A real ATX-ish ARM64 workstation

[konradybcio]

To date, there are only 3 meaningful types of performant and readily available ARM64 desktop workstations:

- M1 Mac Mini / iMac

- Nvidia Xavier NX/AGX

- Honeycomb LX2


All other "look at me, I haz arm" devices are either laptops, puny tablets trying to imitate laptops or plain devboards with an Android image or a quarter-functional, outdated Debian..

They mostly use smartphone chips, sometimes higher frequency-binned and with a more liberal power limit. And they lack HW documentation. Or they are discontinued.


What is truly needed to kickstart the ARM64 ecosystem further is a good desktop workstation. 

In such a board, I reckon there should be:

• some kind of ATX standard form factor to mount it in a regular PC case with a regular PC power supply
  
• socketed RAM
  
• PCIe, USB, SATA, 3.5mm jack, RJ45..
  
• a power and a force reset button
  
• a chip to keep the bootloader on, so as to make the OS installation simpler
  
• targetting Arm SystemReady certification, which given enough work on proper UEFI, would allow for installing OSes like on a normal PC, from a livecd etc..
  
• an OSS-friendly SoC that can heat up as much as it wants - performance is key in this form factor! (without a broken PCIe implementation, I'm looking at you, Raspberry Pi...)
  
• Intel-/AMD-compatible mounting holes for a cooler
  
• DC/PWM fan headers (may be on a separate controller)
  

I was thinking, perhaps Pine could be a good vendor to tackle this? Of course, this is a complicated problem to solve, but discussion is a good thing.

[moonwalkers]

• some kind of ATX standard form factor to mount it in a regular PC case with a regular PC power supply
  
• socketed RAM
  
• PCIe, USB, SATA, 3.5mm jack, RJ45..
  
• a power and a force reset button
  
• a chip to keep the bootloader on, so as to make the OS installation simpler
  
• targetting Arm SystemReady certification, which given enough work on proper UEFI, would allow for installing OSes like on a normal PC, from a livecd etc..
  
• an OSS-friendly SoC that can heat up as much as it wants - performance is key in this form factor! (without a broken PCIe implementation, I'm looking at you, Raspberry Pi...)
  
• Intel-/AMD-compatible mounting holes for a cooler
  
• DC/PWM fan headers (may be on a separate controller)
  

• ...
  
• socketed RAM I hear is tricky because of controller and initialization. AFAIK in x86 systems firmware plays big role in this.
  
• ...
  
• ...
  
• there is already SPI flash in existing Pine64 products. It seems the trick there is to find the right firmware - e.g. right now it's mostly u-boot, and not all versions of u-boot are fully functional. E.g. u-boot on PBP only relatively recently gained ability to use built-in display and keyboard, and power management (i.e., deep sleep) is still hit and miss, depends on the right combination of u-boot and kernel.
  
• That would be awesome. Again though, this is a sizeable investment into firmware development. When even "simple" u-boot can't be mainlined ahead of or at least shortly after product release, I wouldn't hold my hopes high for something like Arm SystemReady certification. And considering up until now Pine64's modus operandi has been "we develop hardware, community develops software"...
  

Pine64's approach has been to develop devices as cheap as possible, so that as many developers as possible can afford them. What you suggest is the opposite approach, something more similar to Tesla's approach (when they started with Roadster, then proceeded with Model S and Model X, moving on to Model 3 and Model Y, and now teasing a mere $25k model). Will it work? I don't know, but I can see why Pine64 would be cautious - unlike cars that users can just start driving right away, computers are useless without software. If I'm not sure whether all the software I want to run will work as well on arm64 as it does on amd64, I'd be much more reluctant to invest into a costly ARM machine no matter how high its performance is. In fact, that's exactly the reason why I still haven't taken the plunge with HiFive's Unmatched.

And by software I mean even something as basic as getting the OS to run with full functionality. On my original Pinebook running Debian I still don't have working Bluetooth. Or any sleep, deep or s2idle. Even getting sound to work was a pain in the ass. But that's OK, it cost me only $89+shipping. Before COVID-19 I used to spend more on dining every couple weeks. When I couldn't get it to work properly even with BSP I ended up just tossing it into storage bin until better times. $200 for PBP is still pretty cheap for a laptop, but I was reluctant to pay even that much until I saw in December 2019 that it is already getting into a better shape than original PB ever was in. Had it been more expensive - I probably would've never bought it until I saw full mainline support (which is still not quite there yet). But now that I did - I learned a bit about ARM as a platform. Enough to pull the original PB out of storage to give it another try, and have a much better success with it than the first time around.

I don't deny there are the most enthusiastic of the enthusiasts out there who are willing to buy a high-cost computer on a sheer hope that they will be smart enough and free enough (time-wise) to get the firmware and drivers for it in shape by themselves. But those are not the majority. So unless Pine64 is willing to commit own resources to developing open source firmware and drivers for their hypothetical high performance Arm workstation - a much safer bet for them is probably to build mind share through sheer numbers of lower-priced devices, which is what they have been doing.