RTX 5090 runs 2°C hotter with thermal paste than liquid metal — Arctic MX-6 vs. stock LM tested

GeForce RTX 5090 Founders Edition
(Image credit: Nvidia)

Nvidia's RTX 5090 Founders Edition is the first Founders Edition graphics card equipped with thermal metal. TechPowerUp tested the RTX 5090's liquid metal by benchmarking the GPU with its stock liquid metal application against the GPU re-pasted with conventional thermal paste. The outlet discovered a two-degree Celsius difference in temperatures.

In TechPowerUp's thermal testing, the RTX 5090 Founders Edition averaged 77.6 degrees Celsius. After replacing the default liquid metal material with Arctic MX-6 thermal paste, the outlet saw a 1.8-degree Celsius temperature elevation, with an average operating temperature of 79.4 degrees Celsius.

TechPowerUp's testing confirms that the RTX 5090 Founders Edition can be swapped to a "cheaper" thermal paste material if necessary. The 1.8-degree difference in temperature between both thermal interface materials is virtually indistinguishable from a real-world performance standpoint. However, slight deviations in fan speed and/or room temperature can potentially cause a larger difference in GPU temperature compared to swapping the RTX 5090 Founders Edition from liquid metal to thermal paste.

The slightly higher 1.8-degree temperature from the thermal paste application also barely transitions the GPU closer to its maximum operating temperature. TechPowerUp discovered that Nvidia raised the throttling point on Blackwell from 90 degrees Celsius to 83 degrees Celsius, as opposed to 83 degrees Celsius on Ada Lovelace. This is just the throttling point, so it stands to reason that Blackwell can potentially operate up into the mid-90 degrees Celsius range without crashing, but this has not been confirmed.

On the RTX 5090 Founders Edition, it remains a mystery when Nvidia's liquid metal application needs replacing. Nvidia has implemented a protective seal around the GPU die that protects the PCB from potential spillage and creates a vacuum seal around the GPU, allegedly stopping the liquid metal from oxidizing. This could significantly improve the liquid metal's longevity, making a possible thermal paste replacement unnecessary.

Like motor oil, thermal paste and metal have a limited shelf life, specifically when under constant use, such as a system operating 24/7. After numerous thermal cycles, thermal paste and liquid metal can dry up, losing their thermal conductivity.

Aaron Klotz
Contributing Writer

Aaron Klotz is a contributing writer for Tom’s Hardware, covering news related to computer hardware such as CPUs, and graphics cards.

  • User of Computers
    Funnily enough, I'd expect PTM to perform better than liquid metal. I don't know why they used LM instead of PTM.
    Reply
  • User of Computers
    paste will dry out extremely fast compared to LM, so I'd expect this delta to increase over time as the paste pumps dries out.
    Reply
  • TCA_ChinChin
    User of Computers said:
    Funnily enough, I'd expect PTM to perform better than liquid metal. I don't know why they used LM instead of PTM.
    I imagine its because I think LM is still better as temperature increases although PTM lasts longer. I suspect that why they engineered such a redundant solution to prevent LM leakage.
    User of Computers said:
    paste will dry out extremely fast compared to LM, so I'd expect this delta to increase over time as the paste pumps dries out.
    Agreed, although i wonder if the extra seals they made for the LM will also help with keeping the paste moist over a longer period of time.

    The Nvidia engineer that did a breakdown with GN specifically mentioned the triple seal was made airtight, not just LM proof.
    Reply
  • HyperMatrix
    LM doesn’t make your chip run cooler. It helps transfer heat to the cold plate/block/shroud faster. If your cooler design gets overwhelmed with heat soak, it will reduce the efficiency/benefit of Liquid Metal.

    Essentially the bigger the gap between the temperature of the die and the cold plate pulling the heat away, the more of an impact LM will have. PTM would not have done better. Because the issue here isn’t the TIM. It’s the cooler itself. Which, while beautiful, is still only 2 slot and outdone by the AIBs I’ve seen reviewed today so far.
    Reply
  • Amdlova
    User of Computers said:
    paste will dry out extremely fast compared to LM, so I'd expect this delta to increase over time as the paste pumps dries out.
    I have here a rx 570 msi wich have almost 30.600 Hr original paste.
    today thermal past look dry but they are working in perfect condition
    Reply
  • bit_user
    HyperMatrix said:
    LM doesn’t make your chip run cooler. It helps transfer heat to the cold plate/block/shroud faster.
    It's the same thing. Moving heat out of the GPU chip faster results in it running at a lower temp.

    HyperMatrix said:
    If your cooler design gets overwhelmed with heat soak, it will reduce the efficiency/benefit of Liquid Metal.
    At steady state, heat is transferred into the heatsink at the same rate the heatsink can dump it into the air. However, that steady-state rate is only fixed for a given die and air temperature.

    The concept of "heat soak" is only something that applies during the warm-up transient. Before the heatsink reaches its steady state temperature, it can absorb heat more rapidly. Once it hits that steady state, the rate at which it can absorb heat from the die will level off somewhat, but it doesn't get "overwhelmed".

    Even once a steady state has been reached, it doesn't mean the heatsink cannot absorb heat at a faster rate, but it will only do so if the die gets hotter. If/when the die temperature increases, the rate of thermal transfer into the heatsink will still increase and the rate at which the heatsink dumps that heat into air also increases (i.e. because the heatsink will also get hotter). This follows from Newton's Law of Cooling, which stipulates that the rate of heat transfer is proportional to the temperature difference.

    HyperMatrix said:
    Because the issue here isn’t the TIM. It’s the cooler itself. Which, while beautiful, is still only 2 slot and outdone by the AIBs I’ve seen reviewed today so far.
    Sure, a better heatsink could benefit more from a reduction of that bottleneck between the GPU and heatsink.
    Reply
  • HyperMatrix
    bit_user said:
    It's the same thing. Moving heat out of the GPU chip faster results in it running at a lower temp.


    At steady state, heat is transferred into the heatsink at the same rate the heatsink can dump it into the air. However, that steady-state rate is only fixed for a given GPU and air temperature.

    The concept of "heat soak" is only something that applies during the warm-up transient. Before the heatsink reaches its steady state temperature, it can absorb heat more rapidly. Once it hits that steady state, the rate at which it can absorb heat from the GPU will level off somewhat, but it doesn't get "overwhelmed".

    Even once a steady state has been reached, it doesn't mean the heatsink cannot absorb more heat, but it will only do so if the GPU gets hotter. If/when the GPU temperature increases, the rate of thermal transfer into the heatsink will still increase and the rate at which the heatsink dumps that heat into air also increases (i.e. because the heatsink will also get hotter). This follows from Newton's Law of Cooling, which stipulates that the rate of heat transfer is proportional to the temperature difference.


    Sure, a better heatsink could benefit more by reducing that bottleneck between the GPU and heatsink.

    You've said the same thing I've said but in a more drawn out fashion. The reason I commented at all was that the article seemed to give the impression that liquid metal was minimally beneficial, whereas the problem as I already state, wasn't the liquid metal, but the cooler design. These are essentially the main considerations:

    Liquid:GPU die <-> Cold plate delta Cold Plate <-> Loop Temp Loop Temp <-> Ambient
    Air:GPU die <-> Cold plate delta Cold Plate <-> Fins Fins <-> Ambient
    If Ambient is 20C, Cold Plate is at 60C, Fins/Heat pipes are at 40C, and GPU is at 80C, the liquid metal only affects the portion of cooling from the 80C GPU to the 60C Cold Plate. Or 33% of the overall temperature difference from ambient. And the smaller the gap between the GPU die temp and cold plate temp, the less of an impact Liquid Metal or any other TIM can have.

    Whereas with an Optimus block on a 700W 3090 Kingpin, using Liquid Metal, I was able to maintain a 7C delta from ambient to GPU die.

    Liquid metal is phenomenal. And I wanted to correct any misunderstands readers would have, attributing the 2C performance difference to liquid metal, as opposed to the rest of the cooler design. As you can see from some of the other comments here.
    Reply
  • JohnyFin
    Funny that people believe that liquid metal is much better. Is not. Thermal solution is similar to creation process of very low temperatures near 0 Kelvin. Is about nano particles which sorrund big particles in vacume environment. There is similarity. Liquid thermal paste can be even worse than normal paste.
    Reply
  • bill001g
    JohnyFin said:
    Funny that people believe that liquid metal is much better. Is not. Thermal solution is similar to creation process of very low temperatures near 0 Kelvin. Is about nano particles which sorrund big particles in vacume environment. There is similarity. Liquid thermal paste can be even worse than normal paste.
    Because most consumers are idiots and only read headlines and put no effort into learning anything. Redbull had to stop saying "it gives you wings" because some idiots kids jumped off a house roof and sued them.
    Reply
  • Tethgar
    HyperMatrix said:
    LM doesn’t make your chip run cooler. It helps transfer heat to the cold plate/block/shroud faster. If your cooler design gets overwhelmed with heat soak, it will reduce the efficiency/benefit of Liquid Metal.

    Essentially the bigger the gap between the temperature of the die and the cold plate pulling the heat away, the more of an impact LM will have. PTM would not have done better. Because the issue here isn’t the TIM. It’s the cooler itself. Which, while beautiful, is still only 2 slot and outdone by the AIBs I’ve seen reviewed today so far.
    what an utterly pointless comment lol. I hate when people state such obvious facts and present it as some sort of intelligent and unique thought. Now tell us about how fans cool worse in warmer environments 🥴
    Reply