Benchmarks Of The 24-Core ARM Socionext 96Boards Developerbox

[grok]

Not bad for a developers box. I sometimes think people mis the point of developers boxes They are for software development. As for the poor processor performance the cores are rather old, this is not a new processor implementation. What would be really interesting is to see a pin compatible chip come out based on AArch64 or what ever is the latest at the time.

It is AArch64, that's why the A53 cores still are everywhere.

There was the Sun Niagara before, which was an 8 core when single core on PC was still a thing. It's slow, makes you wait when you're unpacking software and that uses one thread. Not your 3GHz Intel. It had about the same use case. The servers that serve your downloads, including what we call movie streaming don't need CPU power at all. It just processes requests and sends bytes over the wire. Sure a quad core 3GHz will probably do it but with many processes or threads lightly loaded you can get away with 24 cores at 1GHz (and slower than x86 1GHz) and save a few watts and some space.

Micro ATX dev board is a low volume product, likely much more expensive than the real server hardware and also bigger.

[Anty]

So what is the purpose of this board? It should at least shine in multithreaded tasks but it is heavily beaten by lowest end x86s - where CPU + board is less than $200. Perf / power is also not that great when you take into account x86 CPUs could be undervolted and underclocked and will still win in raw performance.

[wizard69]

Not bad for a developers box. I sometimes think people mis the point of developers boxes They are for software development. As for the poor processor performance the cores are rather old, this is not a new processor implementation. What would be really interesting is to see a pin compatible chip come out based on AArch64 or what ever is the latest at the time.

[NateHubbard]

Bwaha, here come the "desktop performance" ARM chips! I mean, it's just ridiculous How could anyone take these promises seriously. The only bigger gimmick is the ARM servers lol. Most laughable concept ever.

These are low end cores even for a cell phone.
If you're not just trolling, you're reading too much into this.

[ypnos]

Bwaha, here come the "desktop performance" ARM chips! I mean, it's just ridiculous How could anyone take these promises seriously. The only bigger gimmick is the ARM servers lol. Most laughable concept ever.

Server is actually the only use I see for such a system and the results look very promising for that. If the whole system under load only takes < 15 Watts, think about that. The NVME is already taking in a third of that when busy. In the server business power usage is cost factor #1. This could be a really great web server. It is great for serving concurrent requests and it comes with the needed IO capacity. Or as a proxy, this news is a bit old but might be a surprise for you: https://www.datacenterknowledge.com/...center-network

There is much more than HPC in the server market.

[Tomin]

Bwaha, here come the "desktop performance" ARM chips! I mean, it's just ridiculous How could anyone take these promises seriously. The only bigger gimmick is the ARM servers lol. Most laughable concept ever.

You should realise that this chip has power envelope of an Atom chip. It uses five watts not sixty like those Core i processors (and the Ryzen) it was compared with. It is hardly a representative of ARM in the category of desktop PC processors.

I'm just saying that you can't draw such conclusions from this comparison. It's not apples to apples so to speak.

[anarki2]

Bwaha, here come the "desktop performance" ARM chips! I mean, it's just ridiculous How could anyone take these promises seriously. The only bigger gimmick is the ARM servers lol. Most laughable concept ever.

[c117152]

Performance per Watt seems be the only viable metric for ARM

One of their newer 1u 5w racks is designed with real time 32streams x 4kp60 media transcoding: https://www.socionext.com/en/products/assp/SC2A11/ https://www.socionext.com/en/product...loud-solution/

You'll need a few 1200$ Threadrippers / i9 / GPUs and a whole lot in utility bills to cover that much. So, it's Performance per Watt per Dollar depending on the SoC. Though the one here (SC2A11) might be a little limited, it should be appropriate if you have a lot of security cams streaming to a remote server or something of the sorts.

[Weasel]

Obligatory comment from oiaohm about RISC-V superiority and massive performance here.

[reavertm]

Performance per Watt seems be the only viable metric for ARM, too bad due to HW being remotely tested, it's not available.