AMD surprised everyone with the announcement of the Ryzen "XT" processors, internally codenamed "Matisse 2." There's three new processor models at this time: the Ryzen 5 3600XT, Ryzen 7 3800XT, and Ryzen 9 3900XT, the processor in this review. We tested the other two models too, and their reviews are live now. According to AMD, the fundamentals of Matisse 2 are identical to first-generation Matisse—both processors use the Zen 2 architecture, there's no architectural difference, all changes are at the silicon production level and in the microcode. The result of these improvements are higher clock frequencies, higher boost clocks, and better sustainability of frequencies across the board—mostly with single-threaded and low-threaded workloads. The Ryzen XT models are drop-in compatible with previous Zen 2 processors, no BIOS update is needed.

Our performance results confirm that single-threaded apps see a significant benefit. For example, Cinebench 1T is up 2.6%, SuperPi gains 2.6%, and MP3 encoding runs 2.8% faster. However, the world is not single-threaded anymore, thanks to AMD. In our lightly threaded workloads, the differences are smaller (e.g., Photoshop +1.4%, software compilation +0.9%, and WebXPRT 1.7%). Multi-threaded benchmarks see almost no difference, which is as expected, AMD clarified that in their briefings. On average, in applications, there's a 1% performance improvement going from Ryzen 9 3900X to Ryzen 9 3900XT, which is not enough to beat the Core i9-10900K, but still plenty of performance for the most demanding applications. Especially multi-threaded programs will run incredibly well because of the 12-core/24-thread configuration—an enthusiast's wet dream just a few years ago.

These single-thread and low-threaded improvements sound like they're perfect for gaming. Unfortunately, our results tell a different story. There are negligible differences, even at lower resolutions, which I find surprising. In some games, there is a clear improvement; for example, Assassin's Creed Odyssey sees a few percent gained, and Tomb Raider runs slightly faster, just like Wolfenstein. In other games, we're seeing some negative scaling, but the differences are so small, it's hard to know whether they are random run-to-run variations or actually meaningful. I retested all my Zen 2 CPUs back to back in the last days, at the same time as the XT models, with the only difference being that I removed a CPU from the socket to install the other. AMD has been showing bigger gains in eSports titles in their materials, but these already run at very high FPS, so the gains are not that important. Also, with the Core i9-10900K out there, which is still a few percent faster, I doubt these results will change anything for professionals who want to buy the absolute fastest gaming CPU. The differences are small and barely worth mentioning, especially when running at higher resolutions, which have a stronger GPU bottleneck.

Our power and thermal results are much more favorable and show some clear improvements with the new Ryzen 3900XT. Temperatures are a bit more manageable, which is important for a high-end CPU like the 3900XT, and multi-threaded power efficiency is better than anything we've seen to date, almost 50% more energy efficient than Intel's Core i9-10900K! Overclocking potential is improved, too. Back in our Ryzen 9 3900X review, we were able to run the CPU at 4.0 GHz all-core, the Ryzen 9 3900XT reached 4.3 GHz. 4.3 GHz is still not close enough to the maximum single-threaded boost of 4.7 GHz, which means that overclocking will lose you some performance because the overclocked CPU will run 4.3 GHz all the time no matter how many cores are active. The default configuration will be able to respond to loads more dynamically, resulting in higher overall performance. The only exception is when running tasks that will load all cores 100% all day long—here, overclocking can help a bit. We measured the CPU to run around 4.35 GHz by default with all cores loaded. While extremely demanding workloads will have the clocks drop further due to power limits, etc., don't expect huge gains from manual OC; power and heat will be much more difficult to manage in such a situation, too.

AMD is pricing the Ryzen 9 3900XT at $500, which matches the launch price of Ryzen 9 3900X. The problem is that the 3900X has gradually dropped in price and can now be had for $430, and it includes a heatsink out of the box, which the 3900XT lacks. While I agree that most people interested in the 3900XT probably wouldn't run it with the stock cooler—which is quite decent by the way—we've been criticizing Intel for not including a stock cooler with their high-end CPUs for years, too. Looking at the $70 gap between the 3900X and 3900XT I'm having a hard time finding a convincing argument to justify the extra cost. If you check out our price/performance chart, you'll be surprised to see that the AMD Ryzen 9 3900XT managed to reach worse price/performance than Intel's Core i9-10900K—Intel does have slightly higher overall platform cost, though. I would buy the 3900X in a heartbeat because of great pricing and excellent performance, especially if you want to do more than just game, but the 3900XT will be a tough sell at its current price point. AMD is bundling Assassin's Creed: Valhalla with all Ryzen 3000 processors, so it's not a unique selling point for the XT. If AMD can bring pricing of the 3900XT down, possibly to current 3900X levels and below (remember, they're saving $25 by not including a heatsink), the 3900 XT has the potential to become the enthusiast's new darling.

Let's just hope that the XT lineup isn't AMD's vehicle to jack up prices in a stealthy way. If supply of Ryzen "X" models suddenly dries up and only XT is available, people won't have much choice unless they switch to Intel. A lot of our readers are also waiting for the upcoming Zen 3 CPUs and feel they don't need a new CPU that late in the cycle.