Not over-hyped: Honeywell PTM7950 in a lab test and a real game changer for graphics cards

1 - Introduction and technical data 2 - Test methods and equipment 3 - Thermal resistance, minimal thickness 4 - Thermal conductivity and simulations 5 - Repairing and measuring a GeForce RTX 4080 6 - Material analysis 7 - Summary and conclusion

Many manufacturers have been experimenting with the Honeywell PTM7950 thermal pad and graphene pads on graphics cards for some time now. With good success, as AMD board partners in particular (and AMD on the MBA cards) or MSI prove. Paste or no paste, the load on graphics cards in particular is far too heavy a burden for many normal pastes in the long term. In addition to the horizontal forces caused by the very different expansion coefficients of the chip and heatsink (which is why, for example, 1000 cycles are tested), it is above all the higher temperatures and direct contact with the substrate without an extra heat spreader that come into play here. Ageing effects such as pump-out, drying out and simple chemical degradation are not something that today’s GPUs really like in the long term.

Now these parts are not so easy to buy in normal stores, but copycat pads like the Heilos from Thermalright do exactly the same thing. Of course, I will also be testing these pads across the board and I know that many established suppliers are currently working on their own offerings. This will certainly simplify things a lot soon. What I plan to do today consists of two sections. On the one hand, I will test the pad in TIMA5 like a normal paste and on the other hand, I will finally cure the Manli GeForce RTX 4080 Gallardo, which has already been tortured several times. I’ve put together the most important links for you to get you started.

An important foreword on “bulk thermal conductivity” and false marketing promises

I am now deliberately prefacing this with two quotes that not only speak to me from the heart, but also absolutely agree with my laboratory measurements. Conventional pastes cannot achieve much more than 4 to 5 W/(m-K) under the usual conditions on a GPU or CPU in terms of layer thickness, temperature and pressure. Because these quotes are honest and unfortunately correspond to reality, I will use this part from now on as a standard quote in all paste tests of all manufacturers and put it in front. You can’t bend physics.

For those who wonder how you can even arrive at figures above this limit, it should be noted that test conditions can certainly be adapted to achieve astronomically high figures. However, testing in a bucket has nothing to do with reality, even if a known measurement method is used. Without knowledge of the exact circumstances, such values are completely misleading and meaningless. You could give many suppliers credit for simply not knowing any better and just copying the OEM’s data sheets, but it doesn’t make misleading consumers any better.

The previous “reference paste” will only be used for comparisons in the curve diagrams until the automatic, database-based chart generation is ready (soon). Then the best pastes will be used fully automatically in the charts for comparison (can be selected and deselected as with the fans at the time). The Alphacool Apex was and is the tool of choice for me for long-term comparisons, because I have been able to check its behavior for over 12 months. This point will be dropped when we switch to the database.

Real long-term simulations (3000 hours in 1000 cycles up to 90°C) are not feasible in terms of effort. That is why I can only make predictions, but I want them to be understood as such. It is virtually impossible to make scientifically sound statements in just a few days. Yes, you can identify a trend and scale it as a forecast based on existing data, but this is not something that allows you to make really reliable statements. Therefore, I am sorry to say that I have to leave out this important point. However, as far as time permits, I will take community feedback into account and add the comments and long-term experience of third parties to the database as a note in due course, if it seems necessary. In both a positive and negative sense. However, this is a subjective value that has no place in a comparative database.

Unboxing

In addition to the 0.25 mm thick pad, you also get a few adhesive aids for applying smaller sections, which is really helpful (especially the 1 cm x 1 cm on the TIMA5 are very fiddly). The whole thing arrived from China (AliExpress) within 2 weeks as a letter (unfortunately you have to be that patient), but the pad was neatly packaged in a durable box.

The problem with these pads is removing the second protective film and placing it on the test body without destroying it. Even if you have measured everything precisely beforehand, simply leave a little more margin for safety and later remove anything you don’t really need with a scalpel or spatula. Freezing in the **** compartment of the freezer at below -18 °C also helps. However, this not only makes the pad firm, but also brittle. Yes, the thick and thin film both come off better, but with a bit of bad luck you end up with cracks in the pad. And that’s exactly what nobody needs. That’s why the safest way is to use a pre-assembled version with adhesive strips like the one I received. I’ll write more about this later, but I wanted to explicitly point out the useful accessories.

Technical data

Let’s leave the marketing aside and look at the technical data of this paste, which is available in various containers and sizes. We can see that the information on thermal conductivity is missing (red). The green shaded columns contain my measured values, the gray shaded columns contain the completely missing manufacturer’s data.

Further links and basics