MSI GeForce RTX 5090 SUPRIM LIQUID SOC Review – Grilling is now done away from case

1 - Introduction, overview and technical data 2 - Test system and equipment 3 - Teardown: PC, components and cooler 4 - Material analysis and TIMs 5 - Gaming performance 6 - Power consumption, transients and PSU recommendation 7 - Temperatures und clock rates 8 - Thermal imaging and modding 9 - Fan curves and noise 10 - Summary and conclusion

Thermography

An Optris PI640, a radiometric thermal imaging camera with a resolution of 640 × 480 pixels at a frame rate of up to 32 Hz, was used for the recording. It uses an uncooled microbolometer sensor based on VOx (vanadium oxide) and is optimized for industrial measurement tasks with a thermal sensitivity (NETD) of less than 75 mK. Emission correction is carried out using the PiConnect software supplied, which enables calibration, emission factor adjustment and thermal analysis in real time. For electronic components, an emission value of 0.95 is recommended, which has been assumed here as a reference value. The absolute temperature accuracy is ±2 °C or ±2 % of the measured value, depending on the calibration. The thermal separation between the GPU package, memory, VRMs and power connection is clear enough at this resolution to allow reliable conclusions to be drawn about the load distribution.

In idle mode, the MSI RTX 5090 SUPRIM LIQUID SOC exhibits extremely even and uncritical thermal behavior. The measured temperatures are in a very low range. The hotspot is only 36.3 °C, the 12V2X6 connection at 31.7 °C. The memory modules are tempered between 32.0 °C and 33.9 °C, the VRMs remain at around 33 °C, occasionally 35.4 °C. The GPU socket itself is at an inconspicuous 34.2 °C. The heat distribution is homogeneous, there are no recognizable heat nests or conspicuous local hotspots. This behavior indicates that the card is almost completely load-free and proves that it reliably uses passive cooling or very low fan speeds when idle.

The thermography of the MSI RTX 5090 SUPRIM LIQUID SOC shows significant differences in the thermal load of individual components in a direct comparison between gaming and silent mode, depending on the load scenario. In gaming mode (right column), the temperature profile of the card is lower overall, especially for the GPU itself and the power connector. In Torture mode, the 12V2X6 connector reaches a maximum of 64.7 °C, in the gaming scenario even only 62.5 °C. The VRM levels show a temperature of 67.8 °C (Torture) and 66.1 °C (Gaming), which indicates good heat dissipation. The memory modules are also in a moderate range of around 60 to 64 °C. The hotspot is slightly higher at 78.6 °C in the stress test, but remains below critical thresholds. Overall, the heat distribution indicates a balanced thermal load with effective cooling.

The picture is different in silent mode. Throttling the fan speed leads to higher temperatures at the same load. The 12V2X6 connection reaches up to 71.4 °C in the Torture scenario, which is already close to the specification-related load limit of the material. The VRM levels rise to up to 80.7 °C, and the hotspot is also visibly higher than in gaming mode at 83.8 °C (Torture) and 82.6 °C (Gaming). The GPU package itself also gets warmer, with values of up to 62.6 °C. It is noticeable that although the memory chips remain largely the same temperature, the overall heat distribution on the PCB is more homogeneous, which indicates a lower air circulation and slower heat dissipation.

Silent Mode	Gaming Mode

Overall, the thermography confirms the positive effect of the more aggressive fan profile in gaming mode, while silent mode is much more thermally demanding for individual hotspots. The situation is particularly critical at the right VRM stage and around the 12V2X6 connection, where local overheating effects can occur under permanently high loads without sufficient air circulation. A manual adjustment of the fan profile in silent mode could help to make better use of thermal reserves without significantly increasing the noise level.

Modding with Thermal Putty

In this case I used classic pre-cut pads, but thermal putty of the Thermal Grizzly TG Advance type, which is applied in paste form and spreads over the entire surface when mounting pressure is applied. In the MSI RTX 5090 SUPRIM LIQUID SOC, the factory-installed thermal interface has been completely removed and replaced with around 50 grams of TG Advance. Thermal putty of this type offers several advantages compared to fixed pads. The decisive factor is the ability to adapt three-dimensionally to the respective geometry of the components and heat sinks without leaving air pockets. Particularly for components with different heights or irregular surfaces such as VRMs, inductors or memory chips, a significantly more homogeneous heat distribution is achieved.

The material used is characterized by a medium to high viscosity combined with good flow properties under pressure. This enables precise dosing during application, with the defined quantity of 50 grams ensuring that even larger areas such as rear contacts and gaps to cover plates are reliably filled. The typical disadvantages of classic silicone pads such as incomplete contact, ageing cracks or poor reworkability do not occur here. At the same time, the material retains its shape even after several thermal cycles and does not dry out. Another advantage is the mechanical decoupling of sensitive components. The even pressure build-up minimizes the risk of punctual stresses on memory chips or SMDs, for example, which can occur with harder pads. Even in the vertical installation position, the material remains dimensionally stable and does not slip.

Thermographic images of the board show a significant reduction in hotspot temperatures, particularly in the area of the VRMs, as well as even heat distribution across the entire back of the board. This indicates that the putty was not only applied over the entire surface, but also with sufficient contact pressure. The combination of high thermal conductivity, good flow behavior and mechanical stability makes a thermal putty such as TG Advance a professional solution for high-end cards with complex heat sink layouts.

In contrast to classic thermal pastes, thermal putty is not intended for direct use between the GPU die and cooler, but explicitly for secondary thermal zones such as memory, voltage converters and rear panels. When properly applied, this solution can reduce hotspot temperatures by up to 4K compared to very good OEM pads.