At the desktop level I don't get why people care that much about power consumption? It means you have to dissipate more heat, okay, that means you can't use a cheap cooler. But it AFAIK even an extra 100w is cheap in the most expensive areas, especially when contrasted against productivity or cigarette breaks, or people sometimes being 20 minutes late to work...
Actually, air-cooling is marginal with these CPUs, even at stock frequencies. There is no air-cooler which allows them to sustain their frequencies under load.
Water-cooling is pretty much required, and an AIO does not cut it. Still, even with water-cooling you can't really overclock these. They are pretty much at their limit out of the factory.
This could conceivably solved by Intel switching away from silicon TIM to e.g. solder, since the Rth(jc) of the CPUs is much worse at ~0.3 K/W than the thermal resistance of a big CPU air cooler (~0.1 K/W).
The overclocking problems wouldn't be fixed though; there is no easy fix for a CPU that jumps to 400+ watts.
-
An entirely separate issue is that you need to get the heat out of your office. A human dissipates around 50-100 W; you can imagine that a small office crowded with half a dozen people is not pleasant in the summer.
I fully agree, and what's more Intel has the performance to back it up. This chip pulls a lot but it's wicked fast, it's a massive step forward in framerates. It combines the minimum-framerate improvements of HEDT/Ryzen with the single-threaded performance of Kaby Lake. Oh yeah and AVX512 too.
For a sense of perspective here, going from a circa-2012 2600K to a current 7700K is a 40% jump in performance, so it's roughly equivalent to 4-5 years of gains at Intel's usual tempo - only you also have 10 cores on this platform. This thing is an absolute monster for gaming or other tasks that lean heavily on single-threaded performance.
But the power consumption is really the triggering issue for the problems with shitty partner-boards overheating and the TIM. The TIM is really the showstopper right now.
For me its purely a noise issue. Less heat to dissipate means less fan noise(Only speaking about home use here. In an office setting the difference would be unnoticeable).
> In an office setting the difference would be unnoticeable
If you have a hall full of developers, having lots of noisy PCs can be annoying.
Where I work, we optimized for more silent PCs, because all these things do add up, and given the perf/watt-ratio you can get out of modern CPUs, there's no reason for people to need to have noisy PCs.
Even at home, optimizing watt-usage, even for a desktop build, is not completely without merits. All my future projects are planned as fanless as possible. And I know others who do the same. And if Intel can't deliver that, they'll just go buy something Arm-based, like an Rpi 3, which these days are getting good enough to actually do production loads.
I'm not even getting close to wanting a system where the CPU alone can draw 400+ watts.
> an Rpi 3, which these days are getting good enough to actually do production loads
They really aren't, CPU performance is really irrelevant given the RPi's architectural weaknesses. USB was never meant as a system bus and everything has to loop through the kernel stack. Having every single peripheral hanging off a single USB 2.0 bus is crippling for performance. A Pi can't even serve a share at full 100 mbit speed due to bus contention let alone do anything more intensive.
It's very similar to one of Apple's more famous goofs, the Performa 5200/6200 with its left-hand/right-hand bus split that forces the CPU to handle everything.
Some of the clone boards have USB 3.0, SATA, gigabit ethernet, etc and are much better performers in practice despite having slower CPUs "on paper". Or there are little mini-PCs using 5-15W laptop processors that are really nice and run x86 distros/binaries.
All of these are at roughly comparable TCOs to a Pi (they include things like AC adapters that must be purchased separately for the Pi). The RPi is a bad choice for server usage.
> They really aren't, CPU performance is really irrelevant given the RPi's architectural weaknesses
Obviously "production loads" is an undefined term and as such we can discuss infinitely back and forth exactly how much these cheap ARM machines can actually handle.
I also didn't mean to single out the Rpi3 as a universal performer, optimal for everything out there.
My point was that I'm seeing an increased amount of people who are happy with what these cheap boards can do, who 10 years ago would have been forced to buy a server of sorts to cover the same needs.
So now they don't buy servers. Instead they buy cheap, tiny and fan-less ARM-based machines and they're perfectly happy. They even think running ARM is cooler than running Intel, so it's something they brag about.
I'm absolutely not saying I'm going to replace my company's server-farm or my dev-computer with these anytime soon, but Intel cannot completely ignore the power-efficiency aspect either if they want to keep their dominance in the market.
