Hacker News Reader

Comment by c0l0 | Hacker News Reader

Finally one can buy these! The sibling model, ASRock's DeskMeet X600, should exhibit a roughly equal power consumption profile, but offers support for proper ECC UDIMM in up to 4 slots, which makes it a premier choice for a cheap, powerful, reliable and flexible (SOHO) server platform.

Even the previous generation - with AM4 CPUs and DDR4 memory - of these ASRock offerings were easily enough for most users' needs (my wife still has a DeskMini X300 with a Ryzen 5 5600G as her daily driver).

View on HN

Do you know if the motherboard has pins for a discrete TPM 2.0 module, e.g. https://www.asrock.com/mb/spec/product.asp?Model=TPM-SPI & https://download.asrock.com/Manual/TPM-SPI.pdf

Does a discrete TPM offer specific advantages over an fTPM? I suppose it'd be modular but I can't really imagine trusting AMD to run the entire operating system but then saying they can't be trusted to hold a key and being fine with that in the same system. Not that I'm saying they are therefore perfectly secure rather I'm not sure what risks profiles it really avoids so perhaps other advantages exist too?

Sadly, AMD PSP-based fTPMs emulate only a subset ("Mobile TPM" spec) of a discrete client TPM. This could be addressed with a PSP software update, since the Microsoft reference code for TPMs is open-source.

A full TPM implementation is required to use DRTM (SKINIT) for measured Linux Secure Launch, which is useful for home servers to confirm that the boot loader, kernel and optional immutable OS/VMs have not been modified. Unlike "Secure Boot" SRTM, DRTM adds launch integrity with policy defined by the hardware owner. One would typically use SRTM+DRTM.

The TPM emulation in Microsoft Pluton, which is present in some Ryzen chips, might be sufficient for DRTM. It's tragic that AMD fTPM has limited functionality, since AMD provides SKINIT on all SKUs, unlike Intel where TXT DRTM is limited to vPro.

Some ASRock 4x4 (NUC-sized) mini PCs have pins for discrete TPM.

Generally the other way, if you're paranoid -- the fact that it's separate component allows an attacker to MITM the communication.

Not according to the spec sheet or the PCB photos[0][1] they provide, no.

[0]: https://www.asrock.com/nettop/photo/DeskMeet%20X600%20Series... [1]: https://www.asrock.com/nettop/photo/DeskMeet%20X600%20Series...

The DeskMeet X600 looks rather interesting especially re. the ECC RAM.

You don't happen to know of anything similar with more than 2 x 3.5" drive bays per chance? I'm looking for a low powered device with ECC RAM to use as a NAS.

I've been looking at HP Microserver Gen8 boxes, which I think are fairly low power, but I believe they're rather old now.

I think you will need to go for a custom (i.e., one you buy without the mainboard pre-assembled into a complete barebone/system) case to get that. Jonsbo has a few probably interesting options for you, such as the N3[0][1]. I don't know if you could make the DeskMeet mainboards fit in that particular case - but many regular ITX boards for AM4 and AM5 come with ECC support, too.

[0]: https://geizhals.eu/jonsbo-n3-schwarz-n3-black-a3062571.html [1]: https://www.jonsbo.com/en/products/N3.html

Thanks for your links and information re. AM4,5.

> low powered device with ECC RAM to use as a NAS

QNAP has a $600 1U short-depth (11") 4x3.5 2xM.2 2x10GbE 2x2.5GbE 4-32GB DDR4 SODIMM Arm NAS that would benefit from OSS community attention. Based on a Marvell/Armada CN9130 SoC which supports ECC, it has mainline Linux support, and public-but-non-upstream code for uboot. With local serial console and a bit of effort, the QNAP OS can be replaced by Arm Debian/Devuan with ZFS. Rare combo of low power, small size, fast network, ECC memory and upstream-friendly Linux. QNAP also sell a 10GbE router based on the same SoC, which is a successor to the Armada 388 in Helios4 NAS (RIP), https://kobol.io/helios4/

In theory, ex-Apple Qualcomm Nuvia/Oryon Snapdragon boards would make a great NAS, but it may take a while for Linux support to mature. Upcoming AMD "AI 300" Zen5 Strix Point mobile chips are intended to compete with Qualcomm, so they can hopefully be configured with low (15W-54W) power consumption, and ECC support has been good on AMD devices when the BIOS cooperates. Intel Lunar Lake has low power, but Intel ECC support is market segmented to Xeon SKUs, except for the anomalous Coffee Lake Celeron G4930T (35W TDP).

To decouple storage enclosure logistics from the NAS CPU, a mainboard with Thunderbolt or USB4 enables an external NVME enclosure with DIY bridge of multiple SATA ports or PCIe HBA to JBOD.

Thanks for that information re. the ARM based QNAP device. I'd not heard of that before. Would you mind pointing me to the u-boot code you mentioned?

I think I'm right in thinking https://ukstore.qnap.com/catalog/product/view/id/215 is the machine you meant?

Yes, TS-435XeU, forgot to link it.

I've not yet built uboot myself, but these repos claim to have code for CN9130.

https://github.com/SolidRun/cn913x_build/

https://github.com/SolidRun/u-boot/tree/u-boot-v2019.10-marv...

https://solidrun.atlassian.net/wiki/spaces/developer/pages/3...

Oh I clicked on this thread specifically to ask about older generation models, thanks for pointing the name.

My needs are more modest for one particular project so I don't see why I should pay for ddr5 or pcie5 ssds.

Say, do these things support setting an even lower power limit in the bios?

Afaik, that's a matter of the AGESA revision in the firmware and the CPU SKU. In ther real world, I think it actually matters little, since setting software power limits in the OS will be as effective as setting a low-power (35W or 45W) profile in the system firmware in my experience (which is limited to AM4 only so far, however).

Reply on news.ycombinator.com