nice list from wired zone, looks like you want either of the bottom two..
X10SDV System-on-a-Chip (SoC) D-1500 Solutions
Intel Xeon D-1500 Series Discussion
- Thread starter Patrick
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
X10SDV System-on-a-Chip (SoC) D-1500 Solutions
Thank you! The X10SDV-7TP4F looks perfect. The D-1587 version looks even more perfect, but out of my budget.
I've got my Xeon D 1541 SuperMicro 5028D-TN4T which is a lovely little box however despite having SR-IOV in the BIOS it seems Windows Server 2012 R2 isn't happy with it. It claims the following:
To use SR-IOV on this system, the system BIOS must be updated to allow
Windows to control PCI Express. Contact your system manufacturer for an
update., SR-IOV cannot be used on this system as the PCI Express hardware
does not support Access Control Services (ACS) at any root port. Contact
your system vendor for further information.
Before you say I've got ACS switched on in the BIOS along with VT-D, SR-IOV and anything else I think is remotely related. It's also on BIOS 1.1.
For reference I tried Server 2016 TP5 which has a direct Passthrough option for more hardware but it also claims the motherboard doesn't support ACS.
Does anyone have any ideas here? I've tried Proxmox and split out the NIC into Virtual Functions but not tried to assign. However if Windows is complaining like this then something is wrong with either my BIOS settings or the BIOS as a whole.
To use SR-IOV on this system, the system BIOS must be updated to allow
Windows to control PCI Express. Contact your system manufacturer for an
update., SR-IOV cannot be used on this system as the PCI Express hardware
does not support Access Control Services (ACS) at any root port. Contact
your system vendor for further information.
Before you say I've got ACS switched on in the BIOS along with VT-D, SR-IOV and anything else I think is remotely related. It's also on BIOS 1.1.
For reference I tried Server 2016 TP5 which has a direct Passthrough option for more hardware but it also claims the motherboard doesn't support ACS.
Does anyone have any ideas here? I've tried Proxmox and split out the NIC into Virtual Functions but not tried to assign. However if Windows is complaining like this then something is wrong with either my BIOS settings or the BIOS as a whole.
I raised a ticket with SuperMicro today around SR-IOV support as from research a BIOS option of ASPM is missing in the PCI Express configuration part of the BIOS. Well they haven't come back to me just yet however saw someone mention a new BIOS, 1.1a has just been uploaded.
I've flashed this and low and behold the only item I can notice that's changed in the BIOS is ASPM has appeared in the PCI Express section of the BIOS. I've booted up 2012 R2 and it's now not only reporting that SR-IOV can work properly but it does work properly. I've been working on this on and off for a week, cannot imagine you people who brought these systems earlier last year who have had to wait so long. Still I'm happy and just a bit more testing and I can finally start to use my server.
Before that though I'll load up Server 2016 TP5 as I suspect this solves it's issue with Discrete Device Assignment. I've got a half height Radeon 7750 so will give that a shot at trying to pass it to a VM from 2016. Maybe tomorrow though as it's rather late now in the UK
I've flashed this and low and behold the only item I can notice that's changed in the BIOS is ASPM has appeared in the PCI Express section of the BIOS. I've booted up 2012 R2 and it's now not only reporting that SR-IOV can work properly but it does work properly. I've been working on this on and off for a week, cannot imagine you people who brought these systems earlier last year who have had to wait so long. Still I'm happy and just a bit more testing and I can finally start to use my server.
Before that though I'll load up Server 2016 TP5 as I suspect this solves it's issue with Discrete Device Assignment. I've got a half height Radeon 7750 so will give that a shot at trying to pass it to a VM from 2016. Maybe tomorrow though as it's rather late now in the UK
Scott, there is a version of D1587 both with and without cpufan as standard. If you have a 1U chassi you need the one without thou.
So I made some improvments to the shroud and from having 105 celcius I am now down to 80 +- 5 degress celcius. So I guess an optimal shroud is kinda of important (d'Oh!).
So I made some improvments to the shroud and from having 105 celcius I am now down to 80 +- 5 degress celcius. So I guess an optimal shroud is kinda of important (d'Oh!).
The board I'm interested in has just a heatsink. Are they removable/replaceable? If not, how do they perform in an open case like a 3 or 4u?
They work much better with active airflow over the heatsink. SM currently only ships them in their 1U systems where the fans are fairly direct on the heatsink.
Several of us here have found that some form of duct or shroud to direct your fan's airflow onto the CPU heatsink is very helpful.
I don't know your goals ...but... if flex doesn't cut it for you....
I had the same thought over a year ago when the Xeon D 1500's started coming out.
I ended up getting X10SRi-F + E5-2620v3 and building it out to cost the same as the
X10SDV-8C-TLN4F.
The E5-26xxV4 chips look promising too.
I just got an X10SDV-TP8F to replace an older Atom ASRi-2758F.
I was/am using the Supermicro 504-203B 1U chassis.
The old Atom setup fit and the height of the CPU+Heatsink was only 1/2 U so
I was able to place a Noctua 40x10mm fans on top of the heatsink horizontally.
I cannot do the same with X10SDV-TP8F as the CPU+Heatsink height is a full 1U, so
I will have to mount it/them vertically.
So I am not sure If I will try to maintain the 1U form factor and try an shroud:
Air Shroud MCP-310-50501-0B 1 mylar air shroud for SC504/505
Or just convert to 2U like my other nodes.
Anyone else using the 1U SC505/504 chassis?? Are you using the SuperMicro 40mm Fans and shroud??
I'm afraid the 8000RPM fans will kill my quiet setup.
Maybe I can get the SM fan bracket and just use 2 x 40mm Noctua fans instead...hmmm.
What of BDE66?
Concurrent 10 GbE Port Operation May Result in Packet Loss or Reduced Bandwidth
Problem: Due to this erratum, the processor is not able to sustain concurrent, high bandwidth
operation of both 10 GbE ports using small packets (about 112 bytes or less).
Implication: If flow control is enabled, this erratum may result in reduced 10 GbE port bandwidth. If
flow control is not enabled, this erratum may result in packet loss. This erratum may be
observed during the L3 forwarding performance test under RFC2544 using the Intel
Data Plan Development Kit. This erratum has not been observed when exactly one 10
GbE port is active. [1]
Among the other No Fixes, this seems fairly serious for an entire line featuring 10gb. Has anyone tested or encountered this yet?
1 https://www-ssl.intel.com/content/d...-updates/xeon-d-1500-specification-update.pdf
Concurrent 10 GbE Port Operation May Result in Packet Loss or Reduced Bandwidth
Problem: Due to this erratum, the processor is not able to sustain concurrent, high bandwidth
operation of both 10 GbE ports using small packets (about 112 bytes or less).
Implication: If flow control is enabled, this erratum may result in reduced 10 GbE port bandwidth. If
flow control is not enabled, this erratum may result in packet loss. This erratum may be
observed during the L3 forwarding performance test under RFC2544 using the Intel
Data Plan Development Kit. This erratum has not been observed when exactly one 10
GbE port is active. [1]
Among the other No Fixes, this seems fairly serious for an entire line featuring 10gb. Has anyone tested or encountered this yet?
1 https://www-ssl.intel.com/content/d...-updates/xeon-d-1500-specification-update.pdf
@sthquestionaire interesting read. Since most of the tests/ applications we have had for these have been storage and web, likely few have run high bandwidth applications with only that packet size.
I generally wouldn't expect wire speed across all ports with minimum size packets with server hardware. Traditionally, servers are PPS-bound, not really bandwidth-bound. 20 Gbps with 112 byte frames is more than 20Mpps; that's dramatically more than you could handle without special-case low-latency packet handling code (like Intel's DPDK). With 1500-byte frames, full bandwidth across both ports is under 2Mpps, and with TSO/RSO, that probably looks like 1/10th of that to the OS. Cloudflare suggests that 50kpps/core (should really be thread, I'd think--the limit is usually cache misses, and that'd favor threads heavily) is usually about the limit for most servers. Faster CPU speeds don't really help much, because it's not compute-bound, it's cache-miss-bound. Intel has done some work with pre-populating caches with headers from inbound packets; presumably that's one of the tricks that the DPDK is using to get 20Mpps.
If you're trying to build a high-performance piece of network equipment that can do wire speed with 64 byte frames, then maybe a Xeon-D is the wrong tool. For more normal workloads, I don't see how this is a problem.
If you're trying to build a high-performance piece of network equipment that can do wire speed with 64 byte frames, then maybe a Xeon-D is the wrong tool. For more normal workloads, I don't see how this is a problem.
@Scott Laird The point of this errata is that the low latency packet handling code (DPDK) doesn't function properly. These will not compare to a dedicated packet processor, but the errata and problem statement are vague.
If one has a high speed transfer out of one 10gb interface and a few small packets come out of the other, will the errata be triggered?
One case for example: Fast transfers on one interface, small keep-alive packets going out the other.
Depending on clock/cache/config, the 12, 16 and likely 8 core xeon-ds should be able to handle a lot without crashing down on a few small packets.
If one has a high speed transfer out of one 10gb interface and a few small packets come out of the other, will the errata be triggered?
One case for example: Fast transfers on one interface, small keep-alive packets going out the other.
Depending on clock/cache/config, the 12, 16 and likely 8 core xeon-ds should be able to handle a lot without crashing down on a few small packets.
Last edited:
Can anyone recommend a budget RAM and M2 SSD to use as cache in a D1508 board? I'd think I would just want to start with 2 sticks of RAM.
OS is unRaid. Already have 6 SATA data/parity drives. I would be upgrading an old super micro atom 525 board.
I don't use the server for anything beyond basic NAS functionality. However, I may setup something like Plex. I'm having extremely slow read and write issues with my current setup that I'm hoping some new hardware will fix.
OS is unRaid. Already have 6 SATA data/parity drives. I would be upgrading an old super micro atom 525 board.
I don't use the server for anything beyond basic NAS functionality. However, I may setup something like Plex. I'm having extremely slow read and write issues with my current setup that I'm hoping some new hardware will fix.
For a simple file server using Unraid: https://www.amazon.com/Samsung-950-PRO-Internal-MZ-V5P256BW/dp/B015SOI392/
The new SM961 is coming out but the price of the 950 Pro is falling so I wouldn't worry about it.
RAM: Just get 2x 16GB DDR4 RDIMM off of ebay: SAMSUNG 16GB PC4-2133P PC4-17000R DDR4 REG-ECC DIMM M393A2G40DB0-CPB
The new SM961 is coming out but the price of the 950 Pro is falling so I wouldn't worry about it.
RAM: Just get 2x 16GB DDR4 RDIMM off of ebay: SAMSUNG 16GB PC4-2133P PC4-17000R DDR4 REG-ECC DIMM M393A2G40DB0-CPB