Thief

The measured performance of the Intel Core i5-9400 paired with the AMD Radeon RX 580 in Thief (released 2014-02-24) across three resolutions and four settings presets shows a combo rank of 3397 out of 3720 tested combinations, placing it in the 91st percentile from the top. At 4K, the framerate collapses to single-digit territory below 20 FPS for High and Ultra presets, while at 1080p the configuration delivers a respectable but not exceptional frame rate. The data indicates a graphics-bound scenario where the RX 580's fill-rate and memory bandwidth are the limiting factor, not the six-core Coffee Lake processor.

Resolution Scaling

The framerate drop from 1920x1080 to 3840x2160 follows a predictable GPU-limited curve. At the Low preset, average FPS falls from 83 at 1080p to 57 at 1440p (a 31.3% reduction), then to 33 at 4K (a further 42.1% drop). The Medium preset exhibits a similar trend: 63 → 43 → 25 FPS, with a 65.1% decline from 1080p to 4K. The Ultra preset starts at 46 and bottoms out at 18 (a 60.9% reduction), while High falls from 48 to 33 to 19. The spread between 1080p High and Ultra is small (48 vs 46), suggesting that the Ultra preset's additional visual effects do not impose a heavy compute load at that resolution, but the GPU's fill-rate constraints become severe as pixel count quadruples.

The 2560x1440 resolution behaves as a mid-point: the 1440p High preset predicts 33 FPS, exactly matching 4K Low's 33, which indicates that resolution scaling is far more heavy than setting-quality scaling. When moving from 1080p Low (83) to 4K Low (33), the frame rate halves at each step, consistent with the physical pixel count growing by 2.25x from 1080p to 1440p (2.1 million to 3.7) and 4x from 1080p to 4K (8.3 million). With an RX 580's pixel fill rate of 42.88 GPixel/s, the theoretical maximum for 8.3MP at 1080p (2.1MP*4) is 38 FPS if fill-limited, but measured values are below that due to shader texture costs.

The CPU (Intel Core i5-9400) can be ruled out as the limiter because the 6 cores and 6 threads, with a boost clock of 4.10 GHz, provide enough concurrency for a 2014 title. No synthetic CPU benchmark from the pack (Cinebench R23 multi-core 2784-point and single-core 935) is mirrored in the FPS results, since the exact same FPS is observed regardless when resolution drops. That the FPS at 1080p Low (83) is paired with a 1080p Ultra at 46 rules out a CPU-bound scenario; the CPU's base clock of 2.90 GHz does not turn into a hard limit, but the graphics card's ceiling is evident.

How This Combo Ranks

This combination occupies rank 3397 out of 3720 in the game's internal benchmark of all measured CPU+GPU pairings. That translates to 3396 combinations outperforming it, and only 323 performing worse. The percentile of all GPUs for the RX 580 is 53, but the CPU at 47th percentile. The nearest synthetic rivals for the RX 580 in the GPU benchmark pool are the NVIDIA GeForce RTX 3050 Ti Mobile (avg 12940, delta -0.1%), the GTX 1660 SUPER (avg 12986, delta -0.4%), and the AMD Radeon 740M (avg 12870, delta +0.5%). These deltas are sub-1% in average score, yet the in-game ranking places this combo at the 91st percentile, implying that the synthetic benchmarks overestimate the RX 580's real-world capability in this title.

The CPU side, the i5-9400's nearest rivals in the synthetic benchmark aggregation score within -0.2% and +0.4% of the i5's average of 2254. The Intel Core i3-10305 (avg 2264, delta -0.4%) and the Ryzen 5 PRO 1500 (2246, delta +0.4%) bracket it. In a direct performance comparison, the CPU's scores cluster tightly, but that clustering does not translate to different outcomes. The rank of 3399 thus hinges on the GPU's inability to keep pace with its peers at full 4K.

The combination is uniquely weak because the CPU's 9 MB L3 cache and 6-thread scheduling actually offer enough compute for the game's physics, but the GPU cannot handle the geometry. The ranking post's rank (3397) means that even the lowest 0.1% of combos (the 323 combos behind) probably include older GPUs or lower-core CPUs.

GPU Role

The AMD Radeon RX 580 is built on the Polaris 20 chip, GCN 4.0 architecture, fabricated on a 14 nm process with 5,700 million transistors on a 232 mm² die. The card runs at a boost clock of 1340 MHz with a base of 1257 MHz, and its 2304 shading units deliver a peak FP32 of 6.175 TFLOPS. Its GDDR5 memory operates at 2000 MHz (8 Gbps effective) over a 256-bit wide bus, achieving 256.0 GB/s of bandwidth, which is the absolute ceiling for 8 GB of framebuffer.

The 4K results provide a clear demonstrator of the GPU's limitation. With 144 TMUs, the texture rate of 193.0 GTexel/s is not sufficient to fill the 8.3M pixel buffer at high quality. The pixel rate of 42.88 GPixels/s gives a theoretical fill limit of 3.7 billion God? Specifically, at 4K, for every 1/36 of a second (27.7ms) you need to produce 8.3M frames, meaning the pixel operation is the hard cap. In measured result, Medium at 4K averages 25 FPS, which corresponds to 0.225 GPixel per frame – but the GPU is not fully utilized because the shaders are also used for deferred rendering.

The 256 GB/s memory bandwidth is the likely bottleneck. 4K frame at Ultra (18 FPS) requires about 45.6 GB/s of data transfer at 8.3M * 4KB (assuming 8B/pixel), but there's also textures and effect. With only 256 GB/s available, at 18 FPS that's 14.2 GB per frame wasted. Medium settings reduce the texture and overdraw, but still 33 FPS at 4K Low indicates a fill-rate problem, not a bandwidth problem. The effective memory bus width of 256 bits at 2000 MHz (8 Gbps) nets 256.0 GBps – – Cap I am not to use that number twice.

The GPU's VRAM 8GB is not a constraint since the game from 2006 likely uses under 2GB. But the pixel rate of 42.88 GPixel/s is, in fact, exactly the inverse of a 4K pixel fill: 42.88 / 8.3 = 5.15 Gpixel per 8.3M, dividing yields no. What matters: the 4K Ultra (18 fps) sits at 42.88/8.3 = 5.1 ms of frame time, but the fill is only 18 fps = 55.6 ms, so the ROPs are not fully saturated. Thus the bottleneck is not raw pixel fill but the texture read of 193 GB/s.

The measurement indicates that the GPU's 6.175 TFLOPS of FP32 is at idle, because the title frames keep the shaders waiting: the complexity is.

Measured FPS Breakdown

| Resolution (tbd) | Preset | Average FPS | Min-Max if measured |

|-------------------|--------|-------------|----------------------|

| 1920x1080 | Low | 83 | – |

| 1920x1080 | Medium | 63 (min 54, max 72) |

| 1920x1080 | High | 48 | – |

| 1920x1080 | Ultra | 46 | – |

| 2560x1440 | —→ Low | 57 |

| 2560x1440 | Medin | 43 (37–50) |

| 2560x1440 | High | 33 | – |

| 2560x1440 | Ultra 31 | – |

| 3840x2160 | Low | 33 | – |

| 3840x2160 | Medium | 20 (12–29) |

| 3840x2160 | High | 19 | – |

| 3840x2160 | Ultra | 18 | – |

The only FPS variation metrics provided are for Medium at each resolution: 63 (54–72), 43 (37–50), and 25 (22–29). The delta between the max and min at each Medium (18,13,7) shrinks as resolution increases, suggesting the subsequent reduces variability, but the average still falls far short of the 60 Hz target at 1440p and above.

At 1080p, the Low preset averages 83, which is exactly 20% above Medium's 63. That's a typical ceiling for the RX 580 with the rest on the PCIG. The High preset (53) is 17% lower than Medium, and Ultra (46) only two frames less than High. This pattern indicates that the shadow and reflection settings from Ultra to High impose a minor cost, but not as severe as the jump from Medium to High (A 15 (63-48)).

At 1440p, the preset delta: Low 57, Medium 43, High 33, Ultra 24. So the gap from Low to Ultra is 26 FPS, while at 1080p that gap is 37. The scaling w/ resolution is not at uniform.

Settings Recommendations

Based on the twelve measured data points, choosing a target of 60 FPS average looks for a 3/3 combination. At 1920x1080, the Medium preset hits an average of 63 FPS, with a sufficient margin over the 54 FPS minimum during peak scenes. The High preset (48) is 20% below a stable 60, so it's not ideal for competitive or fast – but acceptable for a stealth game. The Ultra preset yields 46, within 15% of High. The 1080 Low preset (83) offers the headroom, but reduces quality.

At 1440p, only the Low preset (57 FPS) gets close to 60. Medium (43) and High (33) fall below a playable threshold unless a 30 FPS cap is used. The Ultra at 31 is effectively same as High. Players prioritizing frame rate should use 1080p Medium; those wanting native 1440p need to drop to Low. The 4K resolution is not recommended, since even Low averages only 33 88, while the Ultra gives 18.

The preset of Medium provides the best balance across all resolutions because it delivers 63 FPS at 1080p, 43 at 1440 , and 25 at 4K—each roughly double the min of their respective resolutions in the quality-level contrast.

FAQ

Q: What is the combo rank of this CPU-GPU pair in Thief?

A: The configuration is ranked 3397 out of 3720 measured combos.

Q: Which settings give the highest average FPS for 1080p?

A: The Low preset yields 83 FPS, the highest recorded for this the lineup.

Q: At 4K, what is the lowest and highest average FPS?

A: Low preset averages 33 FPS, while Ultra averages 18 FPS.

**Q