Thief

The pairing of an Intel Core i7-14700KF and NVIDIA GeForce GTX 1660 SUPER running Thief (2014-02-24) delivers a measured dataset that places this combination at rank 3179 out of 3720 tested combos. The CPU is a 20-core, 28-thread part with a base clock of 3.40 GHz and a boost of 5.60 GHz, belonging to the Raptor Lake architecture, and it sits in the 94th percentile among all CPUs with an average benchmark score of 70163. The GPU, in contrast, is a Turing-based TU116 chip with 6 GB of GDDR6 memory on a 192-bit bus, hitting a percentile of 53 among all GPUs. The data indicates a significant GPU ceiling, with the CPU dramatically outpacing the card in synthetics, and this mismatch shows clearly across the resolution landscape.

Resolution Scaling

The measured FPS table for Thief shows a consistent pattern: as resolution climbs from 1920x1080 to 2560x1440 and then to 3840x2160, the average frame rate drops sharply at every settings level. At the Low preset, the 1920x1080 run averages 120 FPS, but at 2560x1440 it falls to 82, and at 3840x2160 it reaches only 48. Medium follows the same shape: 91 FPS at 1080p, 62 FPS at 1440p, and 37 FPS at 4K. High goes from 69 FPS to 47 FPS to 28 FPS, while Ultra, the most demanding preset, yields 66 FPS, 45 FPS, and 27 FPS across the three resolutions. That is roughly a halving of FPS every time the pixel count jumps by a factor of about 2.25, which mirrors the pixel growth from 1080p to 4K, a typical signature of a GPU-bound workload.

The data also reveals that even lowering to the Low preset cannot keep the game smooth at 4K: 48 FPS is still far off the 120 FPS available at 1080p Low. The drop from 2560x1440 to 3840x2160 is steeper than the drop from 1920x1080 to 2560x1440, indicating that the ROP count and fill rate of the GPU become the bottleneck as the resolution climbs. The GTX 1660 SUPER has 48 ROPs and a pixel rate of 85.68 GPixel/s, numbers that appear modest by 2023 standards. Meanwhile, the CPU side of the equation is strong — the Cinebench R23 multi-core score of 44167 and a single-core score of 6235 show a processor far beyond what a 2014 game would need. The scaling behavior here points to the GPU as the sole constraint, because if the CPU were the limiting factor, we would typically see the frame rate hold steadier when lowering settings at higher resolutions. Here, even at Low settings, 4K yields only 48 FPS, which shows that the GPU cannot keep up with the raw pixel throughput, independent of graphical detail.

The pattern is consistent with a GPU scalar that reaches the limits at around 2560x1440 for any preset. The 1440p High slot yields 47 FPS, and 4K High drops to 28 FPS. This suggests that for this particular title, the GTX 1660 SUPER is best suited to 1080p or perhaps 1440p with medium detail. The existence of the Ultrawide preset further confirms the trend; even at 720p the same GPU would still be the limiting factor, but that is not covered in the data provided. In short, resolution scaling here is heavily cost-driven by GPU load, matching a 2014 title design that wasn’t intended to separately stress CPUs beyond a few cores.

GPU Role

The GPU is a GeForce GTX 1660 SUPER with a base clock of 1530 MHz and a boost clock of 1785 MHz. It brings with it 1408 shading units, 88 TMUs, and 48 ROPs, a texture rate of 157.1 GTexel/s, and 336.0 GB/s of memory bandwidth through a 192-bit bus. This amounts to 6 GB of GDDR6 memory, which is the STORA for the 4K textures and high shadow detail typical of Thief's Ultra preset. The GPU’s benchmark data shows a 3DMark Steel Nomad score of 1278, a Geekbench Vulkan score of 57102, and an average benchmark score of 12986. Its highest scores are in volcano Vulkan and OpenCL tests, but the raw rasterization pieces it lacks shine in the game’s multi-threaded rendering.

The 6 GB VRAM figure is directly relevant here. At 1920x1080, the frame buffer footprint is manageable, but at 3840x2160, the 4K resolution can consume more than 6 GB when streamed large texture sets. The measured FPS at 4K Ultra is 27 average, and at 4K High it is 28, which are both marginal. Even at 4K Low, the drop to 48 suggests that the memory pressure might be the reason why the medium preset’s minimum is 31 FPS and its max is 42 FPS — a 10‑FPS spread that indicates the GPU's processing resources are near their limits. The card’s memory clock effective rate is listed as 14 Gbps, which means the 336 GB/s bandwidth is the allocation that must be shared across the heavy fill-rate demands of the 3840×1200 buffer.

The GPU’s percentile of 53 shows that it sits just in the lower half of all GPUs, but its performance relative to its nearest rivals is close. In the GPU benchmark, the GTX 1660 Super scores 12986 on average, and the nearest 4k comparison shows that the RX 580 scores 12928, a delta of 0.4%, while the RTX 3050 Ti M both also within 0.4%. These numbers indicate that the card’s raw throughput percentile is extremely similar to cards with more modern features, but Thief as a 2014 title does not put any RT or tensor cores to use. The FP32 count of 5.032 TFLOPS and FP16 of 10.05 TFLOPS give a measure of the floating point capability, but the actual frame output at 4K is likely limited by the filter and ROP because the game runs on a classic deferred pipeline.

Further, the GPU is built on a 12 nm process with 6,600 million transistors on a 284 mm² die. That process is old by 2026 standards, but the card was launched with a 125 W TDP, a single 8-pin connector, and a dual-slot form factor. The lack of a game-optimized core clock in the pack means we are not looking at any clock frequency scaled beyond the base and boost. In practice it’s the base and boost, so the data shows the card’s ability to sustain itself over long sessions is consistent, but the frame contours at 4K stay very close—27 and 28 FPS for Ultra and High across two different presets—suggesting the card is already in its ceiling zone.

How This Combo Ranks

The combination of the i7-14700KF and the GTX 1660 SUPER lands at rank 3179 out of a total of 3720 tested combos. That places it in the 85th percentile from the bottom—meaning it outperforms only about 541 other combos. At first glance, the rank seems a paradox: the CPU is in the 94th percentile of all semiconductors, and the GPU is exactly in the 53rd percentile. The explanation lies in the fact that Thief is mostly GPU bound, so a powerful CPU is wasted unless paired with an even stronger GPU.

Looking at the CPU’s nearest rivals, the AMD Ryzen 7 9700F with an average score of 69996 is just 0.2% away, and the Ryzen 9 7940HX with 69875 is 0.4% away, while the Ryzen 9 7950X scores 69515 with a 0.9% delta. The Intel Core Ultra 7 265K sits 764 ahead with a -1.0% delta. These figures show a cluster of high-end CPUs separated by a fraction of a percent in aggregate benchmark scores. However, replacing the GPU would produce a different story; the combo rank here is dragged down by the GPU’s lottery because any modern mid-range GPU will beat the 1660 SUPER by 20% or more. The GPU’s own percentile of 53 means it sits below half of all GPUs, and its closest rival, the RX 580, is effectively identical in average score, which underscores that this card has been super by newer mid-range parts for the last several years.

The deltaPct values from the nearestRival_ory are the only intrigal info we have to gauge the CPU/GPU mismatch. The CPU has a 94th percentile, the GPU has a 53rd percentile, and the resulting combo is lower than the CPU’s rank alone. In fact, if the GPU were in the 94th percentile as well, the combo would not be at rank 3179. It is the GPU lag that creates this hierarchy. For a builder, this means that the bottleneck is unequivocally the GTX 1660 SUPER, and any upgrade to the processing should target the graphics card rather than the processor. The game’s total frame rate at all resolution/setting points is below the GPU’s old curve.

The measured FPS deltaPct across its entire spectrum shows no settings where the CPU is challenged. The CPU’s 20 cores, 28 threads, lots of cache (L1 80 KB per core, L2 2 MB per core, L3 33 MB shared) and its exceptional single-core benchmark (6235) could run a game like Thief with ease, but the minimum FPS values are only present for the Medium preset, and those still exceed the GPU frame times for the most part. As a result, the comboRankInGame attribute stands at 3179, which is not a representation of any CPU weakness but rather a reflection that the GPU is not pairing well with the very fast CPU.

Measured FPS Breakdown

The measured rows are as follows, all with average FPS unless otherwise noted, and the the Medium entries also list min and max FPS.

• 3840x2160 Ultrawide: High 28, Low 48, Medium 37 (min 31, max 42), Ultra 27.
• 2560x1440: High 47, Low 82, Medium 62 (min 53, max 72), Ultra 45.
• 1920x1080: High 69, Low 120, Medium 91 (min 77, max 105), Ultra 66.

At 1080p, the game is fully playable at every setting; Ultra at 66 FPS is only slightly below High at 105. That pairing and the 138 FPS Low indicate a very large headroom at native resolution. When moving to 1440p, the Ultra and High downtrend dramatically, with High hitting 47 and Ultra down to 45. Medium sits at 62, which is noticeably above the Low 82, but the only way to exceed that is to drop to “Low” at 82. The Lam Substance is the 1440p Medium, with a minimum of 53 and a maximum of 72, and gives a relatively tight performance band. 4 four lines on 4K, all four settings belov60 FPS max, with Low is the only preset that catches 48 FPS. Ultra at 27 FPS is virtually a slideshow.

The min/max data for 1080p Medium (77–105), 1440p Medium (53–72), and 4K Medium (31–42) reveal the specific variance. The spread widens by about 20% as resolution climbs, but not enough to give a sense of adaptive geometry. The lack of min/max values for the other settings—the pack only provides these rows for the Medium resolution—means we cannot draw further conclusions across settings, but the absence of synthetic frame times suggests the measurement recorded only the avg/min/max for the medium preset.

The 900p Ultra result is 66 average, which is 31% lower than the High average. The High at 69 is the highest Ultra runs at 105, but the gain is not linear. At 3840x2160, all four presets sit within a 21-FPS window (27 to 48), which is a narrow spread that highlights the fill-rate ceiling of the TU116. This behavior is typical of a pixel-bound scene; the GPU reach its pinne when the shader is trying to do that multiplied work on the same amount of geometry.

FAQ

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

A: The average FPS at 1920x1080 High is 69.

Q: Which resolution and preset combination provides the highest measured FPS?

A: The highest is 1920x1080 Low, with an average of 120 FPS.

Q: Does the GTX 1660 SUPER maintain 60 FPS at 2560x1440 on any preset?

A: At 2560x1440, only the Low preset reaches 82 FPS, while Medium achieves 62 FPS with a minimum of 53. High and Ultra yield 47 and 45, respectively, so those are below the 60 mark.

Q: What is the minimum FPS observed at 3840x2160 on Medium?

A: The minimum FPS is 31, from the Medium row at 3840x2160, which has an average of 37 and a maximum of 42.

Q: Is the CPU a likely bottleneck in this setup?

A: The data strongly suggests no. The CPU scores S$104 in Cinebench R23 single-core and 44167 multi-core, placing it in the 94th percentile overall. In contrast, the GPU only sits in the 53rd percentile, so the measured FPS scaling indicates the GPU is the performance ceiling.

Q: How many total combos are compared in this ranking?

A: The dataset includes 3720 tested combos, and the position of this pair is 3179.

Settings Recommendations

The measured FPS table provides three preset categories at each resolution. At 1080p, the average is always above 66 FPS, even at Ultra. The 1920x1080 Medium row sits at 91 FPS with a minimum of 77 and a maximum of 105, which is an ideal balance between visual quality and stability in the smooth, since it clips above on the entire 60-FPS window. The High preset gives 69 average, but the 105-number on Medium suggests that the render pipeline, while util path is slightly different.

For 1440p, the row that makes the most sense is 2560x1440 Medium, which averages 62 FPS with a minimum of 53 and maximum of 72. That’s a nearly consistent performance experience, and it is the only 1440p setting that has the published min/max range, which indicates stability. Owing to the GPU’s limited fill rate, 1080p Medium (91) outperforms 1440p Medium (62) while the two are a mixed set. If a user wants 60 or slightly lower with the highest pixel count, 1440p Low at 82 FPS gives a higher average but potentially higher low could be inferred from the absence of min/max data. However, the visible data points to 1440p Medium as the functional pick for balanced settings.

4K should be avoided for regular play. The best 4K result is 48 FPS at Low, which is below the threshold of any smooth 60 caption and is below the 1440p Low average of 8.2. In this case the base is so low that the GPU doesn’t have headroom for any finer presets. For those with 4K monitors, the graph strongly suggests lowering the content to 1440p rather than lower settings at 4K, because the 4K Low and 1440p Medium actually produce almost identical frame counts (48 vs 37).

Strictly from the data, the best experience is 1920x1080 Medium, which gives a 91 FPS average and a minimum of 77 FPS. That allows the game to be rendered with minimal hiccups, as shown in the max 105 line. The second highest simultaneous is 1440p Medium at 62 average, but it is lower than the 1080p medium, and its minimum of 53 shows it might dip below 60 on some frames. So, using a 1440p display, the choice is between 1440p Low (82) and 1440p Medium (62). Since the Low row has no min/max, the Medium row provides the guaranteed floor at 53, which may be acceptable for an older title. Thus, the various data supports the following guidelines: at 1080p use Medium for balanced performance; at 1440p use Medium for framerates over 60, or Low for the highest headroom; and no preset at 4K yields a stable frame rate, so lower resolution should be used.