New 4x 2.5GbE and 2x 10GbE Intel Core Firewall and Virtualization Appliance

30

4x 2.5GbE and 2x 10GbE Intel Core Firewall and Virtualization Appliance Performance

The performance section is one where we find another quirk. Here is the quick topology of this system. Our best guess is that the reason the 10th Gen Core U-series is being used here instead of Alder Lake-N is that the Core i5 and Core i7 have 16 PCIe lanes instead of just 9. If four lanes go to the riser, and one each to the 2.5GbE NICs, that is 8 lanes before even getting to the SSD or other components.

AliExpress 4x 2.5GbE And 2x 10GbE Topology
AliExpress 4x 2.5GbE And 2x 10GbE Topology

Next, we are going to use our standard Linux Bench results and Geekbench 5 to discuss the performance of this compared to some of the systems that we have reviewed.

Python Linux 4.4.2 Kernel Compile Benchmark

This is one of the most requested benchmarks for STH over the past few years. The task was simple, we have a standard configuration file, the Linux 4.4.2 kernel from kernel.org, and make the standard auto-generated configuration utilizing every thread in the system. We are expressing results in terms of compiles per hour to make the results easier to read:

Intel Core I7 10510U And Core I5 10210U Linux Kernel Compile Benchmark
Intel Core I7 10510U And Core I5 10210U Linux Kernel Compile Benchmark

On these charts, we have not just the Intel Core i7-10510U, but also the Core i5-10210U highlighted. Since we have both CPUs in the two systems it felt like we should put the results on the same chart since they are both options. While they are Core i5 and Core i7 CPUs, do not be fooled. They are 10th gen Core processors and low-power ones at that.

7-zip Compression Performance

7-zip is a widely used compression/ decompression program that works cross-platform. We started using the program during our early days with Windows testing. It is now part of Linux-Bench.

Intel Core I7 10510U And Core I5 10210U 7zip Compression Benchmark
Intel Core I7 10510U And Core I5 10210U 7zip Compression Benchmark

The CPUs generally are close to the quad-core Intel N200 which is not what we were expecting. Part of that is due to the fact that these systems get up to around 73-75C with the cooling that is offered.

OpenSSL Performance

OpenSSL is widely used to secure communications between servers. This is an important protocol in many server stacks. We first look at our sign tests:

Intel Core I7 10510U And Core I5 10210U OpenSSL Sign Benchmark
Intel Core I7 10510U And Core I5 10210U OpenSSL Sign Benchmark

Here are the verify results:

Intel Core I7 10510U And Core I5 10210U OpenSSL Verify Benchmark
Intel Core I7 10510U And Core I5 10210U OpenSSL Verify Benchmark

Again, we have results closer to the new Alder Lake-N parts that are closer than we were expecting. This is a reason that the upcoming 144-core Sierra Forest CPU with all E-cores is promising. Here we are seeing newer E-cores compete with previous-generation P-cores on a 1:1 basis.

Intel Core i5-10210U and Intel Core i7-10510U Geekbench Results

On the Geekbench side, things got strange. First, the Core i5-10210U and Core i7-10510U Geekbench 5 comparison:

Intel Core I7 10510U And Core I5 10210U GeekBench 5
Intel Core I7 10510U And Core I5 10210U GeekBench 5

The takeaway here is that there is a gain with the Core i7, but it is not an enormous one by any means. When we instead look at the Core i5 to the Alder Lake-N here is what we get:

Intel Core I5 10210U GeekBench 5 Vs Core I3 N305 And N200
Intel Core I5 10210U GeekBench 5 Vs Core I3 N305 And N200

The 10th Gen Intel Core i5 and Core i7 are no match for the Alder Lake-N, especially the 8-core Intel Core i3-N305.

What was extra strange was what happened when we ran Geekbench 6. We had the Core i7 unit running a long-term stress-ng stability test, and it passed. We thought it would fail at some point, but it kept running, albeit hot for weeks on end. We were running benchmarks on the Core i5 in the 8x 2.5GbE system in the meantime and when it got to Geekbench 6, it just quit and the system rebooted.

Intel Core I5 10210U Geekbench 6 Fail Point
Intel Core I5 10210U Geekbench 6 Fail Point

We then tried the Core i7 system after it was done with weeks of stress-ng, and it failed Geekbench 6 as well. We did multiple tests, and it happened several times at different parts of the Geekbench 6 (6.0 and 6.10) multi-core tests. We are not sure why this happens, and it is really a black mark on these systems. It did not happen in other benchmarks, but it was consistent on both systems.

Next, let us get to the power consumption and noise.

30 COMMENTS

  1. I’d honestly love to see the 10Gb units as Debian kubernetes hosts.

    Or even better, find a compatible 4-6 bay SSD chassis (making use of those onboard SATA ports). With how cheap second-hand ~250-500G enterprise SATA drives are, this is potentiallu looking like the most cost effective way to make a strongly perfirmant cephfs cluster.

    I’m thinking the x520 would get the latency values low enough that one could realistically deploy high performance distributed storage in a home lab, without having to sacrifice their firstborn… I might have to give this a shot!!

  2. As i read this article I thought I was reading another installment of “What’s absurd to you is totally normal for us” on the “Fanless tech” website.

    So this device has SATA ports that you may or may not be able to use. That’s a waste of a Dollar or more in costs. And those Molex ports – more waste.

    And those “over the top” CPUs that make the case hot. Does the case get as hot as a Texas concrete road at the peak of summer heat when it doubles as a frying pan?

    Ok, interesting product but…a perfect case of how to stuff 10 pounds of stuff in a 2 ounce baggie.

    No thanks.

  3. This is an excellent little box, add a bunch of SATA drives for bulk storage, and you got yourself an all-in-one node. Get 3 nodes, and now you have a Ceph cluster, and you can use DAC to have each node connect to two other nodes, not even going to need a switch. 2.5G for some inter-node communication and you’re essentially ready for Proxmox and Ceph or Rancher HCI.

  4. @Patrick Did you try ESXi 8/8U1 during your testing? The VMware HCL lists a handful of X520-based NICs as compatible with 8, using the ixgben driver. Wondering if it could be coaxed into working if it’s not out of the box.

  5. The review states, “We then tried the Core i7 system after it was done with weeks of stress-ng, and it failed Geekbench 6 as well. We did multiple tests, and it happened several times at different parts of the Geekbench 6 (6.0 and 6.10) multi-core tests.”

    Since the system is not stable enough to complete one of the tests, how is this going to work as a server? Would extra cooling make it stable? Would down clocking the RAM fix the crashes? What about CPU speeds?

    Without knowing why the system is unstable and how to fix it, I don’t see how the conclusions could be anything other than doesn’t work stay away.

  6. Earlier today, the Youtube thumbnail was nauseatingly positive about these little units, faulty as they were. Then the article says We don’t like the heat, the noise, the TDP, the stability, the chosen components…

  7. Hey guys running around today. Eric – we literally had stress-ng and iperf3 running on this unit for 30 days straight. The only thing that crashed it was Geekbench 6, but not 5. All we can do is say what we found. Steve – we tried 8. I usually try staying safe on VMware since the HCL is so much pickier than other OSes.

    MOSFET – I can tell you that is completely untrue. This was scheduled to go-live at 8AM Pacific. I was heading to go film something in Oregon so everything was scheduled. The thumbnail only says 10GbE 2.5GbE and that is all it has said. The title has not changed either. I am not sure what is “nauseatingly positive” when it is just factually the speed of the ports.

  8. Man, I literally just bought (and had delivered) a Moginsok N100 solution with 4 x 2.5Gbit ports. Not sure how I missed this, as it would have completely changed my calculus.

  9. Since Geekbench crashes at different places each run, this points to a hardware malfunction. My suspicion is continuous stress-ng creates too predictable a load to crash the system.

    If I were to guess, the on-board DC-to-DC power regulators aren’t able to handle sudden changes of load when a fast CPU switches back and forth from idle to turbo boost (or whatever Intel calls it). Such power transients likely happen again and again during Geekbench as it runs each different test.

    Repeated idle followed by all cores busy may not happen frequently in real-world use, but crashing at all is enough to ruin most server deployments I’d have in mind.

    I wonder if the system would be stable after decreasing the maximum CPU clock speed.

  10. @Chrisyopher – Your N100 is great.

    @Eric – we did not see it on the Linux-Bench scripts either. Those are pounding workloads over 24 hours. We only show a very small portion of those scripts here but those would be more short duration bursts and long bursts. There is something amiss, I agree, but it does not seem to be a simple solution.

  11. I wonder if Geekbench is doing something with the iGPU that stress-ng doesn’t touch.

    This whole thing looks like an interesting platform in completely the wrong case and cooling solution.

    It also reinforces that Intel (or someone) needs a low cost low power chip with a reasonable number of PCI lanes. Ideally ECC too, although that’s getting close to Xeon-D and its absurd pricing.

  12. The X520 dual port on PCI-E x4 with 2.0 interface is actually not able to run at full speed (can only perform ~14Gbps when both ports pulling traffic together), the older R86S with Mellanox is OK because the card is running at v3.0. This is a downgrade

  13. @Michael, normally it is the transceiver which takes care of that – most 10GBaseT transceivers can do 2.5 and 5 GbE besides 10GbE.

  14. Oh if only it had SFP28!

    (just kidding)

    Very interesting device with two SFP+ ports. I wonder if they have a new Alder Lake based NAS board coming out, so getting rid of the old stock.

  15. Topton has been having issues with these with the barebones customers. Customers are putting in off the shelf components and they struggle in the heat and hang. I am not sure Topton’s supplied RAM/SSD arent any better than anyones personal depot of parts, but these do run hot and the SSD should probably have a heat sink along with adding a Noctua fan.
    You have to admit, Topton and others have their ears to the home server/firewall market by adding the 10GbE ports. But to keep the prices down they are going through the back of the supplier warehouses looking for the right combination to complete the checklist.
    With as hot as they run, I would surmise they don’t have a very long useful life.

  16. I’d be interested in seeing packet per second benchmarks on these devices running pfsense or opnsense. Having an understanding of how well the platforms perform with a standardized routing or NAT workflow would be interesting.

  17. Something I’m missing in this review are detailed routing tests, since this device is clearly intended to be used as router.
    This means that many people would buy this to put pfsense on it, so it’s nice that you would include stats on that too.
    pfsense lan -> lan (2.5 and 10gbit), lan->wan (both 2.5 and 10gbit with firewall), vpn (ipsec, wireguard etc) speeds.
    More technical people will understand that due to 4x on pci-e 2.0 the 10gbit expierence will be mediocre at best, but non technicial people must also get this kind of info 🙂

  18. What units support Coreboot? What is the purpose of having a firewall that can be accessed at the chip level and defeated by a powerful adversary?

  19. I bought many small PCs for PFSense installations. All the rackmounts have failed after 4 years (PSU or SSD). The fanless were much more reliable but not recommended for non technical users because you never know if the unit is ON or OFF due to the LEDs sometimes inverted betweek disk and power ! It also by default try to boot on a not installed Windows with secure boot instead of PFSense. If you have customers using it on a remote location, it’s an issue. Also, too many PFBlocker rules can lead the unit to be too hot and to crash especially when stupid users put the unit in a closed cabinet !
    My policy is to use second hand DELL servers R210, R220, R230 with the IDRAC Enterprise option and a new SSD. You just have to choose a CPU with AES hardware acceleration in order to imrove OpenVPN and IPSEC speeds. The IDRAC option allows you to repair everything when a software update goes wrong: just ask to the user to plug his mobile phone in usb share connexion mode on a PC in the same LAN and you are done.
    Yes, a rackmount server is much more bulky than a mini pc but these DELL have half the depth of other models like R610 and the 3 Mhz or 3,5 Mhz CPU performs much better on single core tasks than the more expensive models.

  20. I try to order from AliExpress, and my order was canceled, the i7 is out of stock and they informe me that “it is expected that we will launch an 11th generation cost-effective CPU in mid-September, you can buy it at that time.”

  21. I wonder how hard it would be to put this thing in a little NAS case with 6×2.5″ bays and some fans… it’s pretty much ideal, but I definitely like the NAS option.

  22. @Noobuser

    Do you have any updates about this model? I’m looking for a future router with 10GbE SFP ports for home.
    Is stable? Is crashing as the previous ones in tests? Is getting hot?

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.