Overclocked and Non-ECC Hardware for Publishing Papers

brogers · October 11, 2016, 18:14

Hello all! First, I would like to say that I'm happy to finally be apart of this community. It has helped me out tremendously for the last couple of years in undergrad and into grad school, so it's nice to no longer be just a lurker.

I am configuring a few potential builds for doing my doctoral research as our office servers and blades get bogged down by my office mates. I will be funding and building this machine myself. To not get into the semantics on what hardware would be adequate for my needs (another post of questions for another time), I'd like to ask a more etiquette related question.

What is the general consensus in the publishing community (JFM, AIAA Journal and Conference Papers, etc.) about the use of overclocked CPUs (unlocked and overclocked core i series processors) and NON-ECC memory to obtain results intended for publishing? For the memory part of the question, I'm not just referring to DRAM, but also NVIDIA Geforce vs Quadro and Tesla for GPGPU related programming. I'm worried that the use of this hardware is frowned upon due to the increased instability of overclocked CPUs and RAM compared to "workstation" grade hardware.

Thanks in advance. Any help is greatly appreciated.

brogers · October 27, 2016, 10:38

No one has any input on this?

flotus1 · October 27, 2016, 13:16

I have not come across a publishing policy that explicitly discourages the use of consumer hardware and non-ECC memory.
In my opinion you can use whatever hardware you consider adequate. The reason is: you have to do verification and validation anyway and control numerical errors. See e.g. http://www.cfd-online.com/Wiki/Ansys_FAQ#Postprocessing to get started. It is next to impossible that a hardware error slips through a proper v&v process.
Not to mention the countless publications that are actually based on results obtained with consumer hardware without ECC that is often overclocked. One example: https://folding.stanford.edu/

brogers · October 27, 2016, 13:22

Along with V&V, I was also planning on comparing results from factory clocks to overclocked speeds to see if repeatability issues are created (getting the same answer every time is one of the selling points I've heard of ECC RAM).

Publications from things such as folding@home hadn't even occurred to me.

Thanks!

evcelica · November 1, 2016, 11:35

I've been using overclocked CPU and RAM, and non ECC memory in my 5 node cluster for years. Never had a single problem with accuracy.

October 11, 2016, 18:14	Overclocked and Non-ECC Hardware for Publishing Papers	#1
brogers New Member Join Date: Oct 2016 Posts: 3 Rep Power: 10	Hello all! First, I would like to say that I'm happy to finally be apart of this community. It has helped me out tremendously for the last couple of years in undergrad and into grad school, so it's nice to no longer be just a lurker. I am configuring a few potential builds for doing my doctoral research as our office servers and blades get bogged down by my office mates. I will be funding and building this machine myself. To not get into the semantics on what hardware would be adequate for my needs (another post of questions for another time), I'd like to ask a more etiquette related question. What is the general consensus in the publishing community (JFM, AIAA Journal and Conference Papers, etc.) about the use of overclocked CPUs (unlocked and overclocked core i series processors) and NON-ECC memory to obtain results intended for publishing? For the memory part of the question, I'm not just referring to DRAM, but also NVIDIA Geforce vs Quadro and Tesla for GPGPU related programming. I'm worried that the use of this hardware is frowned upon due to the increased instability of overclocked CPUs and RAM compared to "workstation" grade hardware. Thanks in advance. Any help is greatly appreciated.

October 27, 2016, 13:16		#3
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,427 Rep Power: 49	I have not come across a publishing policy that explicitly discourages the use of consumer hardware and non-ECC memory. In my opinion you can use whatever hardware you consider adequate. The reason is: you have to do verification and validation anyway and control numerical errors. See e.g. http://www.cfd-online.com/Wiki/Ansys_FAQ#Postprocessing to get started. It is next to impossible that a hardware error slips through a proper v&v process. Not to mention the countless publications that are actually based on results obtained with consumer hardware without ECC that is often overclocked. One example: https://folding.stanford.edu/ brogers likes this.

October 27, 2016, 10:38		#2
brogers New Member Join Date: Oct 2016 Posts: 3 Rep Power: 10	No one has any input on this?

October 27, 2016, 13:22		#4
brogers New Member Join Date: Oct 2016 Posts: 3 Rep Power: 10	Along with V&V, I was also planning on comparing results from factory clocks to overclocked speeds to see if repeatability issues are created (getting the same answer every time is one of the selling points I've heard of ECC RAM). Publications from things such as folding@home hadn't even occurred to me. Thanks!

November 1, 2016, 11:35		#5
evcelica Senior Member Erik Join Date: Feb 2011 Location: Earth (Land portion) Posts: 1,188 Rep Power: 23	I've been using overclocked CPU and RAM, and non ECC memory in my 5 node cluster for years. Never had a single problem with accuracy.