Cyberpunk 2077 is said to suffer from incorrect SMT detection of AMD’s Ryzen processors. An unofficial patch has been available recently, correcting the problem a priori. Here’s to the theory, what about in practice?
Long awaited by players, Cyberpunk 2077 has suffered since its launch from a large number of problems (who said “buuuuuugs” at the back of the room?), Various and varied. We are interested in an unofficial patch available for a few days on Github and supposed improve performances, especially on the CPU side. The latter includes a number of changes made by the community to the game code, while facilitating their activation.
This patch for Cyberpunk 2077 theoretically allows unbridle the SMT of Ryzen processors (poorly detected by default), but also to remove the countermeasures (not really necessary in a single player video game, but which have a negative impact on performance) intended to mitigate the Specter flaw present in many processors, optimize the management of RAM and graphics or to deactivate TAA antialiasing and Async Compute.
Cyberpunk 2077: we found all the hidden options and how to activate them
Two test platforms, five years apart
To test the effectiveness of this patch, we have set up two very different configurations: an “old” Intel configuration from 2015/2016, fairly representative of a significant portion of the fleet still installed on the market, and a second AMD configuration ” contemporary ”, with a vintage“ 2020 ”processor and graphics card.
In detail, the first configuration is based on an Intel Z170 platform (Asus Z170 Deluxe), with a Core i5-6500 processor (a quad-core model) cooled by a Noctua NH-U12S cooler, 16 GB of Klevv DDR4-2133 (CL15) memory, a GeForce GTX 980 Ti (MSI GTX 980Ti Gaming 6G), a 460 GB Sandisk Ultra II SSD and a Corsair HX 750 power supply.
The second includes an X570 platform (Gigabyte X570 Aorus Master), a Ryzen 5 5600X processor cooled by an AiO watercooling system BeQuiet! Pure Loop 280, 16 GB memory G.Skill DDR4-3600 (CL16), a Gainward RTX 3080 Phoenix GS graphics card, a SSD WD Blue SN550 1 TB (M.2 NVMe) and a power supply Thermaltake Thoughpower iRGB Plus 1050W (80Plus Platinum certified). In both cases, we used a moniteur Philips Momentum 558M1RY for display.
Our test scene (scene 3 of debug mode) consists of three parts: the first with crowds and vehicles takes place in Little China, the second on the cleared roads of Northside is lighter, and we end with a heavier scene, still in Northside but with a higher load of buildings and vehicles. The game has been tested in version 1.04.
Specter Patch and Memory Pool
According to its author, this unofficial patch allows disable software countermeasures to mitigate the Specter flaw present in many recent CPUs. It also allows modify the management of RAM and VRAM. A game configuration file indeed suggests that the use of RAM and VRAM is artificially limited; this patch proposes to remedy this by increasing the amount of memory indicated in this configuration file up to 100% of the amount of memory on board the graphics card, and up to 50% of the available RAM (i.e. 8 Go in our case).
With our “old” configuration, the application of these two modifications in Full HD does not really have the desired effect: performance does not increase, worse they are even slightly lower than those offered by the game.
Same observation in Full HD with “Medium” settings on our Ryzen 5600X / RTX 3080 configuration: the performances with and without the “Specter” and “Memory Pool” modifications are almost identical, whether at the level of framerate medium or framerate minimum.
Switching to QHD with higher graphics settings (therefore with a greater load on the GPU side) even tends to show that these two patches are counterproductive: highest performance is achieved with unmodified clearance !
Patch Async Compute
Enabled by default in the game, theAsync Compute would have, according to the creator of the unofficial patch (yamashi), a negative impact on performance with “old” graphics cards (GTX 10 Series and previous models). Disabling Async Compute could therefore in some cases improve performance slightly. Let’s see what it is with our GeForce GTX 980 Ti.
The difference in performance in Full HD is very slight, but seems very real: with a crowd density set to “Low” or “Medium”, disabling Async Compute (with our GTX 980 Ti) slightly increases the framerate way. There is however no impact on minimum FPS. No need to test in a higher definition, our Core i5-6500 is already on its knees in Full HD.
Patch SMT (Ryzen)
The patch theoretically allows better use of the SMT of Ryzen processors, via a modification of a jump condition in the game code. We have not tested this patch with our Core i5 since it does not have a Hyper-Threading (and anyway, Intel-style SMT is correctly supported by Cyberpunk 2077).
In Full HD with medium settings, we note that the patch has an effect on performance, but only at the level of framerate minimum, which increases by a handful of percent. The framerate mean decreases slightly.
In QHD with “Ultra” graphics settings and textures set to “High”, same observation: average FPS decreases slightly with the active patch, but in return the framerate minimum is a little higher. The difference is not extraordinary, but still good to take. We therefore hope that CD Projekt will correct the poor support for the Ryzen SMT in a future official patch.
Conclusion: a little efficiency, a lot of placebo
If the patch does have an impact on performance when it comes to enabling SMT support on Ryzen processors, the other changes (disabling Specter countermeasures, memory pool changes, etc.) are similar on the other hand more to placebo than to a real effective patch, in any case on our two test configurations at least.
Whether with a small GPU and an older generation graphics card, or a much more recent CPU / GPU pair, the performance impact of the “Specter” and “Memory Pool” changes is non-existent, or even negative. We will therefore be content to apply the “SMT” patch to slightly increase the framerate minimum involved with the Ryzen-based configs (which is not bad, while waiting for an official patch from CD Projekt!), knowing that the framerate mean may drop slightly.
Be careful not to take at face value the few measures that you can find here and there concerning these unofficial modifications of the game: the application of a strict bench protocol, with a perfectly reproducible test scene, is essential to obtain impeccable performance measurements !