Power Draw, Cooling, and Efficiency: AMD Ryzen 9000 Series Processors

Introduction

With the recent Ryzen 9000 series of desktop processors, AMD has bucked the recent trend of ever-increasing power draw on processors. At the top end, its Ryzen 9 9950X maintains the same TDP (Thermal Design Power) as the last-gen 7950X, while every other part (9900X, 9700X, and 9600X) saw reduced TDPs compared to the last-gen counterparts. This reduction in TDP promises reduced temperatures, lower power draw, and better efficiency.

However, this lower TDP for AMD processors is not all that new. The Ryzen 9000 processors’ reduced TDP actually matches the TDP for the otherwise-identically-named non-X SKUs from the Ryzen 7000 series of processors. Additionally, AMD enabled “ECO mode” toggles in its BIOS, which reduced most of the CPUs down to 65 W TDP. With its previous generation, AMD also released its X3D variants of chips, which, due to increased cooling needs from the 3D V-Cache technology, also featured lower TDPs.

Below, we’ve created a table of current and last-gen Ryzen processors (in addition to the Intel 14900K) to highlight the rated TDP of the processors.

Processor	TDP (W)	Cores	Frequency (GHz)
AMD Ryzen 9 9950X	170	16	4.3 / 5.7
AMD Ryzen 9 7950X	170	16	4.5 / 5.7
AMD Ryzen 9 7900X	170	12	4.7 / 5.6
Intel Core i9-14900K	125 (253)	8P+16E	3.2 / 6.0
AMD Ryzen 9 9900X	120	12	4.4 / 5.6
AMD Ryzen 9 7950X3D	120	16	4.2 / 5.7
AMD Ryzen 9 7900X3D	120	12	4.4 / 5.6
AMD Ryzen 7 7800X3D	120	8	4.2 / 5.0
AMD Ryzen 7 7700X	105	8	4.5 / 5.4
AMD Ryzen 5 7600X	105	6	4.7 / 5.3
AMD Ryzen 7 9700X	65	8	3.8 / 5.5
AMD Ryzen 5 9600X	65	6	3.9 / 5.4
AMD Ryzen 9 7900	65	12	3.7 / 5.4
AMD Ryzen 7 7700	65	8	3.8 / 5.3
AMD Ryzen 5 7600X3D	65	6	4.1 / 4.7
AMD Ryzen 5 7600	65	6	3.8 / 5.1

In this article, we will be exploring the power draw, temperatures, and efficiency of two of the new Ryzen 9000 processors: the 9950X and 9700X. We will also be including the 14900K at 125W and 253W as a point of comparison in addition to the 7950X and 7700X. While we wish we could look more broadly at all the above processors, there are lots of great reviewers who have largely already done so, and we encourage interested readers to check them out—Gamer’s Nexus and Anandtech have particularly in-depth reviews. If you want to find out more about the new Ryzen 9000 CPUs in general, we also recommend checking out our main AMD Ryzen 9000 Content Creation Review.

TDP = Power, right?

To quote Betteridge, “no”.

TDP, or Thermal Design Power, is the maximum heat generated by a computer component, measured in Watts. Historically, a TDP rating for a CPU, GPU, or other component told you the maximum power draw that component would make. This set an upper bound for the amount of cooling and power supply you would need for a component. However, some manufacturers now use that term somewhat more loosely these days.

Specifically, AMD determines its TDP values based on a variety of factors that it selects for each processor. Although in some workflows, the processors may average a power draw right around the rated TDP, under heavy workloads, the processors can draw notably more than the rated TDP.

This variance between the TDP and max power draw relates directly to the maximum heat generated. Despite AMD’s claims otherwise, Watts are Watts, and the amount of heat generated in a CPU is equal to the amount of power drawn. Thus, any rating for heat generated must necessarily be the same as the power draw. We even had fun defending the first law of thermodynamics more than 10 years ago in our Gaming PC vs. Space Heater Efficiency article.

Instead of using the actual power draw, AMD uses TDP to indicate what sort of cooler is sufficient for their product. Unfortunately, that does not align with how most cooler manufacturers understand Watts (Noctua no longer even publishes a “TDP capability” for their coolers; see their reasons why here), so it is largely unuseful. Intel has historically done the same but, nowadays, tends to quote base power (unfortunately still often referred to by TDP) and maximum turbo power, both of which are adhered to properly when set in the BIOS.

Because of this, one thing that we measure when doing our testing is the maximum CPU power draw. For AMD, under all-core workloads, the maximum power draw is sustained for the entire workload, meaning that it is functionally the average power draw. Under lightly threaded workloads, this is not the case. For Intel, the average power draw is always equal to (or less than) the configured PL1, while the maximum is usually equal to PL2 unless it is a very light workload.

Power Settings

As we’ve discussed elsewhere, there are a lot of technologies on modern CPUs which serve to increase performance. While some of these are classified as overclocking, others are not. For this testing, following our standard for our reviews, we disable overclocking features but keep most other performance-boosting features enabled.

Specifically, for AMD, we disable ASUS MCE (Multicore enhancement) and AMD PBO (Precision Boost Overdrive) but keep Precision Boost enabled. We also run memory at the maximum supported JEDEC frequency.

For Intel, we run the processors using the “Intel Defaults” profile in the BIOS and manually ensure that PL1 = 125W, PL2 = 253 W, Tau = 56S, and ICCMax = 307 A. For this testing, we also ran one set of benchmarks with PL1 = 253 W, which is often the default on many motherboards. We disabled ABT and MCE and ran the memory at the maximum supported JEDEC frequency.

Test Setup

Our testing setup is basically identical to the standard test platforms we use in all of our CPU reviews. In the past, we have found that the Noctua air coolers have minimal impact on performance given our power settings, so we didn’t do any AIO testing this time. For CPUs, we tested the Ryzen 9 9950X and 7950X—the top-end Ryzen desktop processors, both of which have a TDP of 170 W—as well as the Ryzen 7 9700X and 7700X. The latter two processors occupy the same spot in the product stack, but the Ryzen 7 9700X saw a 40 W TDP reduction from the last-gen counterpart. Although we didn’t test other Ryzen parts, we expect the 9700X to be representative of the other TDP-reduced processors. For comparison’s sake, we also tested the Intel i9-14900K at both PL1 = 125 W and PL1 = 253 W. Although we won’t spend much time directly comparing Intel to AMD, it can be useful as a reference to see just how much dropping power consumption can affect efficiency.

For benchmarks, we looked at three from our standard suite to represent various types of workloads. We choose Photoshop as a representative “real-world” workflow that is relatively lightly threaded. This can stand as a good proxy for a lot of applications that only make use of a few CPU cores most of the time, and which have a variety of types of work in them, from waiting on something to finish to working in real-time. As a representative workflow for heavily multithreaded applications, we settled on our Unreal Engine code compilation benchmark. This stresses CPUs to use as many cores as possible in order to finish the compilation in as little time as possible. Finally, we also tested with Cinebench, which, while somewhat more synthetic, is an industry-standard CPU benchmark that is fairly taxing on processors.

Power Consumption

Starting off with power draw, there are a number of metrics with which we can evaluate the processors. In most cases, the required watt-hours to complete the benchmark is the most useful number, as it represents the total energy required to do a given set of work. Most applications are “fixed-work,” meaning that the hardware is in use until a task has been completed. Thus, a component can reduce the amount of energy required to do the task by either drawing less power while executing the work or completing the work more quickly. We represent a “fixed work” scenario in both Photoshop (a light workload application) and Unreal Engine code compilation (a heavy workload application). However, in some cases, workloads are instead of fixed time. This may be representative of a workload that will be run 24/7, but more frequently, will be a synthetic workload designed to test components or systems. Our Cinebench tests are an example of one of these. For fixed-time workloads, a better measure is simply the average power consumption during the workload, although total watt-hours are directly proportional to this.

The above charts look at both the total watt-hours required for completion and the maximum Watts drawn. Charting the max Watts is tricky because the two processor manufacturers do not boost in similar ways, but it is interesting from the standpoint of knowing the maximum TDP (or thermal load) a cooler would theoretically need to be able to handle.

Starting with Photoshop, a light, fixed-work workload, we found that the most efficient processor was the new Ryzen 7 9700X. Importantly, it was somewhat more efficient than the 7700X by about 1.7 Wh or 10%. This is a solid efficiency gain in the workload gen-on-gen, although much lower than you may expect given the TDP drop; we can see it in the max watts drawn, with the 9700X pulling nearly 50 W less than the 7700X. For the higher-end Ryzen 9 9950X, however, we did not see a drop in energy consumption required to complete the benchmark compared to the previous-generation 7950X. Both processors take about 26 watt-hours to finish, and both pull around 190 watts at peak. This suggests that much of the efficiency gains from Zen 5 are either due to the configured TDP drop or else are being suppressed by more boosting.

Moving on to the Unreal Engine code compilation, we find that the 9700X uses far less energy than the last-gen 7700X by about 25%. Again, this is less than the TDP difference of about 40%, but impressive nonetheless. Much like Photoshop, the 9950X sees little overall change in energy required, in this case around 3%. Interestingly, the 9950X and 7950X both use less energy to complete the benchmark than the 7700X—although the latter draws less power on average, the former two are sufficiently fast to outweigh the higher power draw.

Finally, Cinebench is the only fixed-time benchmark we looked at. In this case, the total energy used really just depends on the average power draw. For most of the benchmark, the average and maximum power draw are nearly the same, although there are lower power sections while a render is initializing or finishing. This does mean that the overall efficiency gains are lower than relative TDP differences, but the 9700X still uses less power than the 7700X and the 9950X less than the 7950X.

One big takeaway from this testing is that despite the TDP staying the same on the 9950X and 7950X, the 9950X has a lower power draw; it is closer to its TDP than the 7950X. Combined with the data from the 9700X, it appears as if this generation saw a “free” 20 W power reduction across the board, in addition to other TDP changes. Ultimately, reducing power draw means less performance at a given efficiency level, though, so with the slightly increased efficiency of Zen 5, we end up seeing little actualized efficiency gains at a fixed TDP.

To summarize, the new Zen 5 parts do draw less power on average, but they only appear to require less energy to complete a workload when the TDP is reduced. That is to say, a 65W version of the 7700X (say, the 7700) would likely have very similar energy requirements for a given workload as the 9700X.

Performance per Watt(-hour)

Another metric on which we can evaluate efficiency is performance per watt-hour (or watt). In the previous section, we asked how much energy was required to finish a workload. Now, we ask how much performance we can get from a unit of energy (or power). In a workload that is wholly fixed-work, this will be basically the same as the above metric. However, if the workload has fixed-time components, a quality or score output, or is time-sensitive, these two metrics can diverge. The two most important benchmarks for this test were Cinebench and Photoshop. Cinebench is fixed work, so we are looking at how many points you get per Watt, while Photoshop outputs a final score that is not wholly dependent on the time taken to complete. Our Unreal Engine compile benchmark is harder to examine in this metric, so we have omitted it for now.

Starting with Cinebench, we find that the 9700x achieves about 73% more performance per watt than the 7700X. This is because while both of them scored fairly identically, the 9700X pulled an average of 57 W less (87 W v 144 W) than the 7700X. This is a very impressive increase in performance per watt, which is made possible both by architectural improvements and a lower TDP. Typically, processors have an efficiency curve where each additional watt of power draw provides less performance improvement. This is demonstrated by the 9950X, which is only about 13% more efficient than the 7950X, largely due to having the same performance while drawing 24 W less.

Moving on to Photoshop, we are technically looking at performance per watt-hour. Again, a Watt-hour is a unit of energy (like Joules) and a Watt of power. The 9700X here attains about 20% higher performance per watt-hour due in part to a lower power draw and, in part, to a higher benchmark score (by around 8%). This is less than we saw from Cinebench. While some of this is because Photoshop has some sections that are fixed-work, part of this is also because the 9700X and 7700X have much more similar power average power draws (58 W and 60 W, respectively). Photoshop is a light enough benchmark that the CPU frequently only uses a few cores, at which point the 7700X and 9700X draw nearly that same amount of power due to boosting algorithms. We see this more clearly with the 9950X, which is only about 10% more efficient by this metric than the 7950X.

CPU Core Temperature

Finally, we wanted to look briefly at the processor temperatures during the tested workloads. For this, we pulled both the average and maximum recorded CPU package temperatures during the benchmark runs. Although we have included both Intel and AMD in these charts, unlike the previous sections, the results between the two manufacturers are less directly comparable–they don’t measure in the same relative locations and so have different deltas between the probe location and any individual core, hotspot, or IHS. Typically, we expect the average temperature to correlate with the average power draw and maximum temperature maximum power draw. However, IHS and TIM properties can affect the results as well.

In Adobe Photoshop, we found that the 9700X and 7700X had nearly identical average temperatures. Since both processors averaged about the same power draw, this makes sense. However, we do see a much higher maximum temperature from the 7700X, likely due to the higher maximum power draw we saw. Overall, Photoshop is a light workload, and cooling is rarely a concern. Still, it is interesting to see how in these more lightly threaded applications, there is not a large difference gen-on-gen. The 9950X shows a similarly small gen-on-gen difference from the 7950X, with both having very similar average and maximum temperatures. In fact, the 9950X has a slightly higher average temperature.

For Cinebench, the results were a bit more dramatic. One thing to note when examining the Cinebench results is that we tested an extended duration of runs, meaning that there was downtime during pre-rendering sections that pulled down the average temperature. Nonetheless, Cinebench has high-intensity sections that stress CPUs, and we found the 97000X to average a massive 25 °C less than the 7700X, with a similar decrease in maximum temperature. Almost certainly, this is due to the much lower average power draw / TDP. Meanwhile, with a much smaller difference in power draw, the 9950X is only a few degrees Celsius cooler than the 7950X in both average and max temperatures.

Unreal Engine, also being a heavy all-core workload, is basically identical to Cinebench, save that the average temperatures are slightly elevated, given the reasons stated above. Regardless, the temperature decrease on the 9700X, in particular, is impressive.

Conclusion

There are a lot of factors to investigate when trying to pin down the efficiency and general power profiles of CPUs, from the cooler used to boosting algorithms to configured power limits. Although we only looked at a small number of benchmarks and CPUs, we still found some very interesting results around the new Ryzen 9000 series of processors.

Although AMD has reduced the TDP for some of their processors, it is interesting how, at a given TDP, the actual maximum sustained power draw has changed gen-on-gen by about 20 W. At a baseline, this means that AMD can increase multi-core performance by 5% with about 10% less power. While not revolutionary, this is certainly a solid improvement. However, due to how AMD’s boosting algorithms work in lightly threaded applications, the new chips are hardly any more efficient and don’t draw less power for those workloads. This is disappointing, as most computer use falls into these sorts of workflows.

The large apparent efficiency gain of the 9700X also has to be tempered by the overall TDP reduction of the part from 105 W to 65 W. In the past, we have seen AMD lose relatively little performance from the 7700X to 7700 or when enabling the 65 W “ECO” mode. As you increase power, the relative efficiency of a CPU tends to decrease, so it is possible that most of what we are seeing is just a result of pulling back on the efficiency curve. Due to how AMD has tweaked TDPs this generation, we expect to see similar results for the 9900X and 9600X, which also dropped 40-50 W.

Overall, the new AMD Ryzen 9000 series of processors is a mixed bag. Although they are a definite generational improvement between lower power draw and increased performance, they pale in comparison to some of the previous microarchitecture updates AMD has managed with Ryzen. Still, computer components have tended to only increase in power draw and heat over time, so we appreciate AMD’s efforts to focus on efficiency in its base configuration rather than push energy consumption to capture every last dreg of performance.

---

Finding the perfect workstation doesn’t have to be complicated. Explore our solutions page for a curated selection of recommended systems for a multitude of applications and workflows, or visit our custom configuration page if you already know the ideal hardware for your needs. If you need assistance with tailoring a system to a unique workflow or have any other questions, we encourage you to reach out to our dedicated technology consultants.