I use this tool to look up separate CPU and GPU benchmark indexes, then test the performance limit in the game. Choose your hardware and resolution above. Each number describes that part’s position in its own benchmark suite, whose games, settings, test hardware and reference score determine what the number means and which comparisons it supports.
Keep the columns separate.
The processor scores come from the Tom’s Hardware CPU benchmarks hierarchy, using the captured table updated August 26, 2026. The graphics-card scores come from the Tom’s Hardware GPU benchmarks hierarchy, using the captured table updated June 24, 2026. These are third-party measurements on the reviewer’s test systems.
Worked examples
These examples show the retained table values. They do not predict the frame rate or limiting component of the pairing.
Neither column reports FPS.
| Selected hardware | CPU suite index, 1080p | GPU suite index at selected resolution | GPU memory |
|---|---|---|---|
| Ryzen 7 7800X3D + GeForce RTX 5070 Ti, 1440p | 85.6 / 100 | 69.8 / 100 | 16 GB |
| Core i5-12600K + GeForce RTX 5080, 1080p | 60.8 / 100 | 81.9 / 100 | 16 GB |
| Core i5-12600K + GeForce RTX 5080, 4K | 60.8 / 100 | 69.8 / 100 | 16 GB |
| Ryzen 7 7700X + Radeon RX 9070, 1080p | 70.6 / 100 | 69.1 / 100 | 16 GB |
The columns use different workloads. A CPU score of 60.8 and a GPU score of 81.9 do not establish a 26% performance loss.
Changing the selected GPU leaves the CPU index unchanged; changing the CPU leaves the GPU index unchanged, because selecting another part does not create a new benchmark of that combination. Both numbers can help shortlist components within their own tables, but choosing an upgrade for a particular game also requires its actual frame-rate results, the tested settings and a target that makes sense for the display you use. I start there, before deciding which component to replace.
Test the limit in your game
- Record the starting conditions. Use a repeatable scene and note the resolution, graphics preset, upscaling, frame generation, FPS, frame times and GPU activity. Check the game’s frame cap and VSync setting before treating a flat frame rate as a hardware limit.
- Change one thing. Repeat the scene at a lower render resolution, keeping the other settings the same. A frame-rate increase shows that reducing rendering work helped in that test. Little change calls for more diagnosis; it does not identify the CPU by itself.
- Inspect the workload. Intel’s GPU/CPU-bound scenario guide uses a captured GPU work queue to distinguish a continuously busy GPU from gaps where it waits. Read that trace alongside CPU work, caps and frame times. One utilization percentage cannot explain every stall.
- Choose a change that addresses the measured limit. If reducing rendering work improves the result, test individual graphics settings before buying hardware. If the GPU is waiting, check the CPU workload, background activity, memory and the game’s own limits. Compare an upgrade using benchmarks for that game and target frame rate.
Increasing resolution adds rendering work; it does not recover CPU performance. I choose that trade for image quality only when the resulting frame times are acceptable.
Test before buying.
Compare parts within the same suite
For processors, use the CPU comparison tool or the 7800X3D vs 9800X3D upgrade guide. For graphics cards, use the GPU hierarchy and check the source’s game-specific FPS results. The PC Builder can help assemble a parts list, but a compatible list is not a benchmark of the finished PC.
Assumptions and sources
- CPU indexes come from the captured Tom’s Hardware CPU hierarchy updated August 26, 2026; GPU indexes come from the captured GPU hierarchy updated June 24, 2026. Source links are above. The dates describe the retained tables, not a continuously refreshed feed.
- CPU 100 means the leader of the captured CPU suite. GPU 100 means the leader of the captured GPU suite at that resolution. The tool applies no extra resolution multiplier and no cross-suite ratio.
- GPU indexes describe native-resolution raster performance. Ray tracing, upscaling and generated frames need their own relevant benchmarks.
- The September 26 correction removes the previous limiter labels and imbalance percentages because the separate indexes do not support those conclusions.