The Passive Cooling Paradigm: Atlast Solutions Ultimate Fanless Core i7 7700T
When deciding on a computer, you have a budget to consider. Let ‘budget’ be an axis and on both far ends life is simple. That is: if you have no money there is no room to cater any needs. If your budget is unlimited you can simply satisfy all your longings as long as these are offered somewhere.
With respect to that budget axis, most people live somewhere in the middle, where life is utterly complicated. Where offerings are abundant, all with their respective value, price features and limitations to consider. The free market adds marketing to the equation, seemingly to further complicate consumers’ lives even further. Some first world problems that need to be solved.
Put your savings on one end of the seesaw and your needs, wishes and cravings on the other. Keep reorganizing your wishes by their perceived weights along the beam, then step back, and reconsider until you reach perfect harmony. Your author can stretch this process for weeks and months on end.
Simply put, I want to eat cake and have it and not feel silly in hindsight.
Of course that is too much. Or is it?
The craves and wishes
Whilst editing audio recordings, I find myself making ‘noise profiles’ often for use in the Audacitys noise reduction filter because the computer I have is the main cause of noise in the room and it is vocal enough to make me want to reduce it. I’ve put ‘not so noisy’ somewhere on top of the wishlist for my new computer.
Watching HD videos let alone converting a video file and even playing chess online (!) can be a demanding task for my 2009 era computer. So much so that these applications make it bend its head lowly and groan as if I had asked it to move Greenland over to Denmark. Second on the wish list is some more power and hard wired features for the loads I treat it on. Hardware support for decoding and encoding would be greatly welcomed.
I got used to the convenience of compact housing and bulky computers are not really my type. I will avoid the big tower design-paradigm if I can.
I don’t play games.
While there are some intriguing and valid use cases for the high-end graphics card apart from gaming, such as training your pet (convolutional) neural network or accelerate video- or image manipulation. However such power comes with a considerable price tag and cooling concerns when put to work in a confined space.
And while I would have loved an occasional flight in X-plane 11, I doubt that occasion would justify the costs. The high-end graphics card didn’t make it to the list.
Weeding the market
A table with many computers of interest with their respective noise production levels was my foremost helpful tool. I picked a ‘best, least noisy in a modest form factor’-computer and ended up not purchasing it because it was past midnight and I don’t buy past midnight unless I am absolutely sure and apparently I wasn’t.
On a fresh new day, just a few hours later. The oracle ‘Google’ brought Atlast! Solutions’ Fanless computers to my attention.
You should know your author has a sweet spot for computers without moving parts because these cannot make unwanted noise, do not suffer mechanical wear and are intrinsically more energy efficient.
The offerings on powerful fanless computers I had seen before however, came with an exclusiveness penalty added to the price tag and came with RS-232 ports. The first I could not afford and the latter was nowhere on the wish list. Apparently I have not been paying attention because today there exists more than solely the expensive industrial offering.
At this point I cheated and adjusted my budget to match the computer’s price.
I bought the:
Atlast! Solutions Ultimate Fanless:
Its casing houses an Intel Core i7 7700T ( 0.8 - 3.8 GHz) on an Asus ROG Strix Z270i Gaming motherboard paired with 16GB of DDR4 at 3000MHz and a Samsung SM961 256GB M.2 PCIe NVMe drive.
Note that the configuration is not fixed. You can choose motherboards, memory modules, drives and whether you would like an optical drive fitted.
On the back:
4xUSB 2.0 / 4x USB 3.0 (one C-type) / 1 HDMI / 1 DisplayPort / 5 audio jacks for (120 dB SNR in playback) 7.1 Surround / Optical S/PDIF / Gigabit ethernet / Bluetooth 4.1 and Wi-Fi 802.11 a,b,g,n & ac
All one is to expect of a modern computer is there, the whole shebang. I did have ‘thunderbolt’ pencil-scribbled somewhere - for future expandability – but an eraser was quickly found. Do note the absence of RS-232 ports!
It is housed in the ‘Impactics C3LH’ case
Its dimensions are 260x240x88 (WxDxH in mm.) for non-metrics out there, this approximates to 10.24x9.45x3.45 inches.
Chassis, bottom and top plate are made of zinc-plated steel.
Heat sinks and front bezel are made of an aluminum alloy.
Its design is uncomplicated. Matte black metal ‘modern industrial’.
It reminds of a high-end amplifier: one button on the front panel, large radiators on either side.
A sturdy build.
Each heatsink has thirty fins to enlarge radiating surface area. With these cooling specifications are rated 50 Watt @ 30 °C (86°F) - or 35 Watt at 50°C (122°F) ambient temperature.
What sets the 7700T CPU apart
Not that much.
The CPUs marked with a T in Intel’s Core series line-up are those that generally cannot handle the higher operating frequencies the 7700 vanilla or Ks can. Feature wise there are no differences. To the best of my knowledge, there are none other than its maximum operating temperature ceiling, operating frequencies and its lower TDP.
Lower operating frequencies allow for lower voltages. The CPU draws less power that way and as a consequence it generates less heat. Intel rated the 7700T at 35W TDP (Thermal Design Power) which is defined as the average heat in Watt a processor generates at default operating frequency with an Intel defined high workload on all cores. Any cooling system must minimally keep up with this rate.
The 7700 vanilla and the 7700K parts for example, are rated 65W and 91W respectively.
Heat generation varies greatly depending on load. The ‘thermald’ daemon shows figures of heat generation of the CPU if started with ‘--no-daemon --loglevel=debug’ arguments. In turbo I have seen the 7700T generates ~42 Watt but while writing this article its usage is very modest:
How is cooling done?
By phase change and capillary action of course! This design uses the brick and mortar of heat transportation - the heatpipe. Very elegant and efficient, the energy needed to transport the heat, comes from the waste heat itself. It is like unwanted guests showing themselves the way out.
Heatpipes are ‘vacuum’ copper (in this case) tubes with a modest amount of liquid inside. Heatpipes transport heat from the source of the heat, the processors heat spreader, to wherever the heat is dissipated into air, the radiators. Where the heatpipe gets hot, the liquid changes phase into gas and spreads trough the pipe, almost instantaneous. Where the tube is colder the gas will condensate. The inner walls of the pipe have a wick-like structure. The wick-like structure makes sure the fluid – in our case distilled water – finds its way back to the hot spot(s) of the pipe. Heat pipes are orders of magnitude more efficient than conductive distribution of heat by a solid rod.
The method may be brilliant, physics dictate the effectiveness of the cooling system, nothing else does.
Ambient temperature influence cooling power of the system. Turbo on all cores will exceed the 35 Watt TDP and if such load is prolonged, the computer may have no choice but to briefly lower operating frequency by the P-state mechanism or by injecting idle time in Linux's kernel scheduler with the Powerclamp driver. There are even some ‘last resorts’ but hopefully the machine will never need these.
Power usage
Yes, I promise to tidy up those cables. Really.
At the local hardware store I found the EM-16 v3, made by Alecto, €19,99.
When I sent a draft of this review to Jonathan of Atlast! Solutions for comments, he told me my power usage readings were higher than his. On the unsafe side of high, actually. Luckily we quickly found the cause: me. Jonathan sent me a document with more sensible BIOS settings than the admittedly, rather opportunistic values I had put into place.
At the same time Alecto answered some questions concerning the error margin on measurements on the EM-16. It is correct within a margin of +/-(3% plus 2W). They realize that this margin makes the meter a rather unsuitable tool in lesser power use conditions. Measuring how much power your electronics use in stand-by, for instance, is impossible with this device. Also, it does not take ‘power factor’ into account. This is too bad, but maybe somewhat excusable since consumers, at least where your author lives, are not charged for their ‘real power’ consumption, only for their apparent power use, and this is a consumer grade meter.
With this in mind:
When idle this system consumes 15 Watt +/-(3% + 2W)
Enjoying ‘Buke and Gase’ at 1080p on Youtube on Chrome unstable, the computer uses 23 Watt +/-(3%+2W)
During a ‘smallpt’ run – the tiny particle tracer that became a quick SMP benchmarking tool – the meter shows 62 Watt +/-(3%+2W).
I found the ‘stress’ utility in the Mint repositories and ran it as per the example in the man page:
# stress --cpu 8 --io 4 --vm 2 --vm-bytes 128M --timeout 10s
That made the small machine use 73W +/-(3%+2W)
Considering the modest power supply is rated at a maximum of 84 Watt, this stress test value is well within its capabilities.
Perfomance benchmarks
Phoronix-test-suite 7.2 is used for benchmarking. Tests are done on Linux Mint 18.2 XFCE.
Linux Mint 18.2 is based on Ubuntu 16.04 LTS and that may, in some respect, be regarded to be too old for the newer i7 7700T. Here, thermald proved too old.
I have installed thermal daemon 1.6 from git because it supports Kaby Lake and is the only thermald that isn’t as eager to throttle as the stock or Ubuntu 17.10 thermald versions are. The git version will let Kaby Lake temperature rise up until 74 C before shifting gears, the older versions were more conservative.
Any Linux kernel >= 4.10 has sufficient support for Kaby Lake CPUs but currently this machine happily runs 4.13-rc2 because I wanted to see if the newly added thermal management improvements made any noticeable difference and I forgot to change it to a more common kernel before I ran these tests.
I decided to duplicate the i5 6600K Skylake CPU tests ( 1508233-HA-INTELSKYL16 ) done by Michael Larabel of Phoronix because it emphasizes on computing performance rather than being system wide collection of tests, this makes the CPUs mentioned in the graphs more comparable and I am not that interested in its gaming performance anyway.
More benchmark results via this OpenBenchmarking.org page.
It appears there is little room for gains at the top. I assume, with -march=native, the most efficient way for modern CPU’s is to use specialized instructions, some SSE4 functions perhaps, to solve the bulk of this job faster than generic x86 instructions can. My reasoning here is the more complex logic will change not that much from generation to generation. In the sense of the more complex operations these Intel chips are all architecturally likely mostly the same and therefore their performance is much alike.
This is all hypothesizing because I was too lazy to profile the binary and disassemble it to test the idea. I welcome being educated on this matter if needed though!
In this small test collection the Atlast! Ultimate fanless does fine. It can compete with its higher clocked, power hungrier, older siblings.
Price, value and conclusion
At the time of writing, the 7th day of August 2017, this configuration is priced at: £1074 ~ €1183 or ~ $1269. Shipping costs depend on where you want the machine to go.
Admittedly your author perhaps suffers a little from confirmation bias which leads him to look for elements that support his choice rather than looking for angles to criticize his purchase. People do this to avoid the risk of feeling silly. I am going to try nevertheless.
- A fundamental critique stems from its design choice to passively cool. If your ambient temperatures are often high and you want consistent, maximum performance under high loads, then you will want to think about airflow and/or air cooling because under high load you will then see it lower its clock rate to keep within heat generation limits.
- One can set BIOS values that will make the computer exceed the power supply’s specifications.
So be cautious while tinkering.
Is there room to further improve the passive cooling paradigm?
Where is the low hanging fruit here? I have a few ideas.
- Make dissipation of heat more efficient?
I imagine thin vertical pipes could be added to the heat sinks so the heat sinks would then benefit from the chimney effect. This is for the case manufacturer, Impactics, to consider, test and experiment on.
- Improve heat transportation and distribution from the CPU to and along the radiator sinks?
Why not use longer heat pipes that rest on the CPU in the middle and reach out to both sinks?
- Make the heat conduct more efficient from the CPU to the heat pipes?
De-lidding the CPU and re-doing its thermal past is an idea.
I think one would benefit most from more efficient dissipation because the computer only starts throttling if the heat sinks are saturated.
Surely, the motherboard with its ‘Republic Of Gamers’ branding and built-in LEDs may give a juvenile impression. It also suggests something else. It is built to withstand some beating.
Intel plans to reuse the same CPU socket, socket 1151, for Kaby Lake’s successor as well. Might I want to upgrade to a faster low power part in a few years time, then that might well be possible.
Really, I am very happy with the computer. In some ways I am genuinely excited because of it’s absence of noise, it’s low power usage whilst at the same time being quite powerful.
This is just about right.
It may not move Greenland over to Denmark but it sure moved me.
I wish to thank Jonathan of Atlast! Solutions for helping me understand what normal thermald behavior looks like, for guiding me to more sensible BIOS settings and generally just being patient, honest and realistic.
I enjoyed writing and researching the various bits in this article and I sincerely hope you had at least a reasonably good read. Thank you! More information on the Atlast PC is available at AtLastSolutions.com.
26 Comments