After receiving the prototype, one of the things we were most excited to start working on was a robust analysis of Project Nova's thermal performance. After all, now that we had the physical case and a bunch of computer hardware at our disposal, we could finally and definitively discover how well various hardware configurations would perform - as well as how certain arrangements of fans could influence component temperatures. That, in turn, would speak volumes about what Project Nova is capable of supporting, and could even inspire some design changes that would offer further improvements!

For those that have only recently heard of Project Nova (and our work), though, this analysis actually isn't new for us - when the digital model of the prototype was being created, James made early physical models using an old computer chassis and some cardboard panels. You can read more about that testing in an earlier blog post, but to summarize, that cheap-and-rapid prototyping let him quickly try a bunch of fan configurations, and figure out which ones would perform optimally. That analysis ultimately helped determine where fan mounts and vents would be placed in the digital model to be manufactured.

The testbed that James used while creating the digital model of Project Nova. Most of this equipment would be re-used once he began testing of the prototype.

The testbed that James used while creating the digital model of Project Nova. Most of this equipment would be re-used once he began testing of the prototype.

With this past work, and present opportunity, then, it was imperative that we perform similarly thorough testing, and see what we could learn about the case we had made. And I'm happy to report that we have completed this testing, and conducted considerable analysis of the results! As such, I wanted to share a taste of some of that work today.



Before I dive into a sample of the data, though, it's important that I explain the hardware and the methodology employed throughout this process. So, let's go through it all real quick.

Just one of the many configurations we've installed and tested inside Project Nova. Here, two GTX 980s flank a 5930K cooled by Noctua's NH-C12P.

First off, we have all of the computer hardware that was used across the tests, inclusive of components (CPU, GPU, motherboard, and so on), coolers, and fans. These included the following:

  • CPU: Intel Core i7 5930K
  • CPU Coolers: NZXT's Kraken X41 with Noctua's NF-A14 industrialPPC-2000; Noctua's NH-C12P and NF-A15
  • Motherboard: Gigabyte GA-X99M-GAMING 5
  • Memory: Crucial Ballistix Sport 32GB
  • Storage: Samsung 840 EVO 256GB
  • Video Card(s): EVGA GeForce GTX 980 (single and SLI), ASUS GeForce GTX 780 DirectCU II
  • Power Supply: Silverstone SX600-G 600W SFX PSU
  • Fans: Noctua's NF-P12's for the front and bottom; Prolimatech's Ultra Sleek Vortex 12 for the top; and Noctua's NF-A9x14 for the rear

Note: CPU and GPUs were run at stock speeds. The CPU fan was plugged into the motherboard header, and the case fans ran at 9V off an external power supply - with the exception of the rear fan, which also had an inline Noctua voltage reducer.

We also have the metrology equipment and software that James used to measure temperatures and such, during idle and load testing. This included:

  • A Brand Electronics 4-1850 Power Meter, which measured the AC power draw of the system.
  • An Amprobe TMD-52 thermometer, which was attached to the middle of the PSU's exhaust grill, and the top of the case, so as to measure exhaust temperatures of the PSU and the case itself.
  • Reed AT-6 non-contact tachometer, which measured the PSU's fan RPM.
  • And finally, HWMonitor, which was used to capture CPU/GPU temperatures, and GPU fan RPMs (as it doesn't support our motherboard correctly, we did not pull CPU fan readings).

Finally, we had specified procedures and loading software, employed across all the configurations that were tested:

  • For the idle tests, configs were left running with only HWMonitor open. 
  • For the load tests, synthetic benchmarks were chosen since we wanted to push the components to their thermal extreme: FurMark at 1920x1080 and 2xMSAA for graphics, along with Prime95's "Blend" setting running on 10 threads (leaving two threads free for FurMark and HWMonitor).
  • For both idle and load tests, measurements were taken after 10 minutes of sustained idle/load states. Ambient temperatures were measured with a room temperature monitor during each run.


By the end of all of this thermal testing, James had measured the performance of twenty-two combinations of fans and hardware, across twenty-four unique tests. and forty-eight individual runs. Needless to say, that's a pretty big pile of data to have! So, what does this information actually look like?

Well, it looks something like the above worksheet! In this particular view, each column represents a configuration and its test results, both at idle and load. Each config # corresponds to a specific combination of hardware and fans that were used, and the ambient temperature for each run is listed so that it can be controlled across any comparisons.

In fact, it's with those comparisons that the fun really begins, since that's what the data is really all about - understanding what's effective in the current cooling scheme, and what isn't. Which, with some basic statistical knowledge and simple arithmetic, we can uncover, as we've done in additional worksheets like the one below:

In this example, we're comparing pairs of configurations that are completely identical, except for the presence of a rear-mounted fan. The colored columns reflect the change in temperatures (Δt)   controlled for differences in ambient (Δa)  that occurred from removing the fan. Which is to say that, if the GPU temperature is marked as .9°C, removing the fan caused a .9°C increase in GPU temps. Conversely, if the CPU temperature is marked as -1.9°C, then removing the fan caused a 1.9°C decrease in CPU temps.

Thus, in isolation, each pairing of configurations helps to show how well a particular fan actually works, given other fans and hardware! Furthermore, by calculating a weighed average across (Δt - Δa) for all pairs, we can also show how well that fan works overall, for all tested builds.

That average for the rear fan looks like this:

...Which tells us that the rear fan gives us ~1-2°C temperature improvements at idle, and negligible improvements at load. Maybe not surprising, given the fan's size, but now we have those suspicions backed up by data!

Better yet, though, is that we have similar analyses for the larger fans on the top and front, as well as for the combinations of fans and drives that can be mounted on the bottom (shown below). Beyond the efficacy of individual fans, this particular worksheet was helpful in demonstrating the thermal implications of bottom-mounted hard drives on other components:



Armed with this new information, we are now actively investigating some design changes to Project Nova that will help to improve the functionality and thermal performance of the case - all thanks to this analysis. We're also, overall, quite happy with the existing results we saw across tests: although there's always room for improvement (especially in small cases like Project Nova), the thermal performance we measured was good to excellent, particularly during the strenuous load tests.

Finally, I'm pleased to say that we've published every bit of the collected data and subsequent analysis on a public spreadsheet that can be accessed here, for everyone in the SFF community (and beyond) to see! In part, we've provided all of this since we think it's important for that information to be available, but we're also keen to maintain the open dialogue and discussion that we've embraced ever since this project began on the forums. Dumping the data we had, for all to access and use, was a natural extension of that.

As always, feedback and thoughts are appreciated, and we invite anyone to share them on our forum thread on SFF Forum!