Ampere Computing Uncovers 80 Core "Cloud-Native" Arm Processor

[londiste]

That's not true lest you forgot Sofia, if you want I can pull up their numbers (billions of $ trying to cram it in the segment) & the models that actually featured in smartphones & of course tablets. Part of the reason they failed was the efficiency was way lower at that time, secondly baseband that were horribly uncompetitive & that's why they had to sell their 5G (modem) division to Apple.

Efficiency is two-part equation. Atoms were more powerful than ARM at the time but did consume more power. It didn't help that it took Intel a few years to get Atoms done proper. Sure, they used billions of dollars to try and shove Atom into the market and failed but this is always the case with established markets. Did Intel actually manage to get baseband into Atoms? When they called it quits some C-level person pretty directly said this market is not for them due to low margins.

 

[R0H1T]

Atoms were more powerful than ARM at the time but did consume more power.

That was till the time QC or Apple hadn't moved to 14/16nm & when they did it was game over for Intel. They actually had a process lead for a couple(?) of years with 22nm FF vs 28nm planar.

Intel a few years to get Atoms done proper.

I'm not sure what proper means in this context? It's not like Apple, ARM, QC were sitting still. Between ARM & Atom cores we've seen how far & how well they've fared in the years gone by. ARM has consistently delivered double digit IPC gains nearly every other year of the last decade. Intel's still not done anything close with Atom, albeit they had a couple of major jumps first with the transition to 22nm & OoO Silvermont(?) & the last couple of iterations going wider.

Did Intel actually manage to get baseband into Atoms?

For Intel's own chips I don't think so, I believe Rockchip did though.

This is Huge: Intel Enters Strategic Agreement with Chinese SoC Maker Rockchip

www.anandtech.com

 

[xkm1948]

Had some brief experience with an ARM based cluster node once, the institution brought it in due to the request from the CS department. Something wrong with backwards compatibility of software. It ended up getting scrapped in less than a year. It was mostly idling as there were crap ton of bugs

Unless they have a whole ecosystem behind their back, supporting hardware, software and maintenance, they won't get very far. This to me feels more like one of the flashy titles trying to get some name recognition.

 

[Assimilator]

The concept of putting Arm chips into servers was dreamed up in the bad old days when Intel was the only game in town regarding performance, and as such they could - and did - charge whatever they wanted. Hence an explosion of companies trying to make Arm for server a thing.

About the same time those companies launched their first-generation products - which were entirely unimpressive - AMD dropped Zen, and the prospect of Arm servers immediately died. Because Zen brought x86 prices back down to the point where the pain of adapting your entire company's software to Arm was no longer worth it. Fortunately for the Arm server companies, they had investors who hadn't yet realised they no longer had a feasible product.

Unfortunately for the Arm server companies, those investors have now woken up and are demanding returns. Unfortunately for those investors, the only way they're going to make a profit is if their chosen Arm server horse gets bought by a company big enough to need lots of s**tty, slow CPUs. The only companies that fit that bill are ones with thousands of s**ty slow VMs, aka the big cloud providers, and Amazon already bought an Arm server company, which leaves only Microsoft or Oracle - neither of which are likely to bet on non-x86 tech.

I'm expecting Amazon to dump their purchase in a year or so, and the other Arm server companies to follow shortly after. The value proposition simply isn't there.

 

[Vya Domus]

You guys cling to the fact that x86 has been here always and there is no way it will go away.

We cling to the fact that x86 has been used in high performance products for decades and ARM wasn't. It's as simple as that, ARM made the conscious choice to focus on the low power mobile industry and was very successful. Incidentally when Intel tried to go for that they failed miserably but they were very successful in the server space, notice the pattern ? It's as if you can't just change the paradigm overnight and make completely different products optimized for an entirely different industry.

It's not ARM per say, it's the fact that the IP and R&D that companies such as Intel and AMD have accumulated over the years cannot be matched by a new comer, it's just not realistic for something like that to happen.

But I will point it out for the last time : ARM itself puts little effort into trying to enter the server game. And that shows, there are a couple of small initiatives here and there like Scalable Vector Extension (SVE) meant to lift some of the burden of the people that want to licence ARM for high performance sever parts but it's no where near enough.

You have to realize that ARM doesn't make anything per say and technically doesn't even care where their IP is used but what they do does influence that and for the time being they are simply not focusing on that.

 

[aQi]

thats a leap ahead. Core count and clock is pretty impressive as per watt.