) i was able to find out more information.

surprisely my Gigabyte MD71-HB0 uses the same MP2955A and supports plat. 8222(C)L so i could read out the values required for the X11SPM-F.
interesting are some bytes, they are different from the MP2965.
address of the MD71 - CPU0 MP2955A is 30:

30,BF,2,True,2555 <- for recognizing the chip, 25 is the vendor ID of monolithic power systems

2555 = MP2955A
2556 = MP2956A
2558 = MP2958A
2565 = MP2965
 
Last edited:

Stephan

Active Member
Apr 21, 2017
401
233
43
Germany
I just read MP2965.pdf and presuming it is somewhat similar to MP2955A:

Can you save this setting permanently ?
What is your output of MFR_EEPROM_CTRL (1Dh)? Maybe chip has PMBUS_RW_PIN_EN disabled which will protect MFR_ICC_MAX from modification. Or a password has been set for PMBUS_RW_PWD_EN. Password can be brute-forced though, only 16-bit, check PWD_CHECK_CMD. There is also ADDR_PH pin but in this case chip would not react on SMBus. Also Page 0 should be selected before rewriting MFR_ICC_MAX.

Edit: There is more... EEPROM_WRITE_PROTECT (10h) and then you have to issue a STORE_USER_ALL (15h) command, to save your new page 0 to EEPROM. If you don't, power-cycle will reset everything to defaults...

Edit 2: There is MFR_ICC_MAX on page 0 for Rail 1 and MFR_ICC_MAX on page 1 for Rail 2. Which one are we talking about? :-o ;-)

30,BC,2,True,FF04 <- ICC_MAX 2 bytes
Are you sure this is BCh and not SMBus command BDh ("MFR_ICC_MAX")?

So Gigabyte has FF 04 which seems to set ICCMAX to 255+4*2=263A?

An RPi has 3.3 volts level on I2C pins and also compatible clock range for I2C (MPS can do 10..1000 kHz). With python-smbus2 it would take less than 10 lines of code to rewrite the MPS chip.
 
Last edited:
  • Like
Reactions: RageBone

RolloZ170

Well-Known Member
Apr 24, 2016
1,640
424
83
55
So Gigabyte has FF 04 which seems to set ICCMAX to 255+4*2=263A?
was wrong register:


MFR_OCP_SET_LEVEL (EEh) The MFR_OCP_SET_LEVEL command on Page 0 is used to set the Rail A OC limit level and the OCP delay time for the three rails.

OCP_LIMIT_ICC_MAX (EFh) The OCP_DA_LIMIT on Page 0 is used to set the single-phase valley current limit of Rail A. The MFR_ICC_MAX command on Page 0 is used to set the ICCMAX of Rail A.


-​
MFR_OCP_SET_LEVEL[EEh 2bytes]​
MFR_ICC_MAX[EFh 1byte]​
-​
MFR_OCP_SET_DELAYTIME | MFR_OCP_SET_LEVEL​
MFR_ICC_MAX​
X11SPM​
45|B2​
E4​
MD71​
45|BC​
FF​
 
Last edited:
  • Like
Reactions: RageBone

nevee

New Member
Jul 29, 2021
14
9
3
RolloZ170,

Yes, I can save it permanently without any issue. In my case everything is read/write.
I have in fact MP2968 but it's similar to MP2965.
I had to power on motherboard (even without CPU), to fully manage the MPS IC's through software.
If I leave motherboard only in standby, I can't do too much.
If you need MP29655A datasheet, you have to contact mpsnow@monolithicpower.com.
They will give you right datasheet.

Please let me know if you need some more details.
 
  • Like
Reactions: RolloZ170

RolloZ170

Well-Known Member
Apr 24, 2016
1,640
424
83
55
Yes, I can save it permanently without any issue. In my case everything is read/write.
I have in fact MP2968 but it's similar to MP2965.
thanks. the X11SPM uses MP2955A (CPU), MP2956A (chipset), 2x MP2958A (memory)

VR Roadmap.jpgVR Roadmap
 
Last edited:

RolloZ170

Well-Known Member
Apr 24, 2016
1,640
424
83
55
By the way, ask them which software and programmer you should use for your chip
thanks, very helpfull. have that software already dl, and the programmers(EVKT-USBI2C-02) are on the way to me.
found the string "MP2955A" in the software (HxD) :)
 
Last edited:

Stephan

Active Member
Apr 21, 2017
401
233
43
Germany
I checked and software mentioned by nevee seems only to support MP2888A MP2965 MP5023. So got lucky that board has MP2965. While there are strings talking about MP2955A, resources like images, dropdown box etc. are hardcoded to these three chips. pastebin dot com slash t59kMuTx could bring a light. Or will show and do something random and completely wrong. Help/About may also lie in your face.

Edit 1: See different paste.
Edit 2: Also check paste yY7HH6qq I think command 0xEF aka MFR_ICC_MAX is what we are after. 1 data byte so 255A possible. Rail1+2.
 
Last edited:

RolloZ170

Well-Known Member
Apr 24, 2016
1,640
424
83
55
I checked and software mentioned by nevee seems only to support MP2888A MP2965 MP5023. So he/she got lucky that his board has MP2965
nevee's C621 WD12 board has MP2968 (pre programmed MP2965) and the software was suggested by MPS.
and there is a universal MPS software(virtual_bench_pro_v4)
 
Last edited: