Sipeed NanoKVM-PCIe
cnx-software.com
cnx-software.com
Apparently it's the Bitmain cryptominer folk? Nice context from Tom's.
https://www.tomshardware.com/tech-industry/artificial-intell...
I'd love to see something like this where the board had a basic video card, so you could use it in a system without any video output. Bonus if it also had a usb controller and a serial port, so it didn't need to loop to plugs (although some of that could happen on the internal side as well)
I have never been able to identify a technical barrier to doing this, the important features most people actually care about are implemented over a 1x PCIe link and USB, plus a couple of GPIOs to twiddle the power/reset button connections. Most OEM implementations also connect to the LPC bus and others on the server board to allow more in depth diagnostics, voltage logging, etc. but those are bonus features and not requirements for a useful product. I do not see any technical reason a useful generic PCIe implementation couldn't be produced, and as a result I have a strong feeling that the lack of such products is an intentional choice by one or more of the vendors involved to increase margins by pushing users who want these features up to entry level server boards instead of sticking a card in a higher-end desktop board that might better fit their needs.
I have one, it works. Look at that board though! Given that component and board density, I don't find it very surprising that it costs around 200$ USD. Maybe you could get clever and design one that tries to work 100% via pcie device functions, and cut down the (physical) complexity of the IPMI device, but then the wide system and interface compatibility of the IPMI board would suffer.
To wit, the card needs to do the following, at a minimum:
1. Provide a VGA/UEFI/VFB-equivalent framebuffer for display capture
2. Provide a USB-device interface to emulate keyboard+mouse.
3. Provide a network interface of some sort for remote control
4. Provide GPIO headers/sinks at the appropriate power levels to replace traditional motherboard front-of-device switches -or- provide a PMBus header to perform equivalent functions with a support PSU.
You can get away with a subset of the above, but without that, you will be missing something critical to enabling lights-out-management. On top of that, a full-featured IPMI board is also generally expected to provide some fashion of environmental monitoring and fan control, which the PAUL does provide. It throws in some extra features as well, and provides some maintenance/lifecycle management capabilities of the board itself.
Providing all the above, is about the same amount of complexity of shipping a SBC and ecosystem itself, made moreso by the compatibility requirements.
You don't have nearly as big a market willing to pay $200 for all that -- its certainly not a matter of adding the chip and flash and dropping some vias.
Even having a standardish system-on-chip and non-bespoke firmware is 'new' (post-2020?). It could have been done 10-15 years ago sure, but noone was willing to do the long tail of tasks needed to get a product to market yet.
I was aware of these but I could have sworn when they came out they at least were only officially supported on ASRock Rack motherboards. That doesn't seem to be the case anymore, or I'm just entirely misremembering.
They do still seem to have the standard ASRock Rack problem of being very interesting to "homelab" and other small volume users but not exactly straightforward to obtain. No one who'll let me buy a single unit brand new has them in stock.
Definitely set an auto-notify on Newegg now that I know they work with generic motherboards, though it's a fair bit more expensive than a NanoKVM it's a lot more appealing to me to actually be a complete standalone internal device that doesn't rely on external wiring to function. The various "PCIe" versions of the PiKVM and its ilk still require an external HDMI connection and sometimes even external power.
That's what they claim. Until another binary closed source firmware upgrade arrives, of course.
and all those things are another OS which you have zero visibility or control, but have to manage just like a production server. there's probably tons of ipmi backdoors all over the place and nobody cares much besides limiting routing to some lan.
Their updater for the standalone unit has all sorts of bug reports.
I cannot figure out how these manufacturers can produce such nice hardware and such horrible software. And repeatedly, repeatedly miss the mark on understanding how important kernel/OSS releases are.
It would also be interesting to make the PCIe card a good enough NIC (please use an effectively universally supported chip) to avoid the loopback cables.
About 99% of the time I want console access I would vastly prefer a proper serial port (≥115200 baud, 8N1, hardware flow control and break condition). Make that and power control (status, on, off, warm reset, cold reset) available via SSH over WireGuard. If done well it requires a lot less bandwidth and is more snappy. While I'm wishing for things please add mosh too ;-).
https://en.wikipedia.org/wiki/NC-SI
Unfortunately, support for NCSI can't be added to most motherboards; it requires the NIC to support it.
The problem with a pcie one for me is that modern motherboards suffer from having hardly any PCIe ports - and when they do they're mashed in close to each other essentially making one useless if you have a decent GPU.
I can see a great deal of trouble capable of coming from a networked device capable of watching the screens 24x7 and potentially intercepting passwords being entered. And those are the legitimate functions for this device. Wouldn’t take much to throw a reverse shell for external access if you wanted to be particularly nefarious.
Not saying there’s any evidence this kvm is malicious. But I probably wouldn’t put it in anything more than one of my toy home lab servers.
For remote access, there is no cloud. But you can BYO tailscale or FRP [0] (note: I really like the FRP idea, as it's trivial to self-host)
For updating, there is a central server. But at least the process seems to be manually-initiated [1].
I am not saying the firmware is backdoor-free, but at least it would be feasible to monitor/block all outgoing network connection attempts, and still have a functional device.
[0] https://wiki.sipeed.com/hardware/en/kvm/NanoKVM/network/tail...
[1] https://wiki.sipeed.com/hardware/en/kvm/NanoKVM/system/updat...
About this KVM, I really like the Tailscale access ability, though I'm worried about the binary blob.
Also, is there Windows / Mac compatibility?
And yes it works fine on windows. I've got one. Haven't tried it on Mac yet though.
Would you count on a Chinese supplier fulfilling their promise given the past history of relationships with open source?
uses exactly the same firmware image as the little usb ones. i also have one of the "full" and "mini" usb ones.
have connected it to the motherboard headers for power, reset, leds etc... including usb directly to motherboard header. documentation is pretty decent, was quite easy.
would still be better if it contained a usb hub on the pcie port, but they've basically just modified their existing design for poe/pcie power in addition to the preexisting usb, and slapped it on a card.
as an homelabber, i'm using HP's iLO on my gen8 microserver for example.
Edit - ChatGPT to the rescue:
> KVM over IP (Keyboard, Video, Mouse over Internet Protocol) is a hardware solution that allows administrators to remotely access and manage servers and other devices at the BIOS level, even when the operating system is not running. It provides complete remote control, enabling tasks like system reboots, troubleshooting, and OS installations as if you were physically present with the machine.