Even before Ryzen 3000 was shown at CES as part of a demo, I had concerns about thermal performance for these CPUs. Way back when this website was in its earliest planning phases during November of last year (a full year ago now), Jim and I had a conversation about whether or not Ryzen 3000 would use chiplets. Believing they probably wouldn’t, I said:
Well they probably won’t be right up next to each other. Except for the CPU dies. Also the heat dissipation would be asymmetrical since the CPU dies would be shoved off to one side… Because you do want the CPU dies to be in the center [for optimal cooling]. And that’s why you don’t really have as much room as it looks like you might. If everything on AM4 was concentrated on one side there would be cooling issues and that’s not ideal.
I was skeptical on chiplets for two reasons: room and heat. Obviously there ended up being enough room for three dies on an AM4 sized CPU since the 3900X and 3950X exist. But my predictions about heat lived up to be much better; Ryzen 3000 suffers from reduced thermal performance. The 3700X and the 2700 have vastly different performance despite being 65 watt TDP CPUs (which is odd considering AMD’s TDP is based on temperature) and having similar power consumption.
Many will be quick to point out that the 3700X has a much denser compute die, which obviously makes heating the IHS more difficult. It is entirely natural to see reduced thermal performance on a new node due to a reduction in surface area if power consumption does not decrease. However, I believed that the orientation of the CPU die(s) is the greater issue. To find out whether or not my assumptions were right, I’ve done extensive testing.
RYZEN 3000 VS. 2000 THERMAL PERFORMANCE
First, let me just demonstrate that Ryzen 3000 does in fact perform worse at the same TDP and power consumption. This is my 3700X against my 2700. They’re both the same TDP according to AMD, which is not an indication of power consumption but an indication of thermal performance. They should have the same temperature more or less. Obviously, they don’t.
This test was performed using the NH-U12A and an IC Graphite Thermal Pad as part of my Ryzen 8 Core Retrospective article. The results clearly show worse thermal performance on the 3700X than the 2700 than there ought to be, especially considering the 3700X is on a much, much more efficient node than the 2700.
TESTING METHODOLOGY
Since I can’t just center the IHS on top of the compute die in my 3700X (which would theoretically give the most optimal performance), I tested this a few different ways. My goal was to achieve the best thermal performance possible from all the possible cooling solutions I can use. Things that could affect thermal performance may be, in my opinion:
- Where thermal paste is placed on the IHS (in the center as usual or right on top of where the CPU die would be)
- Air cooler vs. AIO
- Using a thermal pad instead of paste
Prior to testing, I predicted this would be the hierarchy of possible combinations, with 1 being the best and 6 being the worst:
- Air cooler and thermal pad
- Air cooler and thermal paste above the die
- Air cooler and thermal paste in the center
- AIO and thermal paste above the die
- AIO and thermal paste in the center
- AIO and thermal pad
My hypothesis is that the air cooler and a thermal pad will be the best cooling solution for Ryzen 3000 CPUs. My reasoning comes from my previous testing with thermal pads and paste on either AIO or air. Previous tests have indicated that AIOs perform very poorly when you use a thermal pad whereas air coolers perform just fine. I believed this is the case because AIOs depend on the heat source to be concentrated in the center of the cold plate otherwise the cooler will suffer reduced performance.
Since Ryzen 3000 does not have the CPU die in the center, where most of the heat is coming from, AIOs should perform worse. Air coolers should be able to accept heat further away from the center of the cold plate than AIOs can. Again, these were my assumptions going into this analysis.
With that in mind, the following parts were used:
Test Bench Configuration | Part Name(s) | |
CPU | Ryzen 7 3700X | |
CPU Coolers | Noctua NH-U12A | Arctic Liquid Freezer II 240mm |
CPU Thermal Interfaces | NT-H1 Thermal Paste | Innovative Cooling Graphite Thermal Pad (40 x 40 mm) |
GPU | EVGA GTX 1080Ti FTW3 | |
Motherboard | ASROCK X570 EXTREME 4 | |
Memory | G.SKILL Flare X 3200 MHz CL14 2x8GB | |
Storage | Samsung 850 EVO 250GB | |
PSU | Seasonic Focus 650 Watt Gold |
I only tested Prime95 and Blender for this analysis. In my past cooler reviews I have tested Civilization VI, but starting now I will be dropping that from my benchmarks. This is simply because Civilization VI (as well as other games) don’t really stress CPUs that much, so pretty much any cooler can handle a high end CPU in gaming.
I chose Noctua’s U12A air cooler to represent air coolers in general and Arctic’s Liquid Freezer II 240mm AIO (provided by Arctic, so thanks to them) to represent AIOs in general. This analysis is not a complete review or comparison of the Liquid Freezer II; that will be coming later.
The thermal application is also important for this test. When applying thermal paste to the center of the IHS, I used a small blob of paste that spread out almost over the entire IHS. It did not fully cover the corners though it did get to the edge of the IHS between the corners. As for applying paste above where the compute die would be (in the corner), I used a slightly smaller amount of paste so that it wouldn’t gush out onto the side of the CPU. This covered the corner the compute die was in thoroughly.
I collected CPU temperature data using MSI Afterburner and ambient temperature data using the HDE TA318 Thermometer.
Finally, I have updated the BIOS to ASRock’s most recent beta update which came out on November 1. This should have the latest clock speed tweaks and I have verified that 4.4 GHz is hit on my 3700X, though that isn’t really relevant to this testing since under an all core load all Ryzen CPUs will throttle a little in order to keep cool, stable, and safe.
THERMAL PASTE OR PAD?
Before I get into the results, you might be wondering at this point why I’m testing both thermal paste and pad. Most reviews have found that thermal pads, while convenient, have slightly worse thermal performance than paste. So why bother testing them if it’s just going to be the same margins but with slightly worse performance?
Well, thermal pads do typically have worse performance than paste, yes, and that is because the vertical heat transfer is not as good as thermal paste. Having good heat transfer vertically is important because that is the heat going from the CPU’s IHS to the cold plate of the cooler. I believe thermal pads suffer in this regard due air gaps; paste, being a viscous liquid, can fill these gaps much easier than a solid. Though a thermal pad is theoretically superior in thermal transfer, in practice this was not the case.
But what thermal pads excel at is horizontal heat transfer, which is transferring heat across the whole pad. Performance in this category is unparalleled compared to thermal pastes, even the very best. Though thermal pads struggle to transfer heat from the IHS to a cold plate at one point, it should have many more points of thermal transfer than your typical paste.
This is the reason why I theorized an air cooler with a thermal pad will perform the best. The advantage in getting the heat from the corner of the cold plate to the whole cold plate more evenly should be worth the penalty in vertical heat transfer, except in the case of AIO coolers, which I believe depend so much on that vertical transfer in the center of the cold plate that even on Ryzen 3000 CPUs a thermal pad would still perform worse than paste.
Anyways, let’s get to my findings.
THERMAL PERFORMANCE BENCHMARKS
PRIME95
I decided to run Prime95 on the default blend setting and run it for half an hour on each configuration. Since Prime95 has many different phases that each stress to a different level, I took the peak temperature rather than the average temperature. This is because the most stressful tests often times completed within seconds on the 3700X; this data does not represent one-off temperature spikes.
As we can see, there’s really no substantial difference between any of these configurations besides for the AIO and thermal pad which I correctly predicted would be in last. Though, it’s not far behind by very much. These results are more or less the same between the other 5 combinations, not even 2 degrees off from the best and the worst.
What is most interesting is that the two thermal paste applications on the AIO performed identically and on the air cooler it wasn’t much different. I wondered how much of the IHS my thermal application covered when applying the paste to the corner, and I was very, very surprised at what I saw:
The paste is barely covering a third of the whole IHS at the most, yet performance is almost as good or just as good as if almost the entire IHS was covered. I’m not entirely sure that this proves the orientation of the die is a significant factor in cooling, but I think it proves at the very least that in the areas where the IHS is uncovered, heat dissipation is negligible.
But, on the other hand, the thermal pad was clearly not the best performer on air, and my prediction there has been proven wrong. I will talk about the thermal pad more, but let’s see the results from Blender first.
BLENDER
As we can see, this is very similar to what we saw in Prime95, though this time the thermal pad is the clear loser overall. Additionally, paste applied in the center was better on the air cooler but worse on the AIO compared to applying paste in the corner above the die. This goes against my predictions but since these results are so close to each other, I think it’s not a stretch to assume they are more or less equal performers.
But it is certain at this point that the thermal pad is at least not the best for the air cooler. Why would that be? Well, I had theorized that the pad would be the best because it could take all of the heat concentrated in the corner of the IHS and spread it across the cold plate of the cooler more evenly. Well, I suppose I forgot that the heat from the IHS to the pad has to transfer vertically, where pads suffer, and that was enough of a bottleneck to adversely affect performance, though not by much.
Overall, these results are very similar and 4 of them are almost identical.
CONCLUSION
So, what did we learn from this test? Well, I think what we learned is that there are no tips, tricks, secrets, or techniques for getting significantly better thermal performance out of your cooler when it’s trying to cool a Ryzen 3000 CPU. Whatever the reason for that is, it is absolutely clear that Ryzen 3000 has very poor thermal performance.
Now, what could that reason be? Well, though I theorized that it mostly had to do with the orientation, now I’m not so sure. I think the orientation must have something to do with it; after all, covering almost the entire IHS vs. covering just a corner of it had practically no difference thermally. But, if the orientation was the sole reason for this poorer than expected performance, then surely the corner method would have been the best since it provided the best application of paste to the corner of the IHS.
I believe the larger factor is that Ryzen 3000 is simply very dense. We’re getting to the point where performance gains on new nodes are actually slowing down because the transistors are getting packed closer and closer together while consuming the same amount of power and thus producing the same amount of heat. Less surface area is going to mean less area to spread heat from, and consequentially that hurts thermal performance.
Concerning AIOs, it appears that, at least in the case of Arctic’s Liquid Freezer II, not all AIOs perform badly with thermal pads. I’m not sure why my Corsair H100i was such a poor performer when using a thermal pad; maybe it had to do with mounting pressure or the design of the cooler (which I now am less sure is the reason). Whatever the case is, it seems my concern about thermal pads and AIOs is probably unfounded, but I do think I will be testing both with thermal pads and paste in the future just to make sure.
I think this test also brings up a very good question for CPU cooler reviews in general: should people test coolers on a Ryzen 3000 CPU? On one hand, these two coolers are of a very similar grade and are high end. But on the other hand, I found that the H100i was a better cooler than the U12A when using thermal paste, and when I consider the Liquid Freezer II’s much deeper radiator, I have to wonder if there would be a difference on a CPU like the 2700 or 9900K where the CPU die is right in the middle of the IHS. Is there a point in getting a better cooler for a 3700X?
I don’t really have a good answer to this question. On one hand, there are certainly alot of people who are buying Ryzen 3000 CPUs and it is certain that the industry is going in the direction of chiplets and denser CPUs. On the other hand, there are also alot of people who don’t have Ryzen 3000 and are still on Ryzen 2000, 1000 and Intel CPUs that are all monolithic dies. So, which one is right to test? Chiplet or monolithic CPUs?
I think from here on out, I will be testing both my 2700 and 3700X for CPU cooler reviews just to cover all possible bases. It is certainly much more work on top of also testing with paste and pads, but more data is always better. The new chiplet design of Ryzen 3000 likely complicates things for cooler testing, but I’m a little thankful for it since it makes things more interesting in my opinion.
So, in summary, Ryzen 3000 has very strange thermal behavior that demands further testing which will be done in future reviews. It needs to be determined whether or not these CPUs actually benefit from higher end cooling solutions and, if so, which ones and why. The results of this analysis were certainly interesting, but ultimately not entirely conclusive.