Conclusion
The Ryzen 9 3900X remains the fastest processor in our bench even with SMT disabled, beating the Core i9-9900K across multi-threaded tests. It also beats an untouched Ryzen 7 3700X. It's settled then that 12 physical cores beat 8-core/16-thread. However, the lack of SMT shaves a significant amount of performance off of the Ryzen 9 3900X. This goes to show that AMD's implementation of SMT is a sincere engineering feat and has a tangible performance gain. "Zen" simply wouldn't be the same without SMT, and AMD has been kind/smart enough to provide the feature on all its processor models priced north of $140 rather than keeping it exclusive to those above $300.
That said, we notice another fascinating phenomenon. With SMT disabled, the IPC of the "Zen 2" core seems to go up by roughly 2 percent. This directly impacts performance of software not in need of too many cores, namely games. We theorize that this could be due to two distinct things happening with the processor. For one, with the SMT off its shoulders, power-management of the processor is left to spread its power budget and Precision Boost headroom across fewer logical processors, so each of the cores runs at higher boost frequencies, positively impacting per-core performance. Secondly, to a smaller extent, performance is also benefited by the on-core schedulers not having to juggle resources between two logical processors. We saw hints of this when testing the Core i7-9700K, an 8-core processor that lacks SMT, and comparing it to the 8-core/16-thread i9-9900K. Despite slightly lower clock speeds, the i7-9700K managed to match or slightly exceed the i9-9900K at gaming. There are 12 such "unburdened" cores on a Ryzen 9 3900X, and hence, we see a neat 2% performance boost across gaming tests.
At the academically significant 720p resolution, which provides valuable insights to the processor's IPC and CPU-level bottlenecks, we see the 3900X with SMT off gain 2.2 percent performance averaged across all games. This gain is more pronounced in games such as "Metro Exodus", which sees a staggering 9.6 percent increase in frame rates. "Wolfenstein II" almost sees a 5 percent gain, and "Rage 2" is not far behind with a gain of 4.5 percent. "Assassin's Creed: Odyssey" and "The Witcher 3" are the only titles in our bench that penalize the lack of SMT with a 0.5-1.7 percent performance loss. We will gladly accept the gains we're seeing, especially in games with over 4 percent gains. These are gains we'd expect from light overclocking.
At 1080p (Full HD), the performance gains we see at 720p suddenly disappear in even the games we mentioned above. This is probably because the resolution no longer presents a CPU-limited scenario, nor is it a strain on the mighty GeForce RTX 2080 Ti we're testing with. At 1440p (WQHD), the performance lead begins to come back, but only faintly—1.2 percent averaged across all tests and only under 3 percent on games that benefit from SMT disabling. At 4K Ultra HD resolution, our RTX 2080 Ti thoroughly has its hands full, giving us a GPU-limited scenario. The performance variance between SMT on and SMT off blurs to just 0.2 percent on average, with under 1 percent swings noticed among games influenced by SMT or the lack thereof. These differences can easily be discounted for random variation.
Despite the performance numbers at real-world resolutions, the 720p data enamors us. Disabling SMT definitely appears to improve the per-core performance for games, most of which still aren't very parallelized, and can comfortably spread across 12 physical cores that aren't saddled with SMT. Therefore, if you've built a balls-to-the-wall gaming rig with the 3900X and don't do too much multi-threaded productivity work, you'll gain some FPS by turning SMT off, and turning it back on when you need it, which AMD made slightly easier by putting an SMT toggle in the Ryzen Master software. Applying the SMT toggle still requires a reboot, but it's certainly more convenient than rebooting to BIOS while mashing your Delete key, disabling SMT in the setup program, and then going through a second reboot for the BIOS setting to take effect.
There's a second, major part of this review with the non-gaming applications. Here, the loss of SMT is way more pronounced, particularly with applications that scale well in a multi-threaded environment. WPrime posts a massive 31 percent performance loss, which is roughly 1/3rd the processor's performance. Can you imagine that an intangible feature such as SMT could make such a huge performance contribution? Cinebench R20 multi-threaded also sheds 21 percent (over 1/5th) of its performance.
There are similar performance losses in our rendering tests with SMT disabled. Performance drops roughly 31 percent with Blender and Keyshot and 27 percent with Corona. With SMT-off, the 3900X falls behind both the 3700X and i9-9900K at Tensorflow AI tests. 7-Zip decompression tests show a gargantuan 39 percent performance loss. Media-encoding sprung a surprise with only a 3 percent performance loss seen with H.265 encoding. The H.264 encoder posts a 20 percent loss, though.
The percentage gain effect we saw in gaming was observed very faintly in less-parallelized non-gaming tests, with tests such as SuperPi posting a negligible performance gain and Unreal Engine light-baking giving us a 5 percent performance gain with SMT off. All our web-browsing tests post positive results for the SMT-off setup, with Google Octane running 7.8 percent faster, Mozilla Kraken 3 percent faster, and WebXprt a significant 6.3 percent faster.
With SMT off, the Ryzen 9 3900X idles with 2 W less power-draw (that's enough to power an SSD). It draws 26 W less in multi-threaded workloads (12 percent less, enough to power the LED/CFL lighting in your room). Under extreme stress, 3900X sans SMT runs with 10 W less power (enough to power all the case-fans in a typical build). With gaming, however, power draw is higher because the processor is running all cores in higher boost frequency states and games are able to saturate each logical processor that's a full CPU core. Another aspect is energy efficiency, and this is where SMT off loses big time in multi-threaded tests. The performance gain brought about by SMT outweighs the added power draw.
We leave this review pleasantly surprised. As PC enthusiasts ourselves who game a lot, and whose productivity workload isn't too multi-threaded, we can see the benefit of leaving SMT off on a Ryzen 9 3900X. Gaming performance gets an uplift that's as high as 9 percent if you're really lucky, and the multi-threaded performance of the resulting processor is still higher than 8-core/16-thread processors. 12 physical cores still do carry a lot of weight around. When we do get the itch to benchmark or deal with workloads that can scale across any number of cores, getting a massive 31 percent performance boost is just a flick of a switch away in Ryzen Master, pending reboot, of course.
)