Thief

The Intel Core i7-12700KF is a 12-core, 20-thread desktop processor built on the Alder Lake architecture, with a base clock of 3.60 GHz and a boost clock of 5.00 GHz. Its benchmark profile shows a significant spread between single-thread and multi-thread workloads, which is typical for a hybrid design that balances high-frequency performance cores with efficiency cores. In the context of Thief, a 2014 stealth title, the game’s engine is primarily reliant on a few strong threads for game logic and draw calls, while physics and AI can leverage additional cores. The processor’s 3dmark_single_thread score of 1043 and passmark_single_thread score of 3984 indicate strong per-core performance, which is crucial for maintaining high frame rates in a game that may not scale perfectly across all 20 threads. The Cinebench R23 multicore score of 28979 and the 3dmark_max_threads score of 9983 demonstrate that the CPU has ample headroom for background tasks and any engine threads that do scale, ensuring that the CPU is unlikely to be the bottleneck in most scenarios. The processor’s 25 MB of shared L3 cache also helps keep frequently accessed game data close to the cores, reducing latency that could otherwise impact frame pacing.

The data shows that the CPU’s performance relative to its direct competitors is remarkably tight. The nearest rival, the Intel Core i5-14400, has an average benchmark score of 35336, which is only 0.1% lower than the i7-12700KF’s average of 35355. Similarly, the Intel Core i9-12900HX is 0.2% ahead, and the Intel Core i5-13600T is 0.3% behind. This indicates that for a game like Thief, which is not heavily multithreaded, the i7-12700KF’s advantage in raw multi-core throughput (e.g., a Cinebench R23 multicore score of 28979) may not translate into a tangible in-game performance lead over these rivals. The percentile rank of 89 for the CPU versus all tested CPUs confirms that it sits in the upper echelon of processors, but the game’s performance will be far more sensitive to the GPU and the specific settings used. Given the 3dmark_16_threads score of 9282 versus the 2_threads score of 2065, the scaling efficiency is not linear, suggesting that Thief’s workload will likely fall into the 2-8 thread range, where the CPU’s single-thread performance (3dmark_single_thread score of 1043) will be the deciding factor.

GPU Role — VRAM, clocks, and how they relate to this game's results

The NVIDIA GeForce RTX 4090 is an absolute powerhouse, built on the Ada Lovelace architecture with 24 GB of GDDR6X memory on a 384-bit bus, providing 1.01 TB/s of memory bandwidth. Its base clock is 2235 MHz with a boost clock of 2520 MHz, and it features 16,384 shading units, 512 texture mapping units, and 176 render output units. The GPU’s raw compute capabilities are staggering, with an FP32 throughput of 82.58 TFLOPS and a pixel rate of 443.5 GPixel/s. For Thief, which was released in 2014, this GPU is wildly over-provisioned in terms of both memory capacity and compute power. The 24 GB VRAM far exceeds anything the game’s textures and geometry could demand, meaning there is zero risk of memory capacity bottlenecks at any resolution. The 1.01 TB/s bandwidth ensures that even at 4K with Ultra settings, the GPU can feed its massive shader array without stalling.

The benchmark results confirm this dominance. At 3840x2160 with Ultra settings, the RTX 4090 achieves an average FPS of 82.5, which is a playable frame rate, but notably lower than the 111.1 FPS at Medium and 148 FPS at Low settings at the same resolution. This indicates that even this GPU is not completely immune to the demands of 4K Ultra, though it maintains a very high level of performance. The GPU’s percentile rank of 91 versus all GPUs places it in the top tier, and its nearest rivals are professional workstation cards like the NVIDIA Tesla T4 (which is 0.4% slower in average benchmark score) and the AMD Radeon Pro Vega 56 (0.9% slower). The RTX 4090’s average benchmark score of 66473 is only 0.4% higher than the Tesla T4’s 66733, but in gaming workloads, the 4090’s architecture and driver optimization for DirectX 12 Ultimate (12_2) give it a distinct advantage that the compute-focused Tesla cards lack. The data shows that the GPU is the primary driver of the measured FPS differences across resolutions, with the CPU providing sufficient headroom to avoid limiting the frame rate.

Settings Recommendations — which preset gives the best experience per the measured rows

The measured FPS data provides a clear picture of how the settings presets interact with resolution to deliver the best experience. At 1920x1080, the differences between presets are relatively modest, with Low averaging 196 FPS, Medium at 175.6 FPS, High at 158.5 FPS, and Ultra at 156.8 FPS. The jump from Low to Ultra costs roughly 20% of the frame rate, but the game remains playable at all settings. However, at this resolution, the CPU’s single-thread performance (3dmark_single_thread score of 1043) becomes more relevant, as higher preset levels may increase draw calls and AI computations that stress the CPU. The data suggests that High preset at 1080p is the sweet spot, offering 158.5 FPS while delivering significantly better visual quality than Low, without the marginal FPS drop seen when moving to Ultra (156.8 FPS).

At 2560x1440, the pattern shifts. Low averages 169.7 FPS, Medium 153.4 FPS, High 142.7 FPS, and Ultra 135.5 FPS. The gap between High and Ultra is 7.2 FPS, which is less than the gap between Medium and High (10.7 FPS). This suggests that the Ultra preset at 1440p is well-optimized, providing a visual upgrade for a relatively small performance cost. For gamers with high-refresh-rate 1440p monitors, the Medium preset at 153.4 FPS is a strong choice, as it stays above 150 FPS while offering better visuals than Low. At 3840x2160, the GPU is more heavily taxed. Low averages 148 FPS, Medium 111.1 FPS, High 81.4 FPS, and Ultra 82.5 FPS. The data shows an anomaly: High (81.4 FPS) is slightly slower than Ultra (82.5 FPS), which is within the margin of measurement error but indicates that the Ultra preset is not significantly more demanding than High at this resolution. The recommendation for 4K is to use the Medium preset at 111.1 FPS, as it provides the best balance of visual fidelity and smoothness, comfortably above the 60 FPS threshold. Using Low at 4K (148 FPS) is only advisable if the user prioritizes maximum frame rate over image quality, as the jump to Medium is substantial in both quality and performance.

How This Combo Ranks — use comboRankInGame vs other tested combos

The combination of the Intel Core i7-12700KF and the NVIDIA GeForce RTX 4090 achieves a rank of 65 out of 1642 tested combos for Thief. This places the combo in the top 4% of all systems tested, which is a strong result but not the absolute pinnacle. The data indicates that there are 64 other combos that deliver higher average FPS in this game, likely featuring newer or higher-clocked CPUs that can better feed the RTX 4090 at lower resolutions. The percentile rank of the CPU (89) and GPU (91) are both high, but the combo’s rank of 65 out of 1642 suggests that the CPU is the limiting factor in some scenarios, particularly at 1080p where the FPS is highest (196 FPS on Low). At 1080p, the CPU’s single-thread performance becomes the bottleneck, as the GPU has more than enough headroom to render frames faster than the CPU can issue draw calls.

The deltaPct values for the CPU’s nearest rivals are all within 0.3% of its average benchmark score, indicating that swapping the i7-12700KF for an i5-14400, i9-12900HX, or i5-13600T would yield almost identical in-game performance. This is because Thief’s engine is not highly parallel, and the 12-core/20-thread configuration of the i7-12700KF is not fully utilized. The combo rank of 65 confirms the RTX 4090’s ability to carry the system, as the CPU is only moderately above average in its percentile rank. For comparison, the GPU’s nearest rivals are all professional cards that are not optimized for gaming, so the RTX 4090’s gaming-specific features (like DirectX 12 Ultimate support) give it a clear edge in Thief. Overall, the data shows that this combo is exceptionally well-suited for 4K gaming, where the GPU’s raw power dominates, but the CPU’s modest single-thread performance relative to the very latest processors prevents it from achieving a top-10 ranking.

FAQ

Q: What is the highest average FPS this combo achieves in Thief, and at what settings?

A: The highest measured average FPS is 196, achieved at 1920x1080 with Low settings. At 3840x2160 with Low settings, the average FPS is 148, which is the highest at 4K resolution.

Q: How does the CPU’s multi-threaded performance compare to its single-threaded performance in the context of Thief?

A: The CPU’s Cinebench R23 multicore score is 28979, which is very high, but its single-core score is 4091. Thief is not heavily multithreaded, so the single-thread performance (3dmark_single_thread score of 1043) is more relevant for game logic, while the multicore headroom ensures background tasks do not interfere with frame pacing.

Q: Is there a noticeable difference between High and Ultra settings at 4K resolution?

A: The data shows that at 3840x2160, High averages 81.4 FPS and Ultra averages 82.5 FPS. The difference is only 1.1 FPS, which is within measurement noise, so Ultra offers a visual upgrade at virtually no performance cost for this combo.

Q: What is the best preset for a 1440p monitor with a 144 Hz refresh rate?

A: At 2560x1440, the Medium preset averages 153.4 FPS, which exceeds 144 FPS. The High preset averages 142.7 FPS, which is slightly below 144 FPS. Therefore, Medium is the recommended preset to fully utilize a 144 Hz display.

Q: How does this combo rank compared to other tested systems in Thief?

A: This combo is ranked 65 out of 1642 tested combos, placing it in the top 4% of all systems. The RTX 4090’s high percentile rank (91) is the primary driver, while the CPU’s rank (89) is slightly lower but still above average.

Q: Does the CPU’s performance limit the GPU at any resolution?

A: At 1920x1080 with Low settings, the average FPS is 196, which is very high. The CPU’s single-thread performance (passmark_single_thread score of 3984) likely becomes the limiting factor at this point, as the GPU has ample headroom to render frames faster if the CPU could issue more draw calls.