For a production machine, reliability is everything.
Again, what is "production"? In addition: What is reliability? Asking rhetorically.
There is a wide range of how critical or risky the work is which is being performed on a particular machine. Therefore there is a wide range of what degree of reliability is required, or/ and can be achieved at reasonable cost.
By overclocking you are introducing another possible point of critical failure.
I stand by my point that overclocking does not necessarily sacrifice reliability. I mentioned multicore clocks as a reason for the existence of well-known reliability margins.
Another source of such margins is that while the CPU vendor needs to ensure reliability and longevity of an entire production run of CPUs, a PC owner inclined to overclock merely needs to aim for reliability and longevity of
one particular CPU in a restricted scenario of workloads and environmental conditions. The reliability margin from this source however is ill-defined. It can be determined only with extensive testing but limited certainty.
BTW, while the CPU vendor can and does perform much more extensive testing (on the production run, and prior to the production run, including simulations), the outcome of it all still results in limited certainty. Overclocking does not introduce a new mode of failure, it reduces margin of safety against a mode of failure which already exists. (While respective cooling and power delivery can increase that same margin.)
Further, in a large shop if your machine fails I will unplug your current machine, and take one off the spares rack to replace it. The dead machine will then be fixed, and deadlines will be met.
I agree that the primary way to achieve hardware availability is hardware redundancy. (For desktop PCs: spare machines which can be swapped in on shortest notice.)
You don't own the equipment, unless you work for yourself. And if your tinkering causes deadline problems, then you get to experience the unemployment line.
There are many examples how tinkering can, and routinely does, cause deadline problems. E.g. rolling out a poorly tested software update whose bugs cause downtime. These things cannot be prevented entirely; rather it is largely a cost-benefit consideration how far a company's processes can go to combat such risks.
Please reread EXACTLY what he wrote. He is a *developer* and is running 6700/7700 on his workstation, neither of which support ECC either. ECC is not important for his particular role.
What is a "developer"? What do you know about the workloads that he is running?
So he is performing work which prohibits overclocking, but at the same time supposedly does not benefit from ECC RAM.
Yet ECC RAM in desktop computers isn't unnecessary, it is merely priced too high to be attractive for wider adoption (considering total system costs, not just RAM costs). If you are ready to give up ECC RAM for cost reasons, you may as well be ready to give up more+slower cores in favor of fewer+faster cores, likewise for cost reasons.
In extension, maybe operate those cores at higher than factory-defined clocks for even higher productivity at same hardware costs.
Also, regarding your response to my initial point about not building machines for use at work - you can disagree, but you're wrong. NO reputable company with any sort of sense is going to say "Hey Joe, here is a corporate card - go knock yourself out at Newegg.". That is an exceptionally poorly run IT department that allows this and I've worked in companies ranging from Fortune 10 to smaller orgs.
There are many kinds of "production", and there are many kinds of businesses and corporations. Certain large corporations with datacenters as the very heart of their business develop, build, and maintain their servers in-house, and for very good reasons. At the other end of the spectrum, small businesses with particular needs may benefit from in-house configuration and perhaps even in-house assembly too. In either case it requires sufficient in-house knowledge and experience of course.
Are you claiming that corporations can't have that in-house capability?
[Overclocking at the workplace:]
Again, real companies don't work like this and allow this to happen.
Again, there are many kinds of "production", and there are many kinds of workplaces, businesses, and corporations.
Also, different but not unrelated: There are cloud providers who use desktop harddrives far outside of vendor specifications in huge storage racks. They can do so because they plan and compensate for the increased failure rate.