From: Aron Broom (broomsday_at_gmail.com)
Date: Wed Nov 07 2012 - 10:23:21 CST
Yes, I see the sense here. The extra features are certainly worthwhile.
In fact, I've been meaning to test out LAMMPS. Still, for the cases where
one is running a smaller system (~100k atoms, or ~3k with implicit solvent)
and one can benefit from having many simulations running at once (i.e.
replica exchange, or umbrella sampling/metadynamics) the all GPU option is
Perhaps AMD graphics cards will promote the kind of competition needed to
get the better features without the massive costs? It seems like OpenCL
1.2 is no longer really awful compared to CUDA, no?
I'm curious as to the system cost difference if you build an AMD MB + CPU?
On Wed, Nov 7, 2012 at 9:45 AM, Michael Purdy <mdp3w_at_virginia.edu> wrote:
> Thanks everyone for the input. A clarification on my original post, the
> current Opterons are 2200 series (dual core) and the GTX-690 is plugged
> into a PCIe 2.0 slot.
> So increasing the CPU clock speed here is what I'm considering:
> Dual Xeon E5-2670 Sandy Bridge-EP 2.6 GHz (8-core) CPUs
> GTX-690 (2 GPU, 3072-core)
> 32 GB RAM
> Asus Z9PE-D8 WS motherboard
> motherboard specs
> Intel® Socket 2011 Processors
> Dual Intel® Xeon® processor E5-2600 product family
> Supports Intel® Turbo Boost Technology 2.0
> Chipset Intel® C602
> Memory 8 x DIMM, Max. 64GB, DDR3
> 2133(O.C.)/2000(O.C.)/1866(O.C.)/1600/1333/1066 MHz ECC, Non-ECC,
> Un-buffered Memory
> Quad Channel Memory Architecture
> System Bus 8.0/7.2/6.4 GT/s Intel® QuickPath Interconnect
> Multi-GPU Support Supports NVIDIA® 4-Way SLI™ Technology
> Supports AMD Quad-GPU CrossFireX™ Technology
> The total cost of this workstation will be about $5000 ($3079 for the
> CPUs). Has anyone tried the AMD FX-8350 Vishera 4.0GHz (8-core) $219?
> On 11/07/2012 05:34 AM, Ajasja Ljubetič wrote:
> Can't find the benchmarks right now, but if I remember
> correctly, hyper-threading was not much use. Under linux, there was no
> speedup when using 8 instead of 4 threads (either with or without the
> GPU). Under windows there was about 5-10% gain when using hyper threading--20cf3005138a17549104cdea23f7--
This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:22:14 CST