Sign in with
Sign up | Sign in
Your question

Intel i7-3770K vs. i7-3820 for molecular dynamics.

Last response: in CPUs
Share
September 20, 2012 5:17:11 PM

Hi there,

I'm trying to decide between two workstation builds:

https://dl.dropbox.com/u/5381783/build.pdf

https://dl.dropbox.com/u/5381783/build2.pdf

The components are all the same, except for the motherboard and CPU. One is the i7-3770K, an Ivy Bridge processor; the other is the i7-3820, a Sandy Bridge-E processor.

A few components are missing (power supply, storage HD, video card) because I plan on salvaging them from existing boxes. I have two GTX 480's and a GTX 470 that I plan to install, but in the future we might upgrade them to GTX 680 or whatever comes out in the future.

My application is scientific computing. I need this box to do two things mainly:

1) Rapid execution of single-threaded processes on the CPU 2) High bandwidth and low latency for communication between the CPU and GPU

From what I've heard, the i7-3770K (Ivy Bridge) has the following advantages:

1) More rapid single thread execution by 10-20%.

2) Supports PCI-e 3.0, whereas Sandy Bridge doesn't (?) I've gotten conflicting evidence over this: (http://www.anandtech.com/show/5264/sandy-bridge-e-x79-p...). Only the newest graphics cards support PCI-e 3.0, so this pertains to upgradeability and not the present GPUs.

3) Supports USB 3.0 whereas Sandy Bridge-E does not? This isn't a primary requirement, but since this is a workstation, rapid data transfer to external drives is a big plus.

4) Smaller transistors are "cooler" :) 

The i7-3280 (Sandy Bridge-E) has the following advantages:

1) More memory bandwidth (quad channel, vs. dual channel for Ivy Bridge)

2) More PCI-E lanes for greater GPU/CPU communication bandwidth (40 lanes, vs. 16 for Ivy Bridge)

I'm leaning towards the Sandy Bridge-E processor because of the greater bandwidth. Some people say I don't need 40 PCI-E lanes with two GTX 480 cards - however, scientific computing applications can require very high data throughput (I don't know how to measure exactly how high) and it would be great to have a machine where bandwidth is not an issue. I think this is potentially more important than the performance boost on single thread execution that I'd get from the Ivy Bridge.

Overclocking is a minor concern. I've never overclocked any CPU (even though I've assembled 20+ boxes over the years), and when it comes to work-related hardware I am relatively risk-averse. That said, if one of the CPU can be more easily overclocked by a minor amount, it might affect my decision.

Thanks for reading!
September 20, 2012 5:53:34 PM

An additional note - I'm running molecular dynamics simulations such as GROMACS, TINKER, OpenMM and AMBER. Some of these are single thread processes and others aren't; some of them run on the GPU.
m
0
l
a b à CPUs
September 20, 2012 6:13:02 PM

I don't think the bandwidth is going to matter over the faster processing speed of the 3770K. That is to say the 3770K should be faster in every task.
m
0
l
Related resources
September 20, 2012 6:33:10 PM

FinneousPJ said:
I don't think the bandwidth is going to matter over the faster processing speed of the 3770K. That is to say the 3770K should be faster in every task.


It's tough to determine bandwidth requirements because the data throughput of these jobs is not well characterized or easily measurable.

I would like to have a box where bandwidth is maximized, if only for the purpose of testing whether throughput is a bottleneck or not.
m
0
l
!