16nm ARMv8 SoC for servers and workstations | Hacker News Reader

16nm ARMv8 SoC for servers and workstations

anandtech.com

99 pointsflatroze24 comments

I'm of mixed feelings about this.

I'm in HPC now, and the thing I'm interested in is RIKEN's post-K supercomputer (http://www.aics.riken.jp/fs2020p/en/). Fujitsu is building it, and they're going to be producing the CPUs that will be going into post-K, just as they did for K. But, where the K supercomputer used SPARC, this new one is going to use ARM (https://www.top500.org/news/fujitsu-switches-horses-for-post...). Also, Fujitsu are working with ARM to develop HPC-specific architecture extensions to the ARMv8 spec.

Plus, as Applied Micro has been acquired by MACOM, I wonder how that's going to settle out.

So, I'm waiting to see what Fujitsu comes up with. The Post-K supercomputer will be a very good proving ground.

Thanks for the interesting links! But why the mixed feelings? Do you think vendors should hold off on selling a product that makes them money now to see what comes out of a single, highly specific project in five years?

The only advantage I see here is memory capacity. Well in the x86 landscape we have AMD giving you access of 1TB for less dollar then Intel. And if you include the COST of 1TB memory, the CPU dollar saving from a CPU is negligible. Not to mention the additional work that needs to port software to ARM.

I still dont see ARM going anywhere in server or workstation ( In the near future ).

8 DDR channels. Seems like everyone putting 30+ cores on a chip needs to learn this lesson. Memory bandwidth is the bottleneck. Intel uses HBM stacked on the Xeon Phi. Anyone who has the capability should do likewise or maybe try 8 DDR channels, pin count be damned.

This is what AMD's new Epyc server CPUs are rumored to have.

Quite more than a rumor at this point. The Eypc has 4 Ryzen dies, each of which on an on chip 2 channel memory controller.

http://www.amd.com/en-us/press-releases/Pages/amd-raises-exp...

This looks pretty decent in terms of hitting close to the Xeon 2680v4. But every server ARM play so far has been unobtanium until it's grossly out of date.

I might be excessively cynical, but I suspect that most of these ARM server plays are narrowly targeted at the biggest cloud/advertising companies (Facebook, Amazon, Google, etc.), with the publicity being meant more for investors than potential customers. I remember seeing some server architecture (Cavium ThunderX, maybe?) specifically being advertised as good for social media analytics. Are there more than 10 companies in the world that actually buy their own hardware specifically for that?

There was a pretty bad article about chip making posted a few weeks ago. The one great point it made was that the nm was not comparable across manufacturers. So, I guess my question is how big of a deal is this?

This is from March and nothing has been heard since then. Somehow they totally squandered their lead.

The performance for arm server chips is coming along.

I imagine some being ready to sacrifice performance and going with ARMv8 for them not being flawed with technologies such as Intel AMT.

Some are but not enough to support anything commercial - at least that's what history tells us.

People say they will pay for quality but the vast majority eventually only care about price.

I did a bunch of testing with ARM (ThunderX) for an IPC workload using this codebase: github.com/sargun/wat. Unfortunately, it no longer runs on ARM, because I'm using inline assembly.

The ThunderX's power comes from the fact it's a 48-core CPU, which is awesome for web workloads. The problem comes in that inter-core transfer is ridiculously slow. To make a 64-byte RPC (return / response), was in the area of ~1100 nanoseconds, whereas x86-64 was on the order of 500 nanoseconds on a 2 generation old CPU. Testing on a more modern CPU cut the time further.

I'd love to see ARM either catch-up in single-core performance, or multi-core performance, but until then I don't feel like it's viable.

Interesting! Cavium's network processors running bare-metal have features to ensure no core is ever idle. Maybe at the expense of latency. The idea probably doesn't translate to running a bunch of virtual servers. You probably do better with a shared-memory abstraction than a network. Or, disappointing numbers may just reflect the fact that individual cores are slow. Bad news either way.

According to the article this CPU as compared to the ThunderX has 80% faster integer performance for the whole CPU and 2x faster integer performance per core - so major a improvement:

http://images.anandtech.com/doci/11189/APM-X-Gene-3-performa...

Turns out that number crunching isn't what my CPU is doing most of the time. :-/.

I worry that the only thing they seem to be bragging about is stream, which is generally the worst possible measure of general performance.

When I went to the site, I got a fairly insidious takeover ad: The ad took over the whole page, displaying a 10-second countdown timer and a "Click Here to go to site" link. The "Click Here…" link triggered another pop-up ad (it never took me to the site), and when the timer expired, nothing happened. The URL bar had been changed, so reloading wouldn't help, and the back button took me to a blank screen.

Therefore, I present an outline.com link for the article: https://www.outline.com/cUUPVn

Anandtech has been a mainstay for professional hardware reviews and opinions for more than a decade... I actually just made a HN account to tell you that you might want to check your browser for malware because I've never seen anything like that on there.

They were bought out a few years ago.

Never seen anything like that. Ads, sure, but nothing that obnoxious. I suspect any ad network has the occasional mistake, but generally anandtech is pretty good.

Check your browser and device for malware

I got the same thing too. I think it's a paid ad on their website and not intention by them

Reply on news.ycombinator.com