HP T730 8GB RAM 32GB Storage Thin Client

WANg

Well-Known Member
Jun 10, 2018
1,089
682
113
43
New York, NY
Last edited:

ske4za

Member
Feb 4, 2019
77
37
18
Neither of these seem to be installable by FreeBSD's "pkg"
Just to make sure, you configured AMD under Thermal Sensors in Advanced > Misc > Cryptographic & Thermal Hardware > Thermal Sensors?

If nothing else, this closed bug #10226 shows that he's able to read the temp in dev version 2.5 (based on FreeBSD 12.0).
 

nthu9280

Well-Known Member
Feb 3, 2016
1,622
460
83
San Antonio, TX
I don't recall doing anything special on my T620. I understand we are talking about T730 and different HW / CPU here. It's been couple of years and since I configured it. Don't recall If I needed to change/update any setting in BIOS.

1604355357322.png
 

unmesh

Member
Apr 17, 2017
70
18
8
62
Just to make sure, you configured AMD under Thermal Sensors in Advanced > Misc > Cryptographic & Thermal Hardware > Thermal Sensors?

If nothing else, this closed bug #10226 shows that he's able to read the temp in dev version 2.5 (based on FreeBSD 12.0).
Somehow I missed doing that :-(

It shows the temperature now.

Thanks
 

unmesh

Member
Apr 17, 2017
70
18
8
62
I started experimenting with Proxmox on this and a Lenovo M900 Tiny and thought I'd report on a curious CPU performance finding.

AMD RX427BB 2C4T 2.7-3.6GHz dual-channel DDR3 has a CPUmark of 2700 and pveperf shows CPU BOGOMIPS: 21559.48

I thought the Intel i5-6500T 2.5GHz 4C4T dual-channel DDR4 with a CPUmark of 4796 would be faster regardless of OS but pveperf shows CPU BOGOMIPS: 19999.60

Are BOGOMIPS not a useful indicator of CPU performance?
 

Samir

Well-Known Member
Jul 21, 2017
1,741
621
113
46
I started experimenting with Proxmox on this and a Lenovo M900 Tiny and thought I'd report on a curious CPU performance finding.

AMD RX427BB 2C4T 2.7-3.6GHz dual-channel DDR3 has a CPUmark of 2700 and pveperf shows CPU BOGOMIPS: 21559.48

I thought the Intel i5-6500T 2.5GHz 4C4T dual-channel DDR4 with a CPUmark of 4796 would be faster regardless of OS but pveperf shows CPU BOGOMIPS: 19999.60

Are BOGOMIPS not a useful indicator of CPU performance?
I think that maybe there's some sort of optimization on the AMD that makes bogomips run better, but generally the cpu will be slower as you mentioned.
 

WANg

Well-Known Member
Jun 10, 2018
1,089
682
113
43
New York, NY
I started experimenting with Proxmox on this and a Lenovo M900 Tiny and thought I'd report on a curious CPU performance finding.

AMD RX427BB 2C4T 2.7-3.6GHz dual-channel DDR3 has a CPUmark of 2700 and pveperf shows CPU BOGOMIPS: 21559.48

I thought the Intel i5-6500T 2.5GHz 4C4T dual-channel DDR4 with a CPUmark of 4796 would be faster regardless of OS but pveperf shows CPU BOGOMIPS: 19999.60

Are BOGOMIPS not a useful indicator of CPU performance?
No - it’s defined as the millions of times per second your CPU can be spent doing absolutely nothing - as far as I know BogoMIPS is just there to calibrate the timing for polled I/O in certain device drivers. As a CPU perf indicator? Completely meaningless.

And yes, the RX427BB is closer in performance to a Haswell i3 or i5 Mobile but with a better iGPU…which kinda makes sense since it came out around 2014 and it’s really just a repackaged FX7600p APU.
 

unmesh

Member
Apr 17, 2017
70
18
8
62
No - it’s defined as the millions of times per second your CPU can be spent doing absolutely nothing - as far as I know BogoMIPS is just there to calibrate the timing for polled I/O in certain device drivers. As a CPU perf indicator? Completely meaningless.

And yes, the RX427BB is closer in performance to a Haswell i3 or i5 Mobile but with a better iGPU…which kinda makes sense since it came out around 2014 and it’s really just a repackaged FX7600p APU.
Thanks for the insight. And the education.

I've got a HP NC365T quad NIC in the T730 and can run pfSense as a VM on Proxmox with the ports as LAN bridges but not if they are passed through to the VM (pfSense says em0 and em1 not found). Is it because this NIC does not support SR-IoV or is that not relevant if I am passing the entire PCI card through?
 

WANg

Well-Known Member
Jun 10, 2018
1,089
682
113
43
New York, NY
Thanks for the insight. And the education.

I've got a HP NC365T quad NIC in the T730 and can run pfSense as a VM on Proxmox with the ports as LAN bridges but not if they are passed through to the VM (pfSense says em0 and em1 not found). Is it because this NIC does not support SR-IoV or is that not relevant if I am passing the entire PCI card through?
The RX427BB doesn’t really do SRIOV - well, it does and it doesn’t. It can do it in Linux (not VMWare ESXi) but since it doesn’t do PCIe ACS (access control services) and ARI (alternative routing-ID interpretation) it can only present up to 7 virtual functions and does so in a rather unsafe manner. @arglebargle mentioned that there is a kernel patch around that allows it, but I don’t remember which one. I don’t remember if the NC365T is Intel i350 based…if it is, then it does SRIOV. If it isn’t, then It wouldn’t.
 

unmesh

Member
Apr 17, 2017
70
18
8
62
The RX427BB doesn’t really do SRIOV - well, it does and it doesn’t. It can do it in Linux (not VMWare ESXi) but since it doesn’t do PCIe ACS (access control services) and ARI (alternative routing-ID interpretation) it can only present up to 7 virtual functions and does so in a rather unsafe manner. @arglebargle mentioned that there is a kernel patch around that allows it, but I don’t remember which one. I don’t remember if the NC365T is Intel i350 based…if it is, then it does SRIOV. If it isn’t, then It wouldn’t.
It is a i340
 

unmesh

Member
Apr 17, 2017
70
18
8
62
No, HP NC365T controller is Intel 82571EB, very old chip.
Interesting. Is the following output that mentions 82580 a software artifact?

Code:
#pciconf -lv
igb0@pci0:1:0:0: class=0x020000 card=0x1780103c chip=0x150e8086 rev=0x01 hdr=0x00
vendor = 'Intel Corporation'
device = '82580 Gigabit Network Connection'
class = network
subclass = ethernet
igb1@pci0:1:0:1: class=0x020000 card=0x1780103c chip=0x150e8086 rev=0x01 hdr=0x00
vendor = 'Intel Corporation'
device = '82580 Gigabit Network Connection'
class = network
subclass = ethernet
igb2@pci0:1:0:2: class=0x020000 card=0x1780103c chip=0x150e8086 rev=0x01 hdr=0x00
vendor = 'Intel Corporation'
device = '82580 Gigabit Network Connection'
class = network
subclass = ethernet
igb3@pci0:1:0:3: class=0x020000 card=0x1780103c chip=0x150e8086 rev=0x01 hdr=0x00
vendor = 'Intel Corporation'
device = '82580 Gigabit Network Connection'
class = network
subclass = ethernet
 

RTM

Well-Known Member
Jan 26, 2014
737
270
63
No, HP NC365T controller is Intel 82571EB, very old chip.
I believe that page you linked to is wrong.
The NC364T is two 82571EB's, the NC365T is a single 82580 (the chip that is used on the i340 NIC).

See page 12:

And page 11:

PS: As far as I know the 82580 does not support SR-IOV, for that you want i350 (or alternatively one of the older chips that DO support SR-IOV)

Another edit: I may be wrong here, but assuming that passthrough works as expected (iommu enabled and all that), I would expect that you can pass through the entire controller (but probably not individual ports).
 
Last edited:

unmesh

Member
Apr 17, 2017
70
18
8
62
...
PS: As far as I know the 82580 does not support SR-IOV, for that you want i350 (or alternatively one of the older chips that DO support SR-IOV)
That is what I remember too from the reading I'd done at the time of purchase.

Does one need SR-IOV if one wants to pass through the entire card, though?
 

RTM

Well-Known Member
Jan 26, 2014
737
270
63
That is what I remember too from the reading I'd done at the time of purchase.

Does one need SR-IOV if one wants to pass through the entire card, though?
See my edit, I don't believe that to be necessary.
 

unmesh

Member
Apr 17, 2017
70
18
8
62
@unmesh No need for SR-IOV if you're passing the entire card. Just make sure to blacklist the drivers in Proxmox.
So I've enabled AMD IOMMU, loaded the vfio modules, blacklisted the igb drivers, run initramfs, rebooted and passed through the NIC to a pfSense VM but it still thinks there are no network interfaces. I should redo it and also see if a Linux VM sees the Ethernet ports.

Is there a way to tell that the host is not seeing these interfaces?

Also, do the 4 Ethernet ports have to be in their own IOMMU group with nothing else?
 

mimino

Active Member
Nov 2, 2018
182
68
28
So I've enabled AMD IOMMU, loaded the vfio modules, blacklisted the igb drivers, run initramfs, rebooted and passed through the NIC to a pfSense VM but it still thinks there are no network interfaces. I should redo it and also see if a Linux VM sees the Ethernet ports.

Is there a way to tell that the host is not seeing these interfaces?

Also, do the 4 Ethernet ports have to be in their own IOMMU group with nothing else?
Is em driver loaded in pfSense? What's the output of lspci?
 

unmesh

Member
Apr 17, 2017
70
18
8
62
Is em driver loaded in pfSense? What's the output of lspci?
pfSense complains about not finding em0 and em1 and restarts so I can only look at the boot output as it scrolls by though it is pretty fast. FWIW, pfSense runs on the bare metal just fine.