CoffeeOne wrote on 2020-02-22, 23:31:

Are you interested in newer hardware, too? I have some HP servers here, too. I booted my HP DL585 G7 with moderen CPUs (released end of 2012 😁)

Thank you for checking SIV out on the HP DL585 G7 and I am always interested to know how SIV get's on with multi-socket systems as they highlight issues that single socket ones don't, further if SIV "gets it wrong" then I like to know so I can fix it. I also suspect Martin would like to know how HWiNFO64 does on the G7.

I think all the issues stem from SIV setting up cpu->m_core (maximum cores) rather than cpu->e_core (enabled cores) , if you look at the APICs these are strange for CPU-16, CPU-24, etc.. This caused SIV to get the wrong number of boost states, 7 rather than 2 for CPUs 32->63, see Menu->Hardware->CPUID->CPUs 32->63->CPU-32. Further the temperatures are missing for CPU-32, 40, 48 and 56.

Please try the attached SIV64X 5.47 Core-01 and post the new save files. If things are still strange do SIV64X -DBGCPU -EXIT=12 > SIV_DBGOUT.log | MORE and also post the .log file.

When it's there I also like such as C:\Program Files\SIV64X\SIV_HP_DL585_G7.dmi which contains the raw data that is reported on the [Machine] panel as SIV runs better in test mode when I have it.

I noticed SIV was unable to report the DIMM SPD information and guess there must be an SPD multiplexor. Do any 3rd party utilities report this and if so which ones?

red-ray wrote on 2020-02-23, 09:37:

Thank you for checking SIV out on the HP DL585 G7 and I am always interested to know how SIV get's on with multi-socket systems […]
Show full quote

CoffeeOne wrote on 2020-02-22, 23:31:

Are you interested in newer hardware, too? I have some HP servers here, too. I booted my HP DL585 G7 with moderen CPUs (released end of 2012 😁)

Thank you for checking SIV out on the HP DL585 G7 and I am always interested to know how SIV get's on with multi-socket systems as they highlight issues that single socket ones don't, further if SIV "gets it wrong" then I like to know so I can fix it. I also suspect Martin would like to know how HWiNFO64 does on the G7.

I think all the issues stem from SIV setting up cpu->m_core (maximum cores) rather than cpu->e_core (enabled cores) , if you look at the APICs these are strange for CPU-16, CPU-24, etc.. This caused SIV to get the wrong number of boost states, 7 rather than 2 for CPUs 32->63, see Menu->Hardware->CPUID->CPUs 32->63->CPU-32. Further the temperatures are missing for CPU-32, 40, 48 and 56.

Please try the attached SIV64X 5.47 Core-01 and post the new save files. If things are still strange do SIV64X -DBGCPU -EXIT=12 > SIV_DBGOUT.log | MORE and also post the .log file.

When it's there I also like such as C:\Program Files\SIV64X\SIV_HP_DL585_G7.dmi which contains the raw data that is reported on the [Machine] panel as SIV runs better in test mode when I have it.

I noticed SIV was unable to report the DIMM SPD information and guess there must be an SPD multiplexor. Do any 3rd party utilities report this and if so which ones?

Yes, it does look already good.
Obviously when the machine is doing nothing, it is in constant "turbo" mode, so all cores show 3.2GHz, confirmed also by hwmonitor.

I will attach the dmi and the txt file (zipped).
No tool (at least I don't know one) can read out the DIMM SPD info on these servers.

The same is true for the harddisks / SSDs. They are hidden behind the hardware raid controllers, so windows only sees the logical drive(s).

Just a small cosmetical question about showing the DIMMs, like here 8GB + 8GB +8GB ..... + 8GB.
How will it look like when you have - for example + a machine with 64DIMMs and all 64 are equipped? This 585G7 only has 48 DIMM slots.

CoffeeOne wrote on 2020-02-23, 12:51:

The same is true for the harddisks / SSDs. They are hidden behind the hardware raid controllers, so windows only sees the logical drive(s).

Thank you and all the fixes worked as I hoped they would, but I just spotted SIV is reporting it as having 8 cores with SMT rather than 16 cores! This is down to CPUID 8000001E incorrectly reporting Threads per Core 2🙁 The attached SIV64X 5.47 NoSMT-01 should compensate for this.

As for more DIMMs then see the attached and if HP would tell me how to drive the DL585 G7 SMBus mux I suspect SIV would be able to report the SPD.

Sometimes it possible to read the drives behind a RAID controller and SIV can do this for such as my Intel® X299 Chipset SATA RAID Controller, further SIV can even read my Intel® Volume Management Device NVMe RAID Controller NVMe drive information. One day SIV will report the HP Smart Array P410i Controller drives, but as yet I have not managed to make this work, see Menu->Devices->CSMI RAID

red-ray wrote on 2020-02-23, 13:50:

Thank you and all the fixes worked as I hoped they would, but I just spotted SIV is reporting it as having 8 cores with SMT rath […]
Show full quote

CoffeeOne wrote on 2020-02-23, 12:51:

The same is true for the harddisks / SSDs. They are hidden behind the hardware raid controllers, so windows only sees the logical drive(s).

Thank you and all the fixes worked as I hoped they would, but I just spotted SIV is reporting it as having 8 cores with SMT rather than 16 cores! This is down to CPUID 8000001E incorrectly reporting Threads per Core 2🙁 The attached SIV64X 5.47 NoSMT-01 should compensate for this.

As for more DIMMs then see the attached and if HP would tell me how to drive the DL585 G7 SMBus mux I suspect SIV would be able to report the SPD.

Sometimes it possible to read the drives behind a RAID controller and SIV can do this for such as my Intel® X299 Chipset SATA RAID Controller, further SIV can even read my Intel® Volume Management Device NVMe RAID Controller NVMe drive information. One day SIV will report the HP Smart Array P410i Controller drives, but as yet I have not managed to make this work, see Menu->Devices->CSMI RAID

I did not (yet) test your latest version.
But are you sure that you want to change it?

There was an ongoing discussion about the AMD bulldozer architecture, AMD call the FX processors with 4 module "8-core" cpu. It is somehow true, at least for integer performance, but still 1 module shares some resources of 2 cores.
So it's the same here: AMD 6386 CPU were called 16-core by AMD.

But not by Microsoft anymore, see windows screenshot of the task-manager.

So when SIV reports 2 threads per core, it is inline with MS Windows Server 2012R2.

EDIT:
I am lost in your screenshot.
How many DIMMs are in that machine?
Where can I see it?
I was talking about number of DIMMs not number of cpus/cores.

EDIT one more time: You wrote it already 48 DIMMs.
There is written: 4x Sexa DIMMs (?)

CoffeeOne wrote on 2020-02-23, 14:00:

I did not (yet) test your latest version. But are you sure that you want to change it?

But not by Microsoft anymore, see windows screenshot of the task-manager. So when SIV reports 2 threads per core, it is inline with MS Windows Server 2012R2.

Thank you, now I remember and don't wish to change it given what Windows reports. I also just checked https://en.wikipedia.org/wiki/Opteron#Socket_G34 and SIV reporting Abu Dhabi is correct as is the L3 cache being 2 x 6MB.

I am wondering what other systems you have, do you have any with more than 64 CPUs and/or four sockets?

CoffeeOne wrote on 2020-02-23, 14:00:

How many DIMMs are in that machine?

It had 48 DIMMs, but as they report their temperature SIV reports this rather than their size, same on my system.

red-ray wrote on 2020-02-23, 14:09:

Thank you, now I remember and don't wish to change it given what Windows reports. […]
Show full quote

CoffeeOne wrote on 2020-02-23, 14:00:

I did not (yet) test your latest version. But are you sure that you want to change it?

But not by Microsoft anymore, see windows screenshot of the task-manager. So when SIV reports 2 threads per core, it is inline with MS Windows Server 2012R2.

Thank you, now I remember and don't wish to change it given what Windows reports.

I am wondering what other systems you have, do you have any with more than 64 CPUs and/or four sockets?

CoffeeOne wrote on 2020-02-23, 14:00:

How many DIMMs are in that machine?

It had 48 DIMMs, but as they report their temperature SIV reports this rather than their size, same on my system.

Ah I see, the 2 lines with the temperatures (48 in total) represent the number of DIMMs. OK, that was confusing to me.

Unfortunately I never managed to get an 8-socket machine, I have 3 more 4 socket ones: 2 time HP DL585 G5 (with CPUs, RAM and Raid controllers changed, so they are identical to HP DL585 G6)
and one HP DL580 G7 (with 64DIMMs) 😉

CoffeeOne wrote on 2020-02-23, 14:29:

HP DL580 G7 (with 64DIMMs) 😉

This is of interest, but I suspect both Martin and I will need to improve our SPD reading code. I expect SIV will read the SPD for some, but not all of the DIMMs.

I have a very old (21-Sep-2012) partial SIV save file from a HP DL980 G7

red-ray wrote on 2020-02-23, 14:09:

Thank you, now I remember and don't wish to change it given what Windows reports. I also just checked https://en.wikipedia.org/wiki/Opteron#Socket_G34 and SIV reporting Abu Dhabi is correct as is the L3 cache being 2 x 6MB.

Wait, that's wrong.
L3 is 2 x 8MB = 16 MB per CPU.

https://en.wikipedia.org/wiki/List_of_A ... 22_(32_nm)

Where do you have that 2 x 6MB from?

EDIT: HWiNFO64 shows 12MB for L3 cache, too. It is still wrong 😀

CoffeeOne wrote on 2020-02-23, 14:57:

Wait, that's wrong. L3 is 2 x 8MB = 16 MB per CPU.

Where do you have that 2 x 6MB from?

EDIT: HWiNFO64 shows 12MB for L3 cache, too. It is still wrong 😀

I suspect HWiNFO does not check the CPUID 8000001D data and I also think it has the wrong codename.

In the link it also says Abu Dhabi models have 16 MB of L3 cache (2x8 MB) but only 14 MB is visible with the HT Assist feature activated using 2 MB as a directory cache

The 12MB comes from CPUID 80000006 and the 2 x 6MB from CPUID 8000001D L3 Unified. The Share 8 means it's shared with 8 CPUs, so there is 2 x 6 MB per socket.

red-ray wrote on 2020-02-23, 15:05:

I suspect HWiNFO does not check the CPUID 8000001D data and I also think it has the wrong codename. […]
Show full quote

CoffeeOne wrote on 2020-02-23, 14:57:

Wait, that's wrong. L3 is 2 x 8MB = 16 MB per CPU.

Where do you have that 2 x 6MB from?

EDIT: HWiNFO64 shows 12MB for L3 cache, too. It is still wrong 😀

I suspect HWiNFO does not check the CPUID 8000001D data and I also think it has the wrong codename.

In the link it also says Abu Dhabi models have 16 MB of L3 cache (2x8 MB) but only 14 MB is visible with the HT Assist feature activated using 2 MB as a directory cache

The 12MB comes from CPUID 80000006 and the 2 x 6MB from CPUID 8000001D L3 Unified. The Share 8 means it's shared with 8 CPUs, so there is 2 x 6 MB per socket.

Ah yes, there is HT assist. I was just thinking about BIOS that reports 16MB for both L2 and L3. But 12MB is probably right, also windows display 48MB L3 for all 4 cpus. Bios reports the higher number that it looks nicer, so I take it back 😁

OK,

I run the same version of SIV also on the HP DL580 G7

Unfortunately running File Save Local crashed the remote desktop connection, I guess the machine crashed. I have to connect a monitor to the machine.

CoffeeOne wrote on 2020-02-23, 15:40:

Unfortunately running File Save Local crashed the remote desktop connection, I guess the machine crashed. I have to connect a monitor to the machine.

Thank you for trying SIV and apologies for it causing issues, which section was it processing when the issue happened?

If you do Menu->Tools->Maximum Lines->Set 58 Lines I suspect SIV will only use 3 columns.

SIV looks to have read half the SPD, but the temperatures seem too low, once I get a save file I will be able to tell more.

Windows crashes in the section [smb-bus] of File -> Save Local.
No blue screen, but display goes off, and machine does not react anymore to key-presses. So only power off (or holding powerswitch for 4 seconds) helps.
🙁

CoffeeOne wrote on 2020-02-23, 16:20:

Windows crashes in the section [smb-bus] of File -> Save Local.

Thank you, I see this once in a while and expect there is a strange SMBus device that I need to get SIV to skip.

To find what it is do siv64x -dbgsmb -save=[smb-bus]=SIV_DBGOUT.txt > SIV_DBGOUT.log | more and post the two SIV_DBGOUT files.

Once I get these I will get SIV to always skip that device on HP DL580 G7s.

Note when you do the above I expect the system will lock up again, but this time we should know how to stop it.

red-ray wrote on 2020-02-23, 15:56:

Thank you for trying SIV and apologies for it causing issues, which section was it processing when the issue happened? […]
Show full quote

CoffeeOne wrote on 2020-02-23, 15:40:

Unfortunately running File Save Local crashed the remote desktop connection, I guess the machine crashed. I have to connect a monitor to the machine.

Thank you for trying SIV and apologies for it causing issues, which section was it processing when the issue happened?

If you do Menu->Tools->Maximum Lines->Set 58 Lines I suspect SIV will only use 3 columns.

SIV looks to have read half the SPD, but the temperatures seem too low, once I get a save file I will be able to tell more.

Hi, here is a Save-selected, with the following setting:

In the zip-file, there is the dmi, the txt, and a dump from a crash before.

I just read your latest post, will do that, too.

red-ray wrote on 2020-02-23, 16:46:

Thank you, I see this once in a while and expect there is a strange SMBus device that I need to get SIV to skip. […]
Show full quote

CoffeeOne wrote on 2020-02-23, 16:20:

Windows crashes in the section [smb-bus] of File -> Save Local.

Thank you, I see this once in a while and expect there is a strange SMBus device that I need to get SIV to skip.

To find what it is do siv64x -dbgsmb -save=[smb-bus]=SIV_DBGOUT.txt > SIV_DBGOUT.log | more and post the two SIV_DBGOUT files.

Once I get these I will get SIV to always skip that device on HP DL580 G7s.

Note when you do the above I expect the system will lock up again, but this time we should know how to stop it.

SIV64X - System Information Viewer V5.47 Core-01 WIN-AGPOMB377GN::Administrator Built Feb 23 2020 at 11:26:04 [1252] Command [siv64x -dbgsmb -save=[smb-bus]=SIV_DBGOUT.txt ] [[smb-bus]] [C:\Program Files\SIV64X(1)\siv_dbgout.txt]

[smb-bus] [hardware]

Output:

EDIT: Yes, the machine locked up again. Needs a very long time to reboot unfortunately, it's not like a normal PC

EDIT2:

The program terminated (but no windows lock-up) also when I un-selected only [smb-bus] from save local.
So for the save-local I unselected all 3 smb entries (see above)

CoffeeOne wrote on 2020-02-23, 17:08:

The program terminated (but no windows lock-up) also when I un-selected only [smb-bus] from save local.
So for the save-local I unselected all 3 smb entries (see above)

Which section was being processed when SIV64X.exe crashed out?

It looks to be device [ 1_69 ] that triggered the lockup and the attached SIV64X 5.47 DL580-01 should skip it. I feel it's a good idea to do siv64x -dbgsmb -save=[smb-bus]=SIV_DBGOUT.txt > SIV_DBGOUT.log | more again in case there is a 2nd rouge device. If that runs OK then please go for the full Save Local.

Looking at the save file SIV did quite well as it read 32 lots of SPD data and the 4 Xeon PIRs via the systems 5 SMBuses, it has 1 x PCH + 4 x IMC. Once we get [SMB Bus] working it will be interesting to see if there are any SMBus multiplexors.

red-ray wrote on 2020-02-23, 17:46:

Which section was being processed when SIV64X.exe crashed out?

It looks to be device [ 1_69 ] that triggered the lockup and the attached SIV64X 5.47 DL580-01 should skip it. I feel it's a good idea to do siv64x -dbgsmb -save=[smb-bus]=SIV_DBGOUT.txt > SIV_DBGOUT.log | more again in case there is a 2nd rouge device. If that runs OK then please go for the full Save Local.

Looking at the save file SIV did quite well as it read 32 lots of SPD data and the 4 Xeon PIRs via the systems 5 SMBuses, it has 1 x PCH + 4 x IMC. Once we get [SMB Bus] working it will be interesting to see if there are any SMBus multiplexors.

The first step went well ....

But save local still crashed windows in smb-bus section ....

I reboot the machine one more (the last for today) time in order to send you something

Wow, looked at the monitor in the basement: Machine rebooted, "Uncorrectable Memory Error in Board 4".
Did SIV really find a hardware problem on the machine?
Never have seen that before though.

red-ray wrote on 2020-02-23, 17:46:

Which section was being processed when SIV64X.exe crashed out? […]
Show full quote

CoffeeOne wrote on 2020-02-23, 17:08:

The program terminated (but no windows lock-up) also when I un-selected only [smb-bus] from save local.
So for the save-local I unselected all 3 smb entries (see above)

Which section was being processed when SIV64X.exe crashed out?

It looks to be device [ 1_69 ] that triggered the lockup and the attached SIV64X 5.47 DL580-01 should skip it. I feel it's a good idea to do siv64x -dbgsmb -save=[smb-bus]=SIV_DBGOUT.txt > SIV_DBGOUT.log | more again in case there is a 2nd rouge device. If that runs OK then please go for the full Save Local.

Looking at the save file SIV did quite well as it read 32 lots of SPD data and the 4 Xeon PIRs via the systems 5 SMBuses, it has 1 x PCH + 4 x IMC. Once we get [SMB Bus] working it will be interesting to see if there are any SMBus multiplexors.

I attach a zip file with siv_dbgout.txt, the txt and the dmi file