Benchmarks Of The 24-Core ARM Socionext 96Boards Developerbox

[boboviz]

Not bad for a developers box. I sometimes think people mis the point of developers boxes They are for software development.

All developers i know use Raspberry 3 or Arduino. For less than 50 USD.

[-MacNuke-]

*People say it doesn't have branch prediction at all, but the technical reference manual for it says otherwise.

It does do branch prediction only for instruction fetches. But since it does not execute them it is immune to Spectre and Meltdown.

[L_A_G]

I do find their choice of ARM core to be a little odd seeing how the A53 is supposed to be a simple low power core for mobile devices. We're talking about a core with branch prediction so simple* it's supposedly not even vulnerable to Meltdown or Spectre because of the branch prediction being to spartan. You'd think they would be spoiled for choice when it comes to available cores.

*People say it doesn't have branch prediction at all, but the technical reference manual for it says otherwise.

[coder]

I don't know all boards but does the X2 has multiple PCIe slots and removable RAM?

https://developer.nvidia.com/embedde...son-tx2-devkit

If you need a x16 slot or more RAM, how about this?

https://www.phoronix.com/scan.php?pa...RM-In-My-Hands

Anyway, I'm not saying absolutely no niche exists for this board. I was just responding to wizard69 's unqualified claim that it was a "Not bad for a developers box". Yes, it is bad for a generic developer's box. Better than a Pi 3, but that's a pretty low bar.

[-MacNuke-]

Yeah, but $1200 for this? I'd rather use a Jetson TX2 for half the price. On any low thread-count workloads, the TX2 is way faster.

I don't know all boards but does the X2 has multiple PCIe slots and removable RAM?

[coder]

If you want to develop and test applications and drivers for ARM an i3 gets you nowhere.

Yeah, but $1200 for this? I'd rather use a Jetson TX2 for half the price. On any low thread-count workloads, the TX2 is way faster.

[-MacNuke-]

Yeah, it kinda is bad. Look at its kernel compilation benchmark, in comparison to a dual-core i3! Not to mention that developers are often doing incremental builds where you actually want decent single-thread performance.

If you want to develop and test applications and drivers for ARM an i3 gets you nowhere.

[LoveRPi]

The 25% high performance core to 75% low performance core makes sense in most applications. The Socialnext box is sadly 100% low performance cores. If you look at what Mediatek did for the X20 with 2 high performance, 4 low performance high clock, 4 low performance low clock, it was a pretty well thought out design but they couldn't design a scheduler to utilize such a layout since all of the existing schedulers are pretty bad for asymmetric core performance.

[bobolopolis]

Lot's of people are wondering what this board is for, and since I'm the person that offered Michael access to the board he benchmarked, I figured I'd chime in.

The purpose is right in the name. It's a developer box. Obviously that means different things to different people. For me, this kind of board is fantastic since it provides a native aarch64 development environment with reasonable performance. I primarily work on embedded systems where cross compiling is the norm. As much as I love stuff like Yocto for building images, doing day to day development with cross compilers gets annoying. Boards like this give a developer the ability to pop in an Ubuntu USB stick, install a desktop, and be able to test and develop in a native environment without dealing the all the cross compiler complexity. You can do some of this work in an emulator like qemu, but frequently it's nice to just work on stuff natively.

This particular board supports EFI, so your standard Fedora, Debian, or Ubuntu distros work without much fuss. If you've ever dealt with vendor dev kits that require custom bootloaders and kernels, you'll know this is a huge time saver. Poorly supported vendor forks of uboot or the kernel are one of the biggest annoyances I have when starting work on a new hardware platform.

More broadly, Socionext is mainly using this CPU in servers with thousands of cores, focusing on the performance per watt. If you're doing something that scales well with the number of cores, such as video encoding, this can be a great way to do it. This isn't my particular use case, but I'd expect it to work fairly well.

My main complaint is obviously with the single thread performance, but this is expected from a Cortex A53 core. But even with the weak cores, there are enough of them to make this a nice build server and test platform for native development.

[coder]

Not bad for a developers box. I sometimes think people mis the point of developers boxes They are for software development.

Yeah, it kinda is bad. Look at its kernel compilation benchmark, in comparison to a dual-core i3! Not to mention that developers are often doing incremental builds where you actually want decent single-thread performance.

The only benefit for developers is if they're specifically researching or optimizing multicore scalability of their software. For that, the core/$ ratio isn't too bad. Especially if you don't want a big, hot, loud workstation or server.