Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Baidu rolls out world's first commercial deployment of Marvell ARM server (marvell.com)
35 points by Ecio78 on March 4, 2013 | hide | past | favorite | 10 comments


This article on El Reg gives a better overview I think, including mention of the 40-bit memory addressing on the Armada XP that has been commented on in here already.

http://www.theregister.co.uk/2013/02/20/baidu_marvell_arm_se...


Interestingly I noted that it supports 16GB DDR3 RAM. Since all current ARM cores are 32bit I wonder if they are doing some sort of bank switching to address all the RAM.


Apparently some Marvell CPUs have 64-bit address extension http://www.marvell.com/embedded-processors/armada-xp/


Most likely it's something like x86 PAE, where each process has a 32-bit address space and the page table just has a few extra bits to map into a >4GB physical address space.


As others have said: with PAE. This is included in ARM's Cortex A15 Core, although AIUI Marvell use their own.

Further reading:

https://en.wikipedia.org/wiki/Cortex_A15

http://www.eetimes.com/electronics-news/4206387/ARM7-40bit-v...


I'm a little outside of my area of expertise, but didn't Intel late model 32-bit chips have a form of larger memory addressing? I thought I read somewhere that there was a greater-than-32-bit memory address setup in x86 architectures, but Microsoft products wouldn't take advantage of it, only Linux and other OSs were able to use it.

Even in a modern x86-64 bit architechture, you can do some limited 128 and 256 bit operations. This ARM may have a greater-than 32 bit memory addressing setup, but only support 32 bit opcodes for the main registers.


You're thinking of PAE. It allowed the system as a whole to use more than 4Gib, but it still limited a virtual memory space to 4GiB.


There's a couple of patently untrue statements inside the press release, like "As the first organization in the world to leverage ARM servers for commercial use...", but the move is another sign of the growth of ARM servers.

To me, the real piece of information is near the bottom, "ower the total cost of ownership by 25 percent, compared with previous x86-based server solutions", this is where Intel's challenge really is.

It's not about performance, or even about Watts consumed, it's about the money spent to operate a service, from the very bottom of the stack where electricity is consumed, through the management of the application layer.

Server hardware is going to cost a lot less in the next couple of years, and Intel are going to have to work out how they can make money in this new environment, whether that's through going all out on low end kit, or moving up the management layer.


Is there a way for me to get access to a machine like this if I want to see how well my code runs? If I want to benchmark some code on a crazy i7, I can just fire up an instance on EC2 for a couple of hours.

I'm not aware of anything like that for the newer ARM processors. There's an Ubuntu ARM AMI which lets you ssh straight into an ARM image.. running on QEMU on top of EC2. Not ideal.


Intel: Be afraid, be very afraid.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: