The AMD Zen 4 architecture is finally here, and we're allowed to tell you all about it! The AMD Ryzen 9 7900X is an interesting proposition as it's positioned exactly halfway between the $400 7700X 8-core, and the $700 7950X 16-core. It has double the amount of compute muscle as the 7600X 6-core, at a little less than double its price. At $550, it brings 12 performance-cores to the fight against both the Core i7-12700K and i9-12900K.

We spent the last months working on a new and improved CPU benchmarking suite with tons of new apps and games, and every processor was retested with the latest Windows 11 updates, game patches and drivers. Averaged over our 45 application tests we find the Ryzen 9 7900X end up about 6% faster than the Core i9-12900K averaged across our applications, and an impressive 15% faster than the i7-12700K. The average only tells a part of the story, and you need to pay closer attention to tests that scale across scores, such as 3D rendering, video encoding, code compilation, and enterprise applications, where the 7900X absolutely dominates the top-two Intel chips, mainly because all 12 of its CPU cores are "performance" cores, with ample power limits at their disposal for boost clock residency. AMD has given the 7900X some generous boost frequency spread, going all the way up to 5.60 GHz, but more on this later.

The Ryzen 9 7900X is only a marginally better processor than the 8-core 7700X, if your PC mainly serves as a gaming battlestation. You're not getting a much faster gaming processor. This is mainly because games still don't need more than a handful cores (6 to 8), and the single-chiplet design of the 7700X ensures that all gaming workload is localized to a single die, and handled among a single bunch of CPU cores, which is not the case with the dual-chiplet processors such as the 7900X. This is not to say that the 7900X is bad at gaming at all. Averaged across our game tests, the 7900X is faster than the Ryzen 7 5800X3D by 5%, and about 1.5% faster than the 7700X itself, on account of its higher boost clocks. The 7950X is only 1% faster. The Intel chips still end up slightly better, with the i7-12700K ending up 3% faster than the i9-12900K about 5% up. In early summer, way before we had any Zen 4 samples, I picked titles based on their relevance and popularity and it seems the game list tends to run a bit better on Intel than what a general average over many more titles would end up at. I have plans to test Zen 4 on a wider range of games, 50 is the target, to get a better feel for this. Both AMD Zen 4 and Intel Alder Lake are fantastic processors for gaming and in a subjective test without FPS counter you wouldn't be able to tell their performance apart. For gaming at highest FPS, the processor is secondary, you should rather spend as much money as possible on the graphics card and pair it with a more affordable processor, instead of focusing on brute CPU performance.

AMD is fabricating the Zen 4 compute dies on TSMC's 5 nanometer production process and the IO die is made on TSMC 6 nanometer. Just these numbers alone would suggest that Zen 4 is extremely energy efficient, but it seems clocking Ryzen 7000 at these high frequencies and TDPs has cost AMD quite some efficiency. We've upgraded our power measurement pipeline, and can now measure "chip-only" power consumption (as opposed to "whole system" before), and we record samples much faster, with full integration of the results. This lets us pair up power measurements with individual benchmark runs and perform more complex analyses, check out page 23. The 7900X is a fairly energy-efficient processor, with a 185 W chip-only power-draw in multi-threaded tests, which is 50 W less than that of the 7950X, and a whopping 72 W less power than the i9-12900K. Gaming power-draw is well contained compared to the top Alder Lake chip, too, at 108 W vs. 133 W, although it's much higher than that of the 7700X, which games at just 80 W, and the i7-12700K does so at just 93 W. Absolute power draw figures only tell half the story, the other half is efficiency. The 7900X ends up with comparable Cinebench points per-Watt, and gaming FPS per-Watt to the 7950X. The dual-CCD processors simply end up in a whole different efficiency band to the single-CCD 7700X and 7600X..

What will be a big source of discussion for Zen 4 is the extremely high temperatures. Using our high-end Noctua air cooler, there was a constant struggle trying to keep the CPU from its 95°C thermal limit at which point it will start lowering boost clocks. Even adding all-in-one watercooling wasn't able to make a significant difference. The only AIO that showed meaningful gains was the Arctic Liquid Freezer II, which offers an "offset mounting" configuration that moves the center of the cold plate away from the center of the CPU to sit right on top of the compute dies on Zen 4. This gained us a couple of degrees, so we could stay below the 95°C temperature limit. The differences are minimal, as we've tested in our 7950X review. AMD is very clear in their cooler guidance for 7950X and 7900X, mentioning "240-280 mm liquid", but the 7700X and 7600X are supposed to be paired with "mid-frame tower coolers", which I feel is a bit less than what I would recommend. AMD was also clear that 95°C is the new 65°C: "TJMax is the max safe operating temperature – not the absolute max temperature. In the Ryzen 7000 Series, the processor is designed to run at TJMax 24/7 without risk of damage or deterioration. At 95 degrees it is not running hot, rather it will intentionally go to this temperature as much as possible under load because the power management system knows that this is the ideal way to squeeze the most performance out of the chip without damaging it."

I'm just surprised that temperatures are so high. Maybe it's the smaller die size of the compute dies, paired with high clocks and voltage that leads to these temperatures, or there some compromises were made to achieve compatibility with AM4 coolers, or maybe both. I'm sure in the coming months we'll learn more about the cooling challenges, how to overcome them, and cooler vendors will probably release solutions optimized for Socket AM5. Right now you need to be aware that cooling Zen 4 is much more difficult than Zen 3, and that you should really stop worrying about temperatures. AMD is very clear in their messaging that these temps are normal and expected.

During testing I didn't encounter any major bugs or issues; the whole AM5 / X670 platform works very well considering how many new features it brings; there's one big gotcha though and that's startup duration. When powering on for the first time after a processor install, your system will spend at least a minute with memory training at POST code 15 before the BIOS screen appears. When I first booted up my Zen 4 sample I assumed it was hung and kept resetting/clearing CMOS. After first boot, the super long startup times improve, but even with everything setup you'll stare at blank screen for 30 seconds. To clarify: after a clean system shutdown, without loss of power, when you press the power button, you're still looking at a black screen for 30 seconds, before the BIOS logo appears. I find that an incredibly long time, especially when you're not watching the POST code display that tells you something is happening. AMD and the motherboard manufacturers say they are working on improving this—they must. I'm having doubts that your parents would accept such an experience as an "upgrade", considering their previous computer showed something on-screen within seconds after pressing the power button.
Update Sep 29: I just tested boot times using the newest ASUS 0703 Beta BIOS, which comes with AGESA ComboAM5PI 1.0.0.3 Patch A. No noticeable improvement in memory training times. It takes 38 seconds from pressing the power button (after a clean Windows shutdown), until there ASUS BIOS POST screen shots. After that, the usual BIOS POST stuff happens and Windows still start, which takes another 20 seconds or so.

Just like previous AMD Ryzen processors, all AMD Zen 4 CPUs come with unlocked multipliers, which makes overclocking much easier. Our highest manual all-core OC turned out to be 5.2 GHz, held back by the fact that it's impossible to keep the fully loaded CPU under 100°C, even with a large AIO. This means that you're limited in how much voltage you can feed the beast, and you thus end up with much lower clocks. You could decide to not run Prime95 and test stability with lighter loads only, but that's not what I consider a stable OC. I ran all our benchmarks at 5.2 GHz All-Core OC, and this config ends up roughly matching the stock settings on average, in some rendering tests it does have a tiny advantage, but the differences are small enough to not make all-core OC worth it. The better way to OC Ryzen is using PBO "Auto Overclocking", paired with Curve Optimizer, same as Zen 3. Spending just a few minutes on that will yield you a few percent in OC performance—not a lot, but the times of huge OC gains are a thing of the past—CPU manufacturers are getting better and better at giving you the best performance out of the box. When left alone at stock settings, the 7900X spreads its boost frequency very nicely across cores, with its frequency at 1T load going beyond the 5.6 GHz on the box, and an all-core boost as high as 5.3 GHz under FP stress, 5.275 GHz under SSE-only stress; and 5.2 GHz under AVX stress.

What is surprising on my Ryzen 9 7950X is that at stock, the clock speeds on all cores in the second CCD are considerably lower than in the first CCD. I'm seeing the same behavior with the 7950X, which is dual-CCD, too. The clock difference is clearly noticeable across all thread counts, from single-threaded to 32 threads active, and not related to CPPC2 "best cores". The delta is between 100 and 300 MHz, which is much more than I would have expected from random silicon variations. If the Windows scheduler puts workloads on the second CCD, then these will run a few percent slower than had they been scheduled on the first CCD. I asked AMD about this behavior and whether it's expected, they haven't responded yet, I'll update this review when they do.

As mentioned before, Zen 4 now comes with integrated graphics. These "just work"—if no discrete graphics card is installed, plug the monitor cable in the motherboard, boom, everything works. Windows Update will install the right driver, or you can grab the official AMD Radeon drivers. Overall IGP performance is outstanding, and plenty for everything except serious gaming. Some lighter 3D apps work perfectly fine, too, and get hardware-acceleration, just like all video decode and encode workloads, for video conferencing, for example. While AMD is very clear that the integrated graphics are not for gaming, I still ran our IGP test suite and have to say I'm impressed—the Zen 4 IGP roughly matches the IGP of the Core i9-12900K, which uses Intel's latest Xe architecture and that they've spent a ton of die area on. A real graphics card is still much faster, even the most entry-level Radeon RX 6400 offers 4 times (!) the FPS. For all other typical consumer activities this integrated graphics solution is awesome and they'll be a huge selling factor for cost-optimized compact office systems, a market where Intel has dominated, because an additional graphics card wasn't required.

AMD is pricing the Ryzen 9 7900X at $550, which is also the price at which the 5900X was launched. At this price, it is seen undercutting the $585 that the i9-12900K commands. With application workloads, there can be no doubt that you're getting more muscle than the top Alder Lake, for fewer coins, with the 7900X ending up with 12% more performance per Dollar. The gaming performance of the i9-12900K is 3-5% higher than the 7900X, but at 6.3% more cost. Actually, don't even look at gaming Performance per Dollar if you're building a PC that only plays games and isn't into serious content-creation work. The 7700X and 7600X dominate the bang-for-buck, while performing within 2% of this chip. Overall platform cost of Ryzen 7000 is also very high. The new motherboards are extremely expensive, and there's only X670/X670E available for now, the more affordable B650 boards will be released in October. While Intel supports both DDR4 and DDR5, AMD is focusing only on DDR5, saying that this will eventually help adoption rates of DDR5, which enables higher production volumes, which will bring memory pricing down. Nothing wrong with that, and the performance gains from DDR5 are substantial, it still drives up platform cost. Basically you're looking at $1000+ for Zen 4+board+memory whereas $500 will get you a solid Zen 3 setup that offers a bit slower performance, but at half the price. For gamers, this means you can go one or two tiers higher in your graphics card choice, which will definitely result in an FPS increase. I feel like a lot of creators who are willing to spend this amount of money for 7900X, will rather just opt for the 7950X, which is only $150 more expensive, like 10% of the total platform cost, so not that big a cost increase. Intel's new 13th gen processors are getting announced this week, too. It seems like the CPU market will become even more interesting in the coming months, and I'd expect that prices will go down, leading up to 2023.