If I can spend $5k to save a week of optimization work, it's a good deal. Also buying clockspeed can't create new bugs either, which is a risk with any code change.
If I can spend $5k to save a week of optimization work, it's a good deal. Also buying clockspeed can't create new bugs either, which is a risk with any code change.
That said, companies that need more speed often don't need that extra speed on just one node. So that's really $5K x N nodes. When you consider that just the CPU costs $15K, and not the complete cost of a system, I think it's a lot harder to justify. You better have that next lower tier $10K CPU and know that you need just that little bit more. And then... you probably ought to be investing in scaling out horizontally while that $5K x N buys you time.
Famous last words
Some software does not like high clockspeeds. For example, game engine in GTA V starts acting up when framerate gets too high.