AMD’s new Ryzen 5 3600XT, Ryzen 7 3800XT, and Ryzen 9 3900XT are more of a victory lap for the company than a brand-new CPU stack. These three chips aren’t intended to replace hardware already on the market — they’re intended to cement and emphasize AMD’s overall desktop performance leadership.
Let’s meet the family:
The 3800XT offers the largest performance boost, at plus 200MHz boost, but the gains here are fairly small. AMD has also stripped the cooler out of the packaging for reasons unknown, and that hurts the overall value proposition. AMD’s Wraith coolers may not be top-notch, but they get the job done. We reused the Wraith Prism AMD shipped with the 3700X to test the 3800XT and 3900XT.
Here’s AMD’s official pricing stack:
Right now, the 3600XT is $225 at Amazon, the 3800X is $339 at Newegg, and the 3900X is $424. The new chips are (officially) 11 percent, 18 percent, and 18 percent more expensive than the X-class parts they supplant at the top of the market.
The proverbial fly in the ointment, in this case, is the 3700X. The Ryzen 5 3600X is a $249, six-core CPU, while the Ryzen 7 3700X is an eight-core chip for just $289. For gaming, the difference is likely irrelevant, but well-threaded workloads would prefer the 3700X over the 3600XT.
AMD Controls the Enthusiast Desktop Market
The XT series of processors are effectively an AMD victory lap, possibly with a bit of a middle finger towards Intel thrown in for good measure.
Intel’s six-core 10th Generation CPUs are well-positioned against AMD’s six-core Ryzen 5 CPUs around the $200 – $250 price points, but as we move towards $300, AMD starts to pull ahead in terms of available core count. The Ryzen 7 2700 is $267, while the Ryzen 7 3700X is $289. Intel’s eight-core chips, like the Core i7-10700K, start at $334. In previous years, that kind of pricing reflected the fact that AMD had a great price/performance ratio, but Intel had a slight overall performance advantage. Third-generation Ryzen obliterated most of those differences. 1080p gaming is still a win for Intel in certain titles, but the company’s advantage has shrunk.
The XT-class processors are AMD’s victory lap before the launch of the Zen 3 architecture later this year. It’s not clear what kind of performance gains AMD enthusiasts should expect from those chips, and support for them is officially limited to the X570 and B550 chipsets, with some beta-level support for older hardware expected to be made available under restrictions or by request.
Our AMD testbed used an Asus ROG Strix B550-E motherboard, pictured below:
One thing I quite liked: The board has an M.2 slot above the primary GPU slot, rather than below. This has two advantages: It increases the space between the CPU and the GPU, making it easier to stuff one’s hand into the gap to release a GPU’s safety latch. The other advantage is the welcome space for an M.2 heatsink.
One thing to be aware of, when choosing a motherboard, is the M.2 placement and heatsink clearance issue. Some motherboards that support M.2 NVMe connections limit which ports connect at maximum speed, and some ports are placed so close to the primary GPU, it’s impossible to use a heatsink on the drive without impinging on the GPU cooler. Putting the M.2 slot above the cooler solves this problem.
The only thing I didn’t like about the B550-E — and this is 100 percent a reviewer complaint that most people won’t care about — is the lack of an onboard button for power up and reset. It’s not hard to short a board with a screwdriver to power it up, but the button is always a nice touch.
Our AMD testbeds were configured with 32GB of RAM across four DIMMs of Crucial DDR4-3600 and a Corsair MP600 1TB drive. The latest version of Windows (2004) was used, along with Nvidia GeForce Game Ready Driver 451.48, paired with an RTX 2080 (non Super).
A Note on Gaming Performance: I have been unable to bring the Core i9-10900K sample Intel provided up to the performance figures I recorded for chips like the Core i9-9900K and Core i7-9700K last summer and fall. While I trust my results on those CPUs as accurate at the time they were gathered, I’m in the process of retesting the platforms. I have included Core i9-10900K results here, when they were absent from the review of that CPU, but dropped the Core i9-9900K and 9700K from this coverage.
In aggregate, our Core i9-10900K is a few percent slower than the 9900K or 9700K in most gaming tests, while leading those chips in non-gaming tests. According to Intel, this should not be so (which is why we’re retesting the other CPUs). Retesting the Core i9-10900K in a different motherboard, with different RAM, a different SSD, a different GPU, and a brand-new Windows 10 install failed to fix the problem.
Interestingly, the 3900XT isn’t much faster than the 3900X in Blender’s 1.0Beta2 benchmark, though the 3800XT picks up some performance over the 3800X. As expected, there’s no dramatic shakeout in where the Core i9-10900K falls. It’s faster than the Ryzen 7 3800XT and slower than the Ryzen 9 3900X / XT.
Cinebench R15 shows AMD cutting Intel’s lead in single-core rendering down to just 1.3 percent, also known as an effective tie. For fun, let’s contrast this with where things stood back when the Ryzen 7 1800X debuted:
At the $500 price point, over ~3.3 years, AMD has improved its single-threaded performance from 162 to 217, a gain of 1.34x, and from 1637 to 3168, an improvement of 1.94x. Intel has improved its single-threaded score from 201 to 220, a gain of 1.09x, and its multi-threaded score from 1797 to 2684, an improvement of 1.49x. Price point comparisons don’t line up as neatly with Intel, however — the Core i9-10900K is about half the price of the Core i7-6950X when the latter CPU was new.
Let’s move on to something a little newer:
Our Ryzen 3900XT manages to pick up a little more single-thread performance here, but it again ties the 3900X; the implication here may be that we’re hitting TDP limits. The 3900XT leads the 10900K in both single-core and multi-core tests, by 1.03x and 1.13x, respectively.
Not much new to see in Corona Render, with small uplifts for the 3800XT, flat performance for the 3900XT, and the Core i9-10900K outperformed by the 12-core Ryzen CPUs.
Dolphin EMU is a test of JIT compilation performance by the Dolphin emulator. Intel CPUs have historically done very well in this benchmark, which tends to prefer fast cache, and we see the Core i9-10900K outperforming the AMD options by a fairly wide degree.
Our Qt compile test uses Microsoft Visual Studio Community Edition 2019 and finally offers an example of the 3900XT outperforming the 3900X. The 10900K noses past the 3900X, but falls to the 3900XT.
In non-game tests, the 3800XT and 3900XT don’t exactly distinguish themselves. Game benchmarks are at least a little different, in a few places.
In Ashes, the new XT-class chips pull up the average frame rate a fair bit against the 3900X and 3700X, with the 3900XT pulling ahead of the 3950X.
The 3800XT threw an abnormally high score I’m assuming is a test error right now, but haven’t had time to go back and re-test just yet; 25.2 is a bit high. The higher 1080p framerates, however, show AMD again barely nudging past Intel, even under challenging GPU settings. DXMD takes a hefty hit with MSAA enabled, more akin to what you’d expect from an SSAA implementation.
In Hitman, again, the 3900XT meaningfully distinguishes itself from the 3900X and the 3800XT pulls ahead (less distinguishably) from the 3700X. The Core i9-10900K is still the fastest CPU in this game at 1080p.
Metro Last Light Redux is another game where Intel CPUs — the Core i7-8086K, in this case — manage a narrow win. Every other chip back from the 8086K clusters between 124 – 127 fps.
Rise of the Tomb Raider also shows strong clustering, which isn’t surprising given that we use SSAA for testing here. The legacy Core i7-8086K results are ahead, but the 10900K and Ryzen results all sit between 122 – 126fps.
These Chips Aren’t Meant to be Super-Appealing
In aggregate, the 3900XT and 3800XT are a slight improvement on their predecessors. They’re also significantly more expensive and they aren’t a particularly great value.
If you’re an AMD gamer, they offer a little more oomph in that side of the market, which, coincidentally, is pretty much the only place AMD is still giving up some wins to Intel. What they don’t do is change the status quo or offer the vast majority of gamers a particularly attractive deal.
Then again, there are people in gaming who don’t want great deals. This is an objective fact. Top-end chips from Intel and AMD typically demand steeper price premiums than they justify in terms of performance uplift. Intel has played this market for a number of years, to tremendous financial reward, and AMD is cutting itself a bit of the action as well.
Enthusiasts, I’d wager, will keep their focus on the already-existing X-class chips, not these new XT models, but these cores will sell into OEM systems where system builders may have been looking for an update. The big news for AMD will come when the Ryzen 4000 APU family is released, with rumors of an eight-core chip paired with a much faster onboard GPU, and later this year, when the Zen 3 core finally arrives.
It’s not clear what we should expect AMD to deliver with Zen 3, but a 1.1x – 1.15x IPC uplift target would match AMD’s overall trajectory, and pulling those gains off without reducing clock speeds would be an achievement of its own. Large core count increases are not expected this year, due to the inability of the Windows scheduler to support more than 64 cores in the first place, and AMD not exactly having lots of competition in the high core count space right now.
The 3600XT, 3800XT, and 3900XT aren’t great values for the money, and we recommend AMD enthusiasts investigate some of the company’s less-expensive processors on the whole — but if you’re a Team Red customer who wants the fastest gaming chip you can buy from AMD, a 3900XT is a good way to get it.