SQM Network Performance Comparison: J5005 vs i5-8500 with Intel i350 NIC

Hardware Questions and Recommendations
1

I had the opportunity to test two systems, one with a J5005 CPU and the other with an i5-8500, both equipped with Intel I350 (1 Gbit) NICs. These are older platforms, dating back to 2018.

Test Systems

Dell Wyse 5070 (J5005)
CPU Benchmark Scores (cpubenchmark.net):

  • Multithread Rating: 3,123
  • Single Thread Rating: 1,210

i5-8500 System
CPU Benchmark Scores (cpubenchmark.net):

  • Multithread Rating: 9,540
  • Single Thread Rating: 2,445

NICs Tested

I used two i350-based quad-port NICs. I swapped them between the systems during testing and observed no significant performance differences:

  • FSC Fujitsu D3045-A11 Quad-Port LP Gigabit Ethernet Adapter (Intel i350-AM4)
  • Supermicro AOC-SGP-i4

Test Setup

  • Tool: Netperf, running in an LXC container
  • CPU Allocation: 2x 5000 MHz P-cores pinned to the container
  • Network: Gigabit LAN
  • Intel SpeedStep: Enabled
  • Test: RRUL (Real-Time Response Under Load), using the default profile
  • OpenWrt: OpenWrt 24.10. SQM set to upload and download set to 921600 with default SQM settings.

Notes on Plots

The default RRUL test didn’t allow setting a fixed Y-axis scale, so I ran the test multiple times with separate configurations for download, upload, and ping. I then combined these into vertical image stacks per system. Finally, I merged the images from both systems side-by-side for comparison.

Plots

Can´t get it to visualize properly. Trying to share it over Dropbox instead:

Comments & Questions

The results are nearly identical, despite the CPUs being in completely different classes. Is there anything I can tweak to further optimize performance? I'm hesitant to disable Intel SpeedStep due to the potential increase in heat and power consumption. Or does this indicate that, for a 1 Gbit symmetrical connection, the J5005 is already more than adequate?

EDIT: Fixed the plot because it was not showed properly.
EDIT: Added question

2

So you can open flent with the --gui option and then load (multiple) data files in that GUI. Yu then can actually merge data and configure axis limits for all axes independently.

Unfortunately you plot dies not show at all... but clicking the broken image leads me to dropbox where I can see it.

So yes, for this load both CPUs are fully sifficient.

3

Thanks for sharing info about the GUI features. I haven´t tried it much yet. Currently using a shell script to run tests and merge images to an easily comparable overview.

And thanks for validating the findings that for this use case, even the weaker CPU is sufficient.

4

Don't sell it short - the J-series are more than enough for a gigabit symmetric connection - clocks are everything on x86/amd64

sidenote

I've got an older Lanner box - C3758 8c/8t@2.2GHz, 16gb RAM/64GB SSD/8GB eMMC...

Wireless is a bit odd, with a older QCA wifi5 card (dual band, single radio) and a Sierra Wireless 4G/LTE modem...

Wired - i350 NIC's for all 8 ports (incl 2 SFP), and it has the Aspeed BMC option for out of band management.

It's new-in-box, bought it for a project that was cancelled...

Happy to sell it for someone that can make it work for them...