Hewlett-Packard, the No. 1 maker of servers in the world, may become the first server company in the world to make servers powered by microprocessors that are based on ARM architecture. The company reportedly will use chips specifically designed for servers by Calxeda chip developer.
HP and Calxeda (formerly Smooth-Stone) will be targeting servers based on ultra low-power ARM chips at companies who build large data centers and need to lower energy consumption and their physical footprint, such as Internet and cloud computing companies as well as companies looking to do analysis on their data., according to Dow Jones Newswires.
The first Calxeda reference design for OEM partners and select end-user and developer customers, will be based on an ARM Cortex-A9 quad-core SoC. The system-on-chip will enable users to design servers as dense as 120 ARM quad-core nodes (480 cores) in a 2U enclosure, with an average consumption of about 5W per node (1.25W per core) “including DRAM”, according to the company. The chips are presumably made at Globalfoundries using 28nm or 45nm process technology.
Ultimately, Calxeda plans to provide users ARM architecture-based chips as well as products with the Calxeda microprocessors that will offer 5 – 10 times performance/watt advantage and 15 – 20 times advantage when price is factored in.
Calxeda and HP did not comment on the news-story. Industry checks by X-bit labs indicate that the companies are prepping a joint project.
Tags: HP, Hewlett-Packard, ARM, Calxeda, Cortex
Comments currently:
7
Discussion started: 10/26/11 09:39:55 AM
Latest comment: 10/28/11 01:13:46 AM
Expand all threads |
Collapse all threads
[1-3]
1.
one question if you have so many cores don't you have to program super parallel? and whit if some code parts where single threaded than you would have a big problem. or do they give you some kind of mixed hardware whit some blades that are single threaded powered?
1
1
[
Posted by: massau
|
Date: 10/26/11 09:39:55 AM]
+ expand thread (3 answers)
- collapse thread
Gawd, I mean, really? I'm think I'm going to re-engineer my server room based on your suggestions, apparently servers haven't been parallel centric for the last 30 years, I just didn't read the right hardware reviews about Sandy Bridge's awesome single threaded performance.
0
0
[
Posted by: dukie_bref
|
Date: 10/26/11 06:00:29 PM]
i'm just saying that even if you have 1000cores which would be super fast but you can't make every piece of code parallel. k the 999 other cores can do something else so you will have most of the job done faster. but I still think we need some mixed server de designs like 20big heavy cores for ST and 500smaller cores for the MT
0
0
[
Posted by: massau
|
Date: 10/27/11 09:40:20 AM]
Multiple processors in a server running Linux and hosting a site that is using Apache. Apache runs multiple "instances" of itself, so Linux will balance load by spreading all the multiple Apache processes on each processor. Threading is not needed at this point, but the kernel needs to be set for SMP systems. Basically what HP has is a 2U rack that has 120 computers connected into a network and these computers have four processors that can handle about four "instances" of Apache.
0
0
[
Posted by: tecknurd
|
Date: 10/28/11 12:58:10 AM]
2.
A node is not a CPU, or a socket. A node is a complete indipendent computer with its own "not shared" RAM, and there is not necessary any special comunication channel between each node.
Each node run on a separate OS, so 120 nodes in a 2U chassis, this means that we have 120 complete quad core computers in each 2U chassis.
This architecture is logically good for scalable applications distributed in a network of nodes where communications between parallel processes running on each node is minimized.
A web server where all HTTP queries are done in parallel and all communications between queries is done passing throw a database is probably the best candidate to run on this kind of server.
Telecommunication applications where there is no need to maintain a coherent shared state between nodes seam too a good application field.
The target of this architecture is cloud computing, where each parallel operation is spread between different processes, no between many threads in the same process. In fact potentially each module of the application is potentially running in a separate computer.
So don't think about many threads running in the same machine. This seam just a lot of slow, low power, quad core computers (node) packed in few space.
0
0
[
Posted by: Serenico
|
Date: 10/26/11 11:08:50 AM]
3.
I heard that a X86 CPU can emulate ARM without problem, but the reverse is not possible (take day to complete the task)
Can this ARM server do virtualization?
0
0
[
Posted by: Tukee44
|
Date: 10/26/11 07:29:17 PM]
+ expand thread (1 answer)
- collapse thread
An ARM Cortex-A9 does not have any virtual hardware instruction support. Why would need a virtual environment when there are 120 computers in a 2U rack.
Virtual machines from VMware and Xen are used when high performance processors are in servers. Servers do not require high performance processors, but it is not efficient to use high performance processors in servers compared to many servers. Virtual machines are used to take advantage of high performance processors and use virtual machines to make hundreds of servers from virtual machines.
Do not need to emulate an x86 processor and no it will take a day to complete a task. Emulating an x86 on an ARM based setup is silly when there is already ARM ports. Linux and the GNU community does support ARM processors.
0
0
[
Posted by: tecknurd
|
Date: 10/28/11 01:13:46 AM]
[1-3]
Make Home Page
Add to Favorites
Xbitlabs on Facebook
XbitLabs is using Twitter
Windows Live Alerts
