Anyone wants AArch64 server?
9 plus ones
Shared publicly•View activity
View 41 previous comments
- and others, thanks for your discussion. It opens my mind. That's I never thinked about before since I entered into ARM server area.
Just one more question. :-) It's about ecosystem.
Let's say If we find a manufactor who have ability and willing to build a "ARM PC". They can build the motherboard and case. That's they can control by themselves. Then they need to spend $?0K to buy UEFI sourcode and spend some resource to develope specific UEFI for their motherboard. They also need to ask Redhat, Canonical, Suse, etc to support their motherboard's driver in official Linux release. They need to test different type of PCIe extention cards. They probably will find that some cards, like graphic card, sata controller don't support their platform. And these card vendors would say"sorry, we don't support ARM platform. "These issues are what I observed right now building ARM server. It's like a chicken-egg problem. No enough supports from these ecosystem, people can't ARM platform. But if no enough people use it, these ecosystem will not be mature. What's your point?49w
- How about 96boards' enterprise board? Does it meet your requirement?49w
- not really, 96Boards EE has two options, weird and uATX. The current boards are all weird, for no good reason.
You said a manufacturer would need to make a case: I'd say no to that. They just need to make a standard PC form factor like uATX. Standard ATX power header. PCIe, SATA, USB, Ethernet and video (HDMI or DisplayPort) and audio if the SoC can do it. Headers on the board for PC case front panels, and fans (even if they're not totally needed). Just like a PC board.
Users can buy a case, drives, RAM, PCIe graphics/USB/wifi/extra storage and go nuts. If there aren't enough USB 3.0 or there aren't enough Ethernet get a PCIe card, that kind of thing.
SD cards on board are optional, as are mPCI or mSATA or other desktop/laptop standards. It doesn't need to be "unique." Just like servers will do better if they're not unique.
UEFI is open source! It'll cost engineering time but getting Insyde, AMI, Phoenix or some other vendor to do a port would work too. Most of the suitable chips are server or network oriented anyway so the work is already done.
Done properly it'll boot RedHat off a USB or DVD out of the box, and optionally boot over network (PXE/TFTP/HTTP) as this is all in UEFI already.
There'll be no funky Special SD card Image or requiring magical scripts to make it work or build your own filesystem and kernel - although you could if you wanted to.
In business there's no such thing as the chicken-egg situation. It's supply and demand. We know there is demand - perhaps not for 40,000,000 units like a cell phone but enough to hit economies of scale. And if it does that then another one can be designed, others will enter the market..49w
- also, the vendors that "don't support ARM" for PCIe expansion will probably not support Linux either so what's the difference? I've tested many PCIe cards in ARM systems and they have all worked fine from a device standpoint.
The real issues are entirely down to silly kernel/X bugs like not realizing that you want PCIe graphics as primary framebuffer (thanks +Marcin!) or no drivers available at all. But USB, SATA, graphics, networking, nothing has not worked so far if there was a Linux driver.49w