Frank, thank you very much for your very detailed explanation! I checked it and pins and springs are OK. There are only 2 Pins for the cooler and you can move it so that the springs are become more tention. So it seems that it works as it should work and that there is some lash for the cooler. Temp is about 50° in my supermicro 505 rackmount case without active cooling in a full network rack. I took a better look at the standard shield from the case. and yes, it works if you open every inlay and leave it closed for displayport/hdmi.

Hi thanks for your input. I have now ordered following stuff: APU2C4 Router/Firewall in red chassis 60GB disk USB serial cable AC adapter Shipment is on its way. Looking forward to start learning, I expect to be a board member with a lot of questions to you guys!

@BlueKobold: And the IPMI ports let us say from the Supermicro Intel Atom C2000, C3000 or Xeon D-15xx boards are not nice too? If you don't need IPMI, it's just a waste of electricity. In a low power board the OOB management computer is a significant fraction of the overall power consumption.

@kklouzal: Fiber is a completely different story since FTTP (Fiber To The Prem) brings a standard fiber line into your home in which you can simply plug into an existing fiber channel card, get the PPPoE instructions from your ISP and away you go.  8) FTTN (Fiber To The Node) typically uses copper for the 'Last Mile' going from the distribution box. That last mile is either Coaxial or RJ11. In the case of RJ11 you can still use a PPPoE setup but you're already using RJ11 and that's going to be horrible in terms of speed and latency especially the further you get from the DSLAM.  >:( In a FTTN situation where Coax is used for the last mile you typically have an extremely short distance of copper (in terms of the entire connection length). The consumer has no control of anything past the drop towards their ISP. Basically you have a direct fiber connection except for this 'Last Mile'. Why not directly convert back to fiber, or as close to it as possible once it passes the drop back into our home/office? The Chelsio line of ASIC provides EXCELLENT latency, sub millisecond down to around 25 microseconds. The optimal solution would be to bridge the Ethernet PHY to the Ethernet PHY of a 'modem on chip' ASIC or something a little deeper and more integrated if you're not using a modem on chip solution. I guarantee you would receive at minimum an order of magnitude latency reduction as well as maxing out the line rate supplied from the your purchased plan.  8) :-* https://www.chelsio.com/terminator-6-asic/ [image: HklKIxC.png] In the crude design above we have our Chelsio T6 ASIC interfacing with the 'Modem On Chip ASIC' via one of its two provided ethernet interfaces. The secondary free interface is not wasted and we provide a SFP+ port to be made available for the user to do with as they see fit. The T6 ASIC then interfaces with the host machine and presents itself as two available NIC, one dedicated to WAN which has been coupled with the modem on chip. Throw a huge heatsync over the entire thing with a dual 40mm fans and away we go! Again, I think a card of this particular type design, for at least a first wave of exploring the real user demand for something like this, would be extremely attractive. This doesn't actually change anything compared to a bridge mode cable modem. In fact, it pretty much is a cable modem. That is just the issue here: all of the proposed "solutions" are just normal modems crammed on a PCI(e) card with a ethernet interface to the host. It doesn't change a thing, except physical form factor.

Just to be safe, download a config.xml from the WebUI. If the device crashes completely or the FS becomes corrupted, it's an easy way to get back to work. Something else: when reinstalling pfSense, some hardware requires a specific image to get going. You probably already used the right image: pfSense-memstick-serial-VERSION. If you didn't get a console after boot, you need to check the BIOS as well. pfSense just uses whatever the hardware has, but if it isn't configured properly it will just die as soon as it switches to the right console. If you have spare time, I'd suggest (after backup) trying to install or at least live-boot pfSense, just to make sure your hardware isn't about to die. The APU's usually last a pretty long time.

@Gektor: Hi, i need to add patch to support MSI interrupt in Atheros WiFi drivers, but i don't know how a can do it, please help me. Patch is here: https://patchwork.kernel.org/patch/9999249/ pfSense 2.4.1 This is a Linux kernel patch, not a FreeBSD kernel patch. This will require a C/C++ software engineer with FreeBSD kernel experience to port and make functional. Nobody in pfSense will fix this for you (and definitely not for free). That patch for Linux was done by Qualcomm, maybe you should ask them to also make a patch for FreeBSD. If you want to have more help with this, ask FreeBSD, not pfSense. Once it is in FreeBSD, it will also be in pfSense.

@wzochert: DeLorean, I've seen you mention that you've used the QC5000M in other threads.  Have you upgraded one to 2.4 yet?  It seems that would be a pretty simple and cheap way to solve the AES-NI problem with 2.5 when it comes out. I have build several firewalls with the Asrock QC5000M in a Supermicro 1U chassis, and these work also pretty fine without any problems, very silent, low power draw (arround 17Watt idle @ 220V) and…......AES support :-) Grtz DeLorean

While in Legacy mode i got [ 760 ] generic_netmap_dtor            Restored native NA 0 and finally i got some alerts on suricata  :)

This is indeed a driver and/or firmware issue with this chip and revision. If updating does not work your next best option might be a dual-port mini-PCIe NIC…

I tried downgrading to 2.4 and 2.3.4 with no success. The switch is a Cisco SG350XG-24F, so it has only 10Gb ports. When then interface finally goes UP, it synchronizes as 10Gb with no problem. I cannot try immediately your suggestions about mbuf and num.queue, as the device has to go back to production (I will use RJ45 links instead at the moment), but if I can I will of course try this !

from the Connecting to the serial console The default serial console settings in pfSense 2.2 and later are 115200/8/N/1, meaning: Speed: 115200 Data Bits: 8 Parity Bits: None Stop Bits: 1 Must be set into the Putty or TerraTerm or any other terminal program too! In pfSense is it the default setting.

I read through that a couple of times, and my understanding is that, while it does not prioritize VoIP traffic, fq_codel manages the queue more efficiently, so that the small VoIP packets are worked into available spaces, which means they're not waiting in the queue for other 'stuff' to finish. That's way over simplified, but is it an accurate way to describe it? @belt9: for VoIP, I would strongly recommend you read the linked thread in its entirety, then ask whatever questions you may have either here or in the traffic shaping forum. - playing with fq_codel in 2.4 In short, VoIP traffic suffers badly from high latency as I'm sure you know. pfSense just implemented (as a result on upstream FreeBSD developments) fq_codel. Why do you care? Because fq_codel is very easy to implement over your entire network, it does not require much tuning (usually none at all) and it will do an exceptional job of giving traffic such as VoIP very low latency connections even when your WAN is maxed out. It's currently a CLI-only implementation, but very easy to do (done in <5 minutes).

Thanks for the information. I will not go for this product then, had no idea this was their propose.

I saw an update online from the microsoft for that type of chipset.  From 2010!

I think I've decided to move the C2758 to Site B and use the i3-6100 at Site A.  Will report back my results when this is all up and running.

Can you post the following from console command prompt: usbconfig (find ugen for the modem then do a dump_device_desc for the VID and PID) ls /dev at!entercnd="A710" OK at!udusbcomp? !UDUSBCOMP: 8 OK at!udusbcomp=? 0  - reserved                                    NOT SUPPORTED 1  - DM  AT                                      SUPPORTED 2  - reserved                                    NOT SUPPORTED 3  - reserved                                    NOT SUPPORTED 4  - reserved                                    NOT SUPPORTED 5  - reserved                                    NOT SUPPORTED 6  - DM  NMEA  AT    QMI                        SUPPORTED 7  - DM  NMEA  AT    RMNET1 RMNET2 RMNET3        SUPPORTED 8  - DM  NMEA  AT    MBIM                        SUPPORTED 9  - MBIM                                        SUPPORTED 10 - NMEA MBIM                                    SUPPORTED 11 - DM  MBIM                                    SUPPORTED 12 - DM  NMEA  MBIM                              SUPPORTED 13 - Config1: comp6    Config2: comp8            NOT SUPPORTED 14 - Config1: comp6    Config2: comp9            SUPPORTED 15 - Config1: comp6    Config2: comp10            NOT SUPPORTED 16 - Config1: comp6    Config2: comp11            NOT SUPPORTED 17 - Config1: comp6    Config2: comp12            NOT SUPPORTED 18 - Config1: comp7    Config2: comp8            NOT SUPPORTED 19 - Config1: comp7    Config2: comp9            SUPPORTED 20 - Config1: comp7    Config2: comp10            NOT SUPPORTED 21 - Config1: comp7    Config2: comp11            NOT SUPPORTED 22 - Config1: comp7    Config2: comp12            NOT SUPPORTED OK

For an AM3+ system? I'm not really aware of too many boards with onboard graphics, but tossing away one of the x16 slots on my ASUS M5A99FX PRO R2.0 that I'm using for a FreeNAS build (I already had it and it supports ECC) doesn't hurt so much, especially since I had an extra Quadro 600 card lying around that doesn't consume the slot immediately adjacent to it like many GPUs do.  My board has 4 x16 (physically) slots; 2 of them actually have 16 lanes and the other 2 have 4 lanes.  I'm using one of the x4 slots for the GPU; that leaves 2 x16 and 1 x4 for other devices.  One slot holds a x4 Intel NIC, one of the x16 slots has an x8 SAS HBA in it, and there's still one free, even with the GPU in there. I used to have that same Asrock board as well, it's a fine board. I ran 3 GPUs on it and it didn't break a sweat. There's no reason NICs in those slots would put any undue strain on it.  If you need to dedicate the multi-lane PCIe slots to NICs, I'd look at picking up an old PCI graphics card.  There's also at least one PCIe x1 GPU out there IIRC.  You don't need a lot of bandwidth for the console on pfSense. And finally, you could also look at consolidating NICs by picking up a server pull quad NIC or two. That way you could have 8 physical ports in just 2 PCIe x4 slots. Hope this helps. EDIT:  Just re-read and realized you have 10Gbps cards in there.  So disregard the last bit.  If you really want a quality AM3+ board the M5A99FX PRO R2.0 is great.  2x16 and 2x4 all in x16 physical configuration.  Plus a x1 and a legacy PCI slot for good measure.  Enough for your 10Gbps cards and a GPU.

Network interface devices names will probably change too.

Hi! @pldelisle: Or maybe I could only add an Intel Quad port i350 board to my actual « server » for 80$ CAD on eBay? I know these are made in China… are these boards as good as the genuine ones? I saw this one Some people say that they are, most I believe say that they are not and that the parts and design are of lesser quality. The card you posted is definitely meant to deceive… While the main chip is quite likely a real Intel chip (because Intel sells them to third parties), the Delta chips are clearly fake. Delta is supposed to be embossed, not printed like it is on the one you posted. What I would try to get if I were you would be an HP, IBM or other OEM version of an I340 or I350, preferably new and hopefully from a Canadian (which I assume you are since you used an ebay.ca link) or US seller. If the card is a real one it should look nearly identical to a real Intel card with some possible substitutions. I have both a know real Intel I340-T4 bought full price and from a known legit place and its HP version bought off eBay… They essentially look the same with some subtle differences. The Delta chips on both are embossed which is a good sign. A thread on servethehome (https://forums.servethehome.com/index.php?threads/comparison-intel-i350-t4-genuine-vs-fake.6917/) suggested some of the differences between my cards  were a sign of a counterfeit card but I do not fully agree with they said. They were of course comparing I350s together and I am comparing I340s but I don't believe it makes much difference. The clock crystal is different. The known real Intel has a TXC like servethehome said a real Intel card has but my HP version has a branded clock crystal from a known Intel supplier. I believe what is important here is to have a branded clock crystal (not a no name like the counterfeit cards) preferably from a known Intel supplier. The inductors are also different between both card but this were things get funny… My known real Intel card has the inductors they say are fake while my HP version of the same card has the ones they say are OK… I don't think you should say that a card is a counterfeit solely because less important components (like the clock crystal and inductors) are different, Intel uses different suppliers for some of those parts so they are bound to be different sometimes. You should however always expect to have real embossed Delta chips and things such as clock crystals should be branded, not no name parts… Good luck and have a nice day! Nick