Cooling Performance

At this point, we have a general idea of the performance of the provided configuration of the Comino GRANDO RM-S involving an AMD Ryzen Threadripper PRO 3995WX and four NVIDIA RTX A6000 GPUs all cooled via a custom copper loop. The single 360 mm Alphacool full copper radiator has no directly attached fans; instead, there are three Noctua NF-A14 iPPC 3000 RPM 140 mm fans placed slightly further inside and in line with the airflow field entering the chassis through generous openings cut into the side panels. The three SFX PSUs are also placed perpendicular to the radiator and between it and the fans to allow fresh PSU air intake, and we then get a glimpse at the patented tech Comino has utilized on the custom full-cover CPU and GPU water blocks for increased heat transfer away from the components and into the coolant for heat dissipation via the radiator.

In order to test this, I used a combination of GPU-Z and Comino's own monitoring system to log various metrics under a representative load as seen on page four. This was a "short" 2500 second test to confirm everything was up to par prior to commencing on to the much longer series of tests, and but a small test of the constant 24/7 high-performance throughput Comino promises. While there are dozens of different graphs I could present here, I chose the following as a function of time—average CPU core temperature, GPU core temperature for one of the used RTX A6000 units, the same GPU's VRAM temperature, GPU load, and GPU power draw. Testing was done in the same environmental chamber I use for all my relevant water cooling reviews, with an ambient temperature of 20 °C.


Keep in mind that the averaged values are causing some of the spikes and dips compared to the overall trend lines for the component temperatures as a function of time, but the biggest influencing factor here is the ever-changing load and associated power draw, as seen with the GPUs. The CPU cores hit a maximum of 74 °C with 20 °C ambient, and the GPUs didn't go beyond 62.6 °C for either of the four. There was also minimal discrepancy between GPU #1 and GPU #4 with the CPU and GPU system plumbed in parallel, which suggests the coolant flow rate was still high enough to prevent localized hotspots from making any practical difference. The VRAM isn't cooled as well as the GPU core or VRMs that have that massive deformational cutting-based cold plate design, but 64°C at 20°C ambient is still quite cool—note the more uniform trend here, which was arguably still on the rise before testing completed.


Comino rates the GRANDO RM-S at up to 1400 W of heat dissipation with a 20°C increment over ambient conditions, which is the norm for typical workstation and server cooling. Under these conditions, the rated TDP of the CPU and four GPUs alone exceeds the capacity by 80 W, yet I saw the CPU consistently boost past the stock boost clocks to 4.3 GHz, and the GPU clocks held steady with load, too. There was no throttling, and more impressively, fan speeds barely went past 30% PWM (~500 RPM). In more challenging loads, such as Luxmark, I did notice the fans ramp up to ~800 RPM before the benchmark ended. More typical 24/7 use cases will not load the system fully all the time, so the takeaway is that there is still plenty of cooling headroom and Comino has chosen to target 32 dB(A) noise output instead. As I mentioned before, I never noticed anything louder than the coolant gurgle during scientific performance testing. The GRANDO RM-L should be even better in this regard, with more cooling capacity for the hardware here, which is not usually offered inside the RM-S as the rated cooling capacity is officially being exceeded.