NVIDIA yesterday rolled out its GeForce Game Ready Driver 451.48 WHQL which supports a slew of new API features such as DirectX12 Ultimate, Vulkan 1.1, Hardware Scheduling and CUDA 11.0. You can read more about them in detail here.
In this post, we have a look at the impact of Hardware-Accelerated GPU Scheduling on gaming performance. On paper, this lets the GPU handle the memory allocation which in turn should result in better utilization and therefore, better lows as well as averages.
We tested this feature using a Zotac GeForce RTX 2080 Ti AMP and an AMD Ryzen 5 3600X paired with 16GB of dual-channel 3600MHz CL16 RAM. In our tests, although the results did show a marginal increase, the reasons for it are a bit surprising. Let’s begin:
1080p Gaming Benchmarks w/ Hardware Accelerated GPU Scheduling
As you can see, out of all four titles tested, only Metro Exodus doesn’t get an uplift. Instead the performance seems to deteriorate by a just 1 FPS. Since this is within the margin of error, we’ll chalk it as a tie. The other three games: Borderlands 3, Assassins’ Creed Odyssey and Origins, all see an appreciable gain in performance. The averages as well as the lows see an uplift of around 5 FPS across all three titles.
After many readers asked for 4K tests, here they are across half a dozen titles:
As you can see, it’s the same games that benefit the most from GPU scheduling. Older titles like Wildlands and Far Cry 5 see no benefit. Borderlands 3 and the Assassins’ Creed titles see the most prominent uplift.
How GPU Scheduling Improves Performance
This bit was somewhat confusing. We checked the GPU core clocks, temps, utilization, and even the package power. Nothing changed on switching between the two modes. The CPU wasn’t affected either. So what gives? What actually changed? Turns out that enabling Hardware-Accelerated GPU Scheduling improves the
system memory as well as the GPU memory utilization and allocation. Turning it on reduced not only the virtual memory and main memory utilization, but also the VRAM consumption (update). We reckon this is where the extra performance is coming from. Have a look at the graphs below:
Update: After intensive testing, it was found that in general, mainly the VRAM consumption is improved upon turning GPU scheduling on. While the virtual and main memory consumption is reduced is some titles, it’s marginal and not prevalent across config and game tested (example the GTX 1650 S test).
- VM Commit: Virtual memory (size of page file)
- Main Mem Used: Physical Memory (or RAM) used
- ON: GPU Scheduling on; OFF: GPU Scheduling off
As you can see, the amount of page memory (virtual memory) and main memory used decreases with GPU scheduling turned on. We also found that greater the delta between the two metrics, the greater performance you’ll get by turning the feature on.
Update: Some games like Borderlands 3 which use excessive amounts of VRAM and random stutters sees a notably smoother gameplay, with a reduction in VRAM.
Most other games like Assassins’ Creed Odyssey also see a drop in VRAM consumption upon turning on GPU scheduling. In general, we saw that greater the performance gain, the greater will be the VRAM drop upon turning GPU scheduling on. Furthermore, running applications in the background while gaming further amplifies this effect.
4K Results: System and Virtual Memory
4K Results: VRAM Consumption
Strangely, even games like Far Cry 5 which see no apparent performance boost upon turning GPU scheduling on see a steep decrease in VRAM and main memory consumption. My guess is that these titles are shader bound, meaning the performance is being limited by the shader processing power or the CPU, while titles like Assassins’ Creed and Borderlands 3, on the other hand, are memory bound games. As a result, they see a slightly improvement with GPU scheduling.