Non-FAH: Anton 2: ASIC for Molecular Dynamics Simulation

Moderators: Site Moderators, PandeGroup

Non-FAH: Anton 2: ASIC for Molecular Dynamics Simulation

Postby #64-Lux » Thu Aug 14, 2014 9:18 pm

Anton 2: A 2nd-Generation ASIC for Molecular Dynamics Simulation
1 CPU = 66 cores at 1,6 GHz. 2.5 Tb/s on-chip bandwidth. Over 250 billion interactions/s per CPU.
Big emphasis on interchip communication was an eye-opening for me. Obvious, of course. :)

Some info regarding relevant strengths of F@H and Anton: here and here.

Edit: Couple of links about Intel's Xeon Phi - similar number of cores at nearly the same speed, but general purpose: details and specs. Next year's Knights Landing.
#64-Lux
 
Posts: 54
Joined: Sat Jan 26, 2008 5:35 pm

Re: Non-FAH: Anton 2: ASIC for Molecular Dynamics Simulation

Postby bruce » Thu Aug 14, 2014 11:27 pm

Two interesting facts I discovered by reading the 2008 paper on Anton (1) - 512 Nodes

* It uses ~116.5 KW, including ~ 30 KW just for cooled water.
I don't think many of us would consider one as part of our farm.

* We optimize for latency!
– Not throughput
– Not utilization

How many Donors even consider latency?

As far as Anton 2 is concerned, I can't find any information about when it might be available to the general public or whether there will be a version that might be used by FAH in home computers. It seems that their only objective is to build a supercomputer that competes with Blue Gene
bruce
 
Posts: 22322
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Non-FAH: Anton 2: ASIC for Molecular Dynamics Simulation

Postby #64-Lux » Fri Aug 15, 2014 5:08 am

I don't think Anton2/x is intended to be generally available. It's a very niche piece of hardware, that in my opinion loses the mass-market appeal compared to Xheon Phi. Still, the Anton project started long before Xeon Phi was available and has apparently got some really good results (otherwise version 2 would not have been made). I'm sure 512-node Anton2 is a computational beast that can achieve quite long timescales (in computational MD sense) and complements/supplements the F@H computational projects very nicely ;).
#64-Lux
 
Posts: 54
Joined: Sat Jan 26, 2008 5:35 pm

Re: Non-FAH: Anton 2: ASIC for Molecular Dynamics Simulation

Postby #64-Lux » Tue Feb 24, 2015 7:28 pm

In this video Vijay Pande compares FAH and Anton.
#64-Lux
 
Posts: 54
Joined: Sat Jan 26, 2008 5:35 pm


Return to Discussions of General-FAH topics

Who is online

Users browsing this forum: Yandex [Bot] and 3 guests

cron