NSG S0, once out, will most likely be the go-to case for such setups. Until then, an HDPLEX H5 is cool.
My desk has a H5 on it, housing an i7 8700 (non-K) and a GTX 1060. The TIM under the heatspreader is replaced with Thermal Grizzly Conductonaut and Thermal Grizzly Kryonaut is used as every other TIM that the case setup needs. The CPU is on stock clocks with a voltage offset of -30 mV. The GPU has the power target reduced to 90% and clocks increased by 130 MHz, so that it is effectively undervolted as well. The PSU is a Seasonic Ultra Prime Titanium 650. Prime95 with AVX throttles really, really fast, under a minute, perhaps, but is a very unrealistic load. Non-AVX stress tests and FurMark take a while to start throttling (20 minutes?), as the thermal capacity of the aluminum case is quite big. After hours of gaming, the GPU and CPU float around 80 C while providing full stock performance. I don't do 3D rendering (other than in-game) or video en/decoding, so have not had long, real-world, full loads to see how temperatures behave with those.
From the discussion I've had and forums I've read, I think that people are afraid of putting more power in passive cases and having their components at "high" temperatures, despite those being rated for them.
>Prime95 with AVX throttles really, really fast, under a minute, perhaps, but is a very unrealistic load
I suppose blender would thermal throttle the cpu as well. If you run any non-Xeon/non-Laptop Intel chip (greater than 2k series) and care about temperatures - delid the bugger. (Xeons are soldered, laptop chips don't have IHS). Intel uses something that's worse than toothpaste, plus tons of glue between the die and the IHS. If you see temperature deltas under full load more than 9-10C between the cores, the thermal paste between the die and the IHS might have missing spots or have dried out.
In your case removal of the IHS altogether would provide decent results.
You might wish to check the VRMs, they are rated at 125C but if the case is hellishly hot inside, they might not be able to dissipate the heat.
Somewhat unrelated, but maybe you can shoot me down since you seem to have some experience?
Metal is an incredibly good conductor on its own, and the properties of thermal paste (typically) are just barely better than air. So long as your cpu and heatsink are fairly flat surfaces and mashed together physically, it seems like either forgoing or having the absolute minimum amount of paste is ideal. I've used a razor to leave an absolutely minimal layer of paste (e.g. filling in sub-millimeter surface structure) on my latest build, and cpu temperatures are well within a reasonable range. But I'm also not trying to OC the cpu or anything.
>...and the properties of thermal paste (typically) are just barely better than air.
I am not certain how you have managed to come to such a conclusion. Thermal conductivity of air is around 0.03W/(m·K)[0]. Good thermal, non-conductive paste is like 12.5W/(m·K)[1] (or 400 times better than air). Conductive ones are in the region of ~40-80 W/(m·K) and Aluminium is 237W/(m·K). Also air also expands pushing the cooler and CPU away.
Normally you if choose between "too much" and "too little" paste, you pick the former. The pressure pushes out the unneeded amounts.
Don't, you need to multiply the raw conductivity by the linear distance occupied by the thermal paste? I presume that distance will be at least two orders of magnitude larger than that occupied by air in a metal contact only setup.
I would be extremely surprised if increased pressure due to air at higher temperature played any role whatsoever unless the bolts connecting the heatsink and cpu were very loose. If anything, I'd expect the increased conductivity of air at higher temperatures to dominate.
I'd also expect there to be effects at the metal-paste and paste-metal interfaces which reduce the effective system conductivity (i.e. phonons are much more likely to reflect in this scenario than in a metal-metal interface).
It's very impractical/expensive for mass products to make the surfaces in question so flat that no thermal paste would be needed. Many tests and reviews have been done. Even if top-of-the-line coolers came with perfectly flat surfaces, Intel's heatspreader is not -- otherwise it would cost so much more. Also, heatsinks can be applied with a lot of force, which usually pushes out the "unneeded" part of the thermal paste. In a bind, even lipstick, toothpaste, chocolate and other silly compounds work better than nothing, so I'm not surprised that you're getting ok results even with a touch of thermal paste.
A fun thing to try is using a modern low-end CPU (latest i3s, Pentiums, Celerons) without its cooler. Not advised by Intel, of course, but you might get into your OS of choice even before it starts throttling. I'm somewhat comforted by the fact that a CPU automatically powers of once it reaches something above 100 C (103 maybe?) and throttles a few degrees before that. Those temperatures shouldn't leave the silicon damaged.
In practice, thermal paste is a must. If you don't like those (I personally don't, they get everywhere by accident and can be tough to remove), try getting an IC Graphite Thermal Pad which is reusable and rivals really good, if not the best thermals pastes, according to the limited number of reviews I've seen. I think that its practicality beats better results in non-highest-end applications.
Its really not that hard to create flat surfaces- anyone with a lathe, a hard cutting instrument, and a bit of fine grinding material can probably get a contact bond between two pieces of metal.
Blender might throttle. Does it come with benchmarks/tests that I could use to gauge thermal performance?
The CPU is delidded! I've got another i3 4300 delidded as well running under a NoFan CR-80EH. Delid + Conductonaut + Kryonaut made the difference between throttling vs hovering around 90 C in FurMark + Prime 95. When integrated graphics aren't used, the CPU runs cooler, of course, and didn't throttle with MX4 thermal paste and no delid.
I do fear that VRMs are running too hot. When selecting components, I picked those that come with some heatsinks on VRMs at least. The motherboard is an AsRock Fatal1ty Z370 Gaming-ITX/ac (non ITX motherboard wouldn't fit in the case anyway with an ATX power supply). The graphics card is Gigabyte's cheapest offering and has a small sink across the VRMs. I'm hoping that undervolting will help keep the VRMs in check.
NSG S0, once out, will most likely be the go-to case for such setups. Until then, an HDPLEX H5 is cool.
My desk has a H5 on it, housing an i7 8700 (non-K) and a GTX 1060. The TIM under the heatspreader is replaced with Thermal Grizzly Conductonaut and Thermal Grizzly Kryonaut is used as every other TIM that the case setup needs. The CPU is on stock clocks with a voltage offset of -30 mV. The GPU has the power target reduced to 90% and clocks increased by 130 MHz, so that it is effectively undervolted as well. The PSU is a Seasonic Ultra Prime Titanium 650. Prime95 with AVX throttles really, really fast, under a minute, perhaps, but is a very unrealistic load. Non-AVX stress tests and FurMark take a while to start throttling (20 minutes?), as the thermal capacity of the aluminum case is quite big. After hours of gaming, the GPU and CPU float around 80 C while providing full stock performance. I don't do 3D rendering (other than in-game) or video en/decoding, so have not had long, real-world, full loads to see how temperatures behave with those.
From the discussion I've had and forums I've read, I think that people are afraid of putting more power in passive cases and having their components at "high" temperatures, despite those being rated for them.