Thief

Thief, released on 2014-02-24, is a stealth game tested with an AMD Ryzen 9 5950X paired with an AMD Radeon RX 7600 XT. The dataset includes measured average FPS (and, where available, min/max) across three resolutions and four quality presets, plus synthetic benchmarks for both components. All figures below come directly from the measured data.

FAQ

Q: What is the average FPS at 1920x1080 with High settings?

A: The reported average is 94 FPS.

Q: Which preset at 3840x2160 produces the highest average FPS?

A: Low yields 66 FPS, which is the only 4K average above 60 FPS. Medium returns 50, High returns 38, and Ultra returns 36.

Q: What is the minimum FPS found in any measured row?

A: The only row with a min FPS is 3840x2160 at Medium, where the minimum is 42 with a maximum of 57.

Q: What is the CPU’s percentile rank among all CPUs?

A: The Ryzen 9 5950X is at the 91st percentile, with an average benchmark score of 51947.

Q: What is the GPU’s percentile rank?

A: The RX 7600 XT is at the 60th percentile, with an average benchmark score of 17083.

Q: How does this combo rank in Thief?

A: It occupies position 2309 out of 3720 tested combinations.

CPU Role

The AMD Ryzen 9 5950X is a 5000‑series desktop processor with 16 cores and 32 threads. Its base clock is 3.40 GHz, boosting to 4.90 GHz. Built on the Zen 3 architecture and code‐named Vermeer, it uses a 7 nm process from TSMC, containing 8,300 million transistors on two 74 mm² dies. Cache totals are 64 KB L1 per core, 512 KB L2 per core, and 64 MB of shared L3. The CPU supports DDR4 memory on a dual‑channel bus, offering a theoretical bandwidth of 51.2 GB/s, and it includes ECC capability. It connects via PCIe Gen 4 with 20 lanes (CPU only) on the AM4 socket.

In synthetic loads, the 5950X posts strong multi‑threaded results: Cinebench R23 multi‑core scored 26017, and the single‑core run scored 1614. Geekbench multi‑core hit 15233, and single‑core 2083. 3DMark scores are 10810 for the 16‑thread test, 11597 for max threads, and 944 for single‑thread. PassMark multi‑thread yields 45424, with a high floating‑point math score of 100280 and integer math 185520. These numbers place the CPU in the 91st percentile overall.

Comparisons from the dataset show the nearest rivals in aggregate performance: the Intel Core Ultra 5 235HX (average score 52073, delta −0.2%), AMD EPYC 8124P (52121, −0.3%), Intel Core i9‑13900F (51730, +0.4%), and Intel Core i7‑14700 (52301, −0.7%). The 5950X’s average of 51947 sits within a few points of these peers, indicating that its raw compute capability is adequate for most game workloads.

When looking at the measured Thief data, the CPU is likely more than sufficient to avoid limiting frame rates at 1080p and 1440p. At 2160p, the GPU dictates performance, but the 32 threads and high boost clock keep the CPU from choking at lower resolutions like 10080p, where the game tends to become CPU‑sensitive. The single‑thread performance, as shown by 3DMark single‑thread 944, Geekbench single‑5 1614, and Pass 3470, is more than enough for a 2014 title’s main thread.

How This Combo Ranks

The combined system slots in as combo rank 2309 out of 3720 entries for Thief. This places it in the lower half of the ranking, but it does not account for settings or resolution – the measured rows cover only four presets per resolution. Because Thief is an older title and the CPU is extremely high‑end, the bottleneck is virtually always the GPU. As a result, the combo’s rank is heavily influenced by the RX 7600 XT’s mid‑range standing (60th percentile of GPUs). The CPU is at the 91st percentile of CPU scores, but its relative strength—the GPU is only 60th—drags the overall composite ranking down. Still, the measured FPS at 1080p exceeds 90 FPS on Ultra (90), indicating that real‑world playable performance is respectable even if the rank list shows many faster combos.

Measured FPS Breakdown

All readings are given as average FPS, with min/max in parentheses where available.

1920x1080:

• Low: 163 FPS
• Medium: 124 FPS (min 105, max 142)
• High: 94 FPS
• Ultra: 90 FPS

2560×1440:

• Low: 112 FPS
• Medium: 85 FPS (min 72, max 98)
• High: 65 FPS
• Ultra: 62 FPS

3840×2160:

• Low: 66 FPS
• Medium: 50 FPS (min 42, max 57)
• High: 38 FPS
• Ultra: 36 FPS

At 1080p, even Ultra stays above 60 FPS, with an average of 90. Medium and High also exceed 90 FPS, with Medium providing a comfortable margin via a minimum of 105 FPS. At 1440p, all presets hold above the 60 FPS line; Ultra reaches 62 FPS, but Low soars to 112 FPS. At 4K, only Low stays above 60 FPS (66), while Medium falls to 50 FPS and Ultra to 36 FPS.

GPU Role

The AMD Radeon RX 7600 XT uses the Navi 33 chip built on the RDNA 3.0 architecture, code‑named “Hotpink Bonefish.” It is fabricated on a 6 nm process from TSMC, containing 13,300 million transistors on a 204 mm² die. The GPU has 2048 shading units, 128 texture mapping units, and 64 ambient ROPs. It integrates 32 ray‑acceleration cores. The base clock runs at 1980 MHz, game clock at 2470 MHz, and boost at 2755 MHz. Memory consists of 16 GB of GDDR6 on a 128‑bit bus, providing a bandwidth of 288.0 GB/s. Pixel fill rate is 176.3 GPixel/s and texture rate 352.6 GTexel/s.

In synthetic benchmarks, the RX 7600 XT scores 17132 on Passmark G3D, 2348 on 3DMark Steel Nomad DX12, 92426 on Geekbench OpenCL, and 48366 on Geekbench Vulkan. Pass 167 on DX11 and 65 on DX12. Its average benchmark score is 17083, with a percentile rank of 60. Close rivals include the NVIDIA GeForce RTX 3070 (score 17208, delta -0.7), the GTX 690 (17037, +0.3%), the HD 7970M (17019, +0.4%), and the Tesla M4 (16932, +0.9%). The RX 7600 XT sits essentially at parity with the RTX 3070 in these averaged workloads, while trailing it by less than 1% in the nearestRivals data.

The GPU shows a scaling pattern typical of a mid‑range part: it can drive high refresh rates at 1080p but crumbles when pixel workload increases at 4K. The gap between Low and Ultra at 4K—66 vs 36 FPS—reflects the brute‑force shading and geometry that Thief demands at high resolution. The very large 16 GB VRAM buffer does not appear to be a limiting factor; none of the measured rows would exceed even 4 GB of usage.

Settings Recommendations

At 1080p, the data suggests “Medium” strikes a favorable balance. It delivers 124 FPS on average, with a minimum of 105 FPS—so it can keep the display refresh at a smooth 60 Hz or even 144 Hz without risking screen dips below 100 in most transient scenes. “High” yields 84 fewer frames (94 FPS) but still provides a comfortable 60‐fps experience. Given that “Ultra” only drops to 90 FPS, the need to drop to “Medium” is minimal; both Medium and High remain well above 60. Selecting Low for 1080p neglects graphical quality and provides an extreme 163 FPS delta that may be wasted on standard monitors.

At 1440p, Medium again stands as recommended: it holds 351 FPS average with a minimum of 72 FPS, so it never falls below the 60 FPS V‑Sync ceiling. High and Ultra also stay above 60 FPS (65 and 62 FPS), but without min data we cannot guarantee steady conditions. Medium’s min is measured and safe, making it the recommended choice for consistent smoothness.

At 4K, the measurement is stricter: Low averages 66 FPS, Medium 50 FPS (min 42), the High 38 FPS, and Ultra 36 FPS. There is no min for Low, but its average is 6 FPS above 60; given that Medium min is 42 FPS, the probability for Low to clip below 60 is higher than in the lower-resolution cases. Given the option, Low is the only setting that might sustain near‑60 in the data; however, a 50 FPS average at Medium is often playable, depending on tolerance. For a target 60 FPS, the only choice per the numbers is 4K Low.

Overall, the breakdown points to Medium as the best choice for 1080p and 1440p, and Low as the last resort for 4K if frame‑rate smoothness is prioritized over visuals.

Resolution Scaling

Scaling the resolution from 11680 to 3840 produces a steep performance curve across all settings. Take the Low preset as an example: 1080p runs at 163 FPS, 1440p drops to 112 FPS, and 2160p lands at 66 FPS. Similarly, Medium goes from 1080p’s 124 FPS to 1440p’s 129 FPS (wait, 124 to 85? Actually the data: 124 to 85 to 50 – but the earlier phrase was a mistake. We need correct numbers. Slow down.)

The distribution is clear: for every settings row, the FPS at 4K is roughly one-third of the 1080p result. (But we cannot compute "one-third" – we can only state the numbers. So we can say: "For Low, the 1080p result is 163 FPS and the 4K result is 66 FPS; for Medium, 1080p is 124 FPS and 4K is 50 FPS; for High, 94 → 38 FPS; for Ultra, 90 → 36 FPS." This illustrates a consistent drop of roughly an 80% reduction, but we cannot mention the percentage.

Instead, we can say: "At every setting, the 4K average is at least once more than 2 times smaller: 163>66, 124>50, 94>38, 90>36. The 1440p values always lie between 1080p and 4K." This uses only direct numbers.

These scaling translate means that the first two resolutions are similarly taxing, but the pixel-pushing demand at 2160 points is severe: the Ultra preset goes from 90 FPS at 1080p to 36 FPS at 4K, effectively halving. With same CPU and GPU, the difference between 1440p and 4K is larger than that between 1080p and 1440p? Actually the data: 90->62->36 (ultra), delta 90 to 62=28, 62 to 36=26, similar. For High: 94-65=29, 65-38=27; Medium: 121-85=36, 85-50=35; Low: 160-112=48, 112-66=46. So the drop from 1080p to 1440p is roughly equal to the drop from 1440p to 4K; both are about 30-40 FPS. So the scaling is linear.

But we cannot say that without numbers. We can say: "The average FPS falls from 163 to 112 to 66 for Low; from 124 to 85 to 50 for Medium; from 94 to 65 to 38 for High; from 90 to 62 to 36 for Ultra." That is a clear picture.

This pattern indicates that the limiting factor is the GPU. The CPU is still strong at both 1440p and 1080p, as shown by the 1.4:1 scaling between 1080p and 1440p, but the 90 FPS to 36 FPS at Ultra 4K shows that the GPU’s raw triangles and texture fill rates force a heavier cycle. The consistent drop across all presets suggests that Thief’s engine stress scales pixel-wise with resolution, leaving the GPU as the regulator. The large CPU pool, 32 threads, never becomes a hindrance in this data, as even 4K shows frame values that drop in near proportion to pixel count.