Analysis of DX12 thread usage in the Rise of the Tomb Raider DX thread discussion
Analysis of DX12 thread usage in the Rise of the Tomb Raider DX thread discussion
This is my first serious exploration into this subject. If you spot any mistakes or have questions about what I wrote, just let me know. This piece was inspired by a gaming session where I played Rise of the Tomb Raider and noticed it heavily utilized all six cores and many threads on my Ryzen 5 1600. I closely checked the CPU usage chart for each core and found that they seemed to follow similar patterns, which made me wonder if the game was truly using every core effectively or just repeating tasks across multiple ones. My performance dropped in DX11 overall, so I decided to dig deeper by disabling threads and cores to see when frame rates started to suffer due to fewer threads.
I used Rise of the Tomb Raider’s built-in benchmark with specific settings: medium preset, high textures, anisotropic filtering at 16x, no DOF, motion blur/vignette off, tessellation on, and SMA as anti-aliasing. I also enabled DX12 and ran it in exclusive fullscreen at 3440x1440. To test core and thread differences, I would disable cores or threads before booting into Windows 10, then launch ROTTR. I monitored CPU and GPU usage with MSI Afterburner, HWINFO64, and Task Manager, keeping an eye on temps and power draw.
For the benchmark, I ran it six times at medium settings, recording average, minimum, and maximum frame rates. At the end, I saved screenshots of all monitoring tools for analysis. The system in question runs an overclocked AMD Ryzen 5 1600 at 3.8GHz with a high-core count setup. It has a strong GPU (EVGA GTX 1070 Ti SC) and custom liquid cooling.
Starting from six cores and twelve threads, the average frame rate was 78.45fps, dropping to 44.71fps in the lowest scenario. When I simulated with four cores and eight threads, the average improved slightly to 79.2fps, but the minimums were much lower (37.72fps). The core count difference was clear: more cores generally led to better performance. Running a six-core 12-thread setup gave 78.45fps on average, while a four-core 8-thread version lagged by about two fps. The minimums were consistently lower in the higher core count, though not dramatically so.
When I tested a three-core six-thread configuration, the average dropped to 77.39fps with a minimum of 39.95fps—still worse than the six-core version but less so than the six-core eight-thread setup. The CPU usage was consistently higher, though overall performance remained decent. A two-core four-thread setup gave even lower averages (40.17fps) and fewer frames, making it the worst among all tested.
Looking at GPU usage, I noticed more noticeable dips in the four-core 8-thread version, with longer pauses and higher CPU load. In contrast, the six-core configurations maintained smoother gameplay. The minimum frame rates were similar across most setups, though the four-core versions struggled more under sustained load.
Overall, the game seemed to favor using more cores over threads, though a higher core count didn’t always translate to better performance—especially when thread usage was limited. For gaming, having at least four cores is usually sufficient, but adding more can help in demanding titles. The data suggests that Rise of the Tomb Raider benefits from a balanced approach between cores and threads.
For further details, you can view the spreadsheet I used: [link](https://docs.google.com/spreadsheets/d/1...sp=sharing)
And here are the graphs from my testing: [graphs link]
The main point is that your 1070 acts as a bottleneck; I bypassed the text and focused on the images.
The game really excels at handling several tasks at once, which surprised me.
Just disable the cores and check if their usage increases. It would take a lot of effort and time, but I understand why you'd prefer it.
My 1070 Ti feels a bit limited at that resolution, but I think the data remains useful because there are noticeable differences between each set. Even though 1080p shows more contrast between the 4-core and 6-core/12-thread configurations, I believe both would still work well in practice. I might try another test at 1080p to be sure.
I aimed to replicate lower-end processor designs closely, keeping in mind that disabling cores via Windows doesn’t let you pick between cores and threads.
This result, with minimal impact on performance even after shutting down most of the CPU, suggests a significant GPU limitation.
Mainly due to the strong ability to distribute workload across multiple threads