Marvell plans 40-nm ARM server processors | Hacker News Reader

Marvell plans 40-nm ARM server processors

eetimes.com

22 pointswmf9 comments

I hope they don't wait until Microsoft launches Windows Server 2014 Enterprise Edition for ARM Servers...

Seriously. We want that tech now and we have enterprise-grade OSs to run on them.

Is it really such a great deal? Enterprises usually pay top dollars for top Xeons/Opterons because they get a LOT of perfomance of a single chip, single motherboard, single rack. If they wanted to save on CPUs they could easily buy sub-100$ quad cores. I somehow can't believe that it will be such a good deal - there has to be something (cost, complexity and low throughput of interconnects?) that made Intel's engineers not to choose this way.

Yeah I don't quite get that either.

Apparently Marvell thinks that there's a market that's being overlooked for systems with lots of low-power, highly-efficient cores.

I guess I can see how that might work; it would be competing less with big Xeon or Opteron boxes than with blade systems. I've seen companies that for whatever reason are not comfortable with the idea of virtualizing multiple logical servers on one big physical server, but instead have lots of high-density blades. Different way of thinking. Except that many of them are running flavors of Windows on the blades, I could see some sort of ARM SoC based solution having some appeal.

Let's say you have a 4-core x86 running 20 processes. Imagine you can have, for a fraction of the operating cost, the same performance out of a 16-core ARM chip.

You get a lot more performance per thread on a Xeon/Opteron, but you could get about the same, if not more, aggregate performance from a different chip if you match your processor architecture to the workload. Another example: imagine doing lots of floating-point on x86s compared to doing it on GPUs. Even if GPUs require another processor to control them and keep them busy, the GFLOP per dollar is much more favorable to them.

Sun had some interesting numbers on Niagara-baser boxes serving lots of simultaneous requests. Too bad those machines never got the attention they deserved.

I guess that depends on how much is sacrificed on the altar of x86 compatibility. I was under the impression that it was a lot.

The cost of providing power to datacenters is going up both in absolute and percentage. You pay once to heat the server in waste energy then again for air conditioning to cool them back down. The ARM chips provide better perf per watt but the upper limit per ARM chip is way lower. To use them, it requires distributed software which is hard but we're getting there.

It would fail with the same reason itanium failed sometime ago. Who has the time and resources to port and tune every single software piece onto ARM instruction set.

Amazon(AWS) showed interest for these kinda uprocessors sometime ago for low end customers.

> Who has the time and resources to port and tune every single software piece onto ARM instruction set.

The Debian project.

Isn't tuning the LAMP stack would be good enough for many users?

Reply on news.ycombinator.com