4k gaming met 9800X3D bij Hardware Unboxed
Additionale tekst van GN onder de video:
I'd also like to thank Gamers Nexus Steve who wrote this for me, but we didn't end up adding it to the video. So please watch their review to support them if you haven't already:
YouTube: RIP Intel: AMD Ryzen 7 9800X3D CPU Review & Benchmarks vs. 7800X3D, ...
From Gamers Nexus:
We currently run tests at 1080p and 1440p for CPU gaming benchmarks, though we mostly rely on 1080p results for comparison. Although we didn't bother for the 9800X3D review, we typically publish 1-3 1440p charts in games that are still mostly CPU-bound for perspective.
There are a lot of ways to approach reviews. We view bottleneck testing as a separate content piece or follow-up, as it also starts getting into territory of functionally producing a GPU benchmark.
What matters is a consistent philosophy: Our primary philosophy is to isolate components as much as possible, then as standalone or separate feature pieces, we run 'combined' tests that mix variables in ways we wouldn't for a standardized reviews. For us, reviews are standardized, meaning all parts (more or less) follow the same test practices. Introducing more judgment calls introduces more room for inconsistency in human decision making, so we try to avoid these wherever possible to keep comparisons fair. Choosing those practices is based upon ensuring we can show the biggest differences in components with reasonably likely workloads.
A few things to remember with benchmarks that are borderline GPU-bound:
- You can no longer fully isolate how much of the performance behavior is due to the CPU, which can obscure or completely hide issues. These issues include: poor frametime pacing, inconsistent frametime delivery, in-game simulation time error due to a low-end CPU dropping animation consistency despite good frame pacing, and overall quality of the experience. This is not only because it becomes more difficult to isolate if issues such as micro stutters are caused by the CPU or GPU, but also because the limitation may completely sidestep major issues with a CPU. One example would be Total War: Warhammer 3, which has a known and specific issue with scheduling on high thread count Intel CPUs in particular. This issue can be hidden or minimized by a heavy GPU bind, and so 4K / Ultra testing would potentially mean we miss a major problem that would directly impact user experience.
- Drawing upon this: We don't test for the experience in only that game, but we use it as a representative of potentially dozens of games that could have that behavior. In the same example, we want that indicator of performance for these reasons: (1) If a user actually does just play in a CPU bind for that game, they need to know that even a high-end parts can perform poorly if CPU-bound; (2) if, in the future, a new GPU launches that shifts the bind back to the CPU, which is likely, we need to be aware of that in the original review so that consumers can plan for their build 2-3 years in the future and not feel burned by a purchase; (3) if the game may represent behavior in other games, it is important to surface a behavior to begin the conversation and search for more or deeper problems. It's not possible to test every single game -- although HUB certainly tries -- and so using fully CPU-bound results as an analog to a wider gaming subset means we know what to investigate, whereas a GPU bind may totally hide that (or may surface GPU issues, which are erroneously attributed to the CPU).
One thing to also remember with modern 1080p testing is that it also represents some situations for DLSS, FSR, or XeSS usage at "4K" (upscaled).
A great example of all of this is to look at common parts from 4-5 years ago, then see how they have diverged with time. If we had been GPU-bound, we'd have never known what that divergence might be.
Finally: One of the major challenges with GPU-bound benchmarks in a CPU review is that the more variable ceiling caused by intermittent GPU 'overload' means CPU results will rarely stack-up in the hierarchy most people expect. This requires additional explanation to ensure responsible use of the data, as it wouldn't be odd to have a "better" CPU (by hierarchy) below a "worse" CPU if both are externally bound.
We still think that high resolution testing is useful for separate deep dives or in GPU bottleneck or GPU review content.