We found the Missing Performance: Zen 5 Tested with SMT Disabled 78

We found the Missing Performance: Zen 5 Tested with SMT Disabled

Test Setup »

Individual Results

On this page we'll highlight several cases, to explain why we're thinking that something is different with Zen 5 and its multithreaded scaling. Technically this is cherry-picking, but we want to show the interesting ones and put them in some context. We're not trying to hide anything, the full list of test results is included on the following pages, presented in the same way as in our regular processor reviews.

Our argument with this article is not that SMT is the way to go, but rather that there's something in the SMT or non-SMT behavior that affects performance on Zen 5 more than on previous processors.

Gaming at 4K Overall


Let's start with gaming performance at 4K. This resolution is highly GPU limited, so the differences due to the CPU are expected to be small, I still find it quite interesting. Here we see the Ryzen 7 9700X without SMT gain 1.5% over the processor with SMT running. While 1.5% doesn't sound like a lot, it's actually quite a significant gap when the ranking is considered. The 9700X moves from being "similar to Zen 4 and the midrange Intel CPUs" to "second-fastest right behind 7800X3D and matching 14900K."

We can also see that removing the power limit and pushing the clocks further with PBO really can't make much difference—an additional 0.2% is gained.

Baldur's Gate 3 / 4K


Here's a good example for the gains from disabling SMT. The 9700X gains 7.2%, whereas the 7800X3D loses 11% and the 7700X gains only 1.9%. This is on the same setup, with the same everything, just the CPU swapped out back-to-back. This confirms that on Zen 5 something is different in the multi-thread scaling than on Zen 4, and X3D.

Spider-Man / 4K


Here's another good example. While all CPUs are climbing without SMT, the delta with the 9700X is bigger as on the others, it runs 5.1% faster, the 7700X gets a 3.6% boost and there's no change with the 7800X3D.

Alan Wake 2 / 4K


Things are different in Alan Wake 2. Here every CPU is a little bit faster with SMT disabled, but the differences are minimal.

Cyberpunk 2077 / Full HD


Switching gears to 1080p Full HD and Cyberpunk 2077. Cyberpunk's engine has been developed from the ground up to make good use of a lot of threads, and it scales very well. When we disabled SMT, we reduced the number of cores that are usable by Windows from 16 to 8—per definition SMT is able to run two concurrent threads on each core. The result is that Cyberpunk runs at slightly lower FPS across the board. 7800X3D: -0.7%, 9700X: -3.8%, 7700X: -5.4%. Just generally disabling SMT everywhere will cost performance in some cases.

Baldur's Gate 3 / Full HD


This isn't an effect of the resolution though. If we return to Baldur's Gate 3, this time at 1080p, we still see similar scaling like at 4K, where the 9700X definitely improves a lot more than the other CPUs: by +6.8%, 7700X: +3.3%, 7800X3D: +2.1%.

Remnant II / Full HD


Remnant II on the other hand really just likes to run without SMT, no matter the architecture or cache, all CPUs gain roughly the same performance from disabling SMT: 7700X: +6.5%, 9700X: +6.7%, 7800X3D: +6.4%.

Gaming at Full HD Overall


The averages at 1080p are again, clearly in favor of the 9700X without SMT. It clocks a 2.6% performance increase on average, the 7800X3D is minimally slower without SMT, but still the fastest CPU in our test group. The 7700X without SMT gets a meager 0.7% performance uplift. With overclocking, the 9700X gets really close to the top and is able to pretty much match the 14900K.

Baldur's Gate / 720p


Still, there's definitely cases where having a lot of L3 cache can produce FPS numbers that no other CPU can achieve, no matter how many cores or their frequency. X3D just wins here without any doubt. The underlying reason is that certain parts of the game engine will fit into L3 cache on X3D, which means costly trips to system memory are eliminated and everything happens within the processor.

Emulators


Switching to our application workloads, emulation sees very good gains, again roughly +7% for 9700X without SMT, while the comparison CPUs don't gain nearly as much.

AI / Machine Learning


Wow! No SMT, so half the threads, and still the 8c/8t Ryzen 7 9700X still murders every other CPU in our test group. The other processors are showing small losses, which again is strong evidence that something is different with Zen 5 when it comes to SMT.

Browser Performance


Web browsing is generally single-threaded or low-threaded, and here, too, massive gains—to never-before-seen levels of performance. Look where the Intel 14900K is in that list.

Media Encoding


Last but not least, we have media encoding, H.264 in particular. This test is special because it's a case where the stock 7700X is faster than the 9700X—something completely unexpected. Turning off SMT lowers performance, as expected, because there are fewer threads that can run at the same time, but the interesting part is that now 9700X is faster than 7700X, which is the expected result. Going from 2.4% slower to 2.2% faster just by disabling SMT is definitely an interesting result.

Cinebench


Heavily threaded, especially synthetic apps will definitely see a major performance hit due to the lack of SMT—this is expected and the reason why SMT/HT was invented in the first place.


At the other end of the spectrum, a purely single-threaded workload like Cinebench 1T runs virtually the same, because only a single core is loaded and nothing is fighting for shared resources.
Next Page »Test Setup
View as single page
Aug 10th, 2024 14:55 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts