[Beowulf] Linux 3.7 and ARM

Vincent Diepeveen diep at xs4all.nl
Thu Dec 13 06:00:44 PST 2012


In HPC never as long as there is competition in HPC.

Big fat thick manycores will dominate of course.

Otherwise you'll eat 10x power especially for the interconnects which  
is the biggest problem to solve.

Just do the math.

Price of 1 Tesla/Xeon Phi is on average $3k.

For $3600 you've got 1.3Tflops worth of Nvidia.

Those ARMs they will always use little cores and clock them too high  
(as that sells a tad in phones).

Even if you can buy them in for $10 a piece low power. It's 4 cores @  
1.0Ghz using 6 watts. Then RAM eats some
and losses everywhere you eat effectively 10 watt probably, assuming  
a very very efficient psu.

Then interconnect will eat less. Say 5 watt a cpu.

15 watt in total.

So you need for 1.3Tflop then 1300 / 8 = 162.5 = 163 CPU's.

15 watt * 162.5 = 2437.5 watt.

That's for an ARM chip that still needs to get produced at a recent  
production process versus EXISTING crunching power.

So you can easily add factor 2 to the ARMs ineffiency rating.

What's price of all those interconnects going to be? I don't know.  
The current price of the leading chip, the Nvidia K20,
it's of course a market price they thought they could ask at Nvidia,  
it's not the production price, which probably is factor 20 cheaper.

On Dec 12, 2012, at 10:57 AM, Hearns, John wrote:

> Good article:
>
> http://www.zdnet.com/linux-3-7-arrives-arm-developers- 
> rejoice-7000008638/
>
>
>
> When do we think 64 bit ARM will come in to general use in HPC?
>
>
>
> Also TCP Fast Open looks interesting  https://lwn.net/Articles/508865/
>
>
>
>
>
> John Hearns | CFD Hardware Specialist | McLaren Racing Limited
> McLaren Technology Centre, Chertsey Road, Woking, Surrey GU21 4YH, UK
>
>
> T:  +44 (0) 1483 262000
>
> D:  +44 (0) 1483 262352
>
> F:  +44 (0) 1483 261928
> E:  john.hearns at mclaren.com
>
> W: www.mclaren.com
>
>
>
> The contents of this email are confidential and for the exclusive  
> use of the intended recipient. If you receive this email in error  
> you should not copy it, retransmit it, use it or disclose its  
> contents but should return it to the sender immediately and delete  
> your copy.
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin  
> Computing
> To change your subscription (digest mode or unsubscribe) visit  
> http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list