Thief

# CPU Role — cores, clocks, and how they relate to this game's results

The AMD Ryzen 9 8945HX is a 16-core, 32-thread mobile processor built on Zen 4 architecture, codenamed Dragon Range. Its base clock of 2.50 GHz and boost clock of 5.40 GHz provide substantial single-thread headroom, which matters for a 2014 stealth title like Thief where game logic and AI routines often rely on fewer threads. The processor's 64 MB of L3 cache and 1 MB L2 per core help keep frequently accessed game data close, reducing latency in scene transitions typical of stealth gameplay.

The CPU's benchmark scores reflect strong multi-threaded performance. In Cinebench R23, it scores 42,713 multi-core and 6,030 single-core. The Passmark multi-thread score of 51,405 and single-thread score of 3,907 indicate the CPU's capability to handle both lightly threaded game loops and background tasks simultaneously. The R20 scores of 17,939 multi-core and 2,532 single-core align with this pattern. The CPU's percentile versus all CPUs is 95, meaning it sits in the top 5% of processors in the database — a position that suggests it is unlikely to bottleneck the GPU in most scenarios.

Its nearest rivals in CPU benchmarks show how tightly grouped the top tier is. The Intel Core Ultra 9 285HX scores 76,155 average, just 0.1% higher than the Ryzen 9 8945HX's 76,212 average. The AMD Ryzen 9 9950X3D scores 75,779, a 0.6% deficit. The AMD EPYC Embedded 8224P and AMD Ryzen Threadripper PRO 9945WX both score slightly higher, at 76,492 and 76,513 respectively, each about 0.4% ahead. This clustering implies that for Thief, CPU performance differences among top-tier chips are negligible — the game's frame rates will hinge more on GPU throughput.

The transistor count of 13,140 million across a 2x 71 mm² die size, manufactured on TSMC's 5 nm process, indicates high density but also efficiency. The 55 W TDP is notable for a 16-core part, suggesting the mobile platform can sustain high boost clocks without excessive thermal load. The integrated Radeon 610M graphics offer a fallback, but in this benchmark configuration, the discrete GPU handles rendering.

For Thief specifically, the CPU's 32 threads allow the game engine to spawn background tasks without starving the primary render thread. The Passmark physics score of 2,168 and floating-point math score of 117,453 suggest strong compute capability for any physics-based interactions in the game's environments. The data compression score of 677,755 and encryption score of 40,836 are less relevant to gaming but indicate the CPU's overall throughput headroom.

Given the measured frame rates, the CPU appears to provide ample single-core performance for Thief's engine, which was designed for hardware available around its 2014 release. The boost clock of 5.40 GHz is particularly relevant — older game engines often have a primary thread that cannot utilize many cores, so raw clock speed becomes decisive. With 105 FPS at 1080p High settings, the CPU is clearly not holding the system back, as the GPU still has headroom to produce three-digit frame rates.

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

The NVIDIA GeForce RTX 5070 is built on the Blackwell 2.0 architecture with the GB205 chip, manufactured on TSMC's 5 nm process. It carries 12 GB of GDDR7 memory on a 192-bit bus, delivering 672.0 GB/s of bandwidth. The memory clock runs at 1750 MHz, translating to 28 Gbps effective. The GPU's base clock is 2325 MHz with a boost of 2512 MHz. These specifications determine how quickly textures and geometry can be fed to the 6,144 shading units.

The GPU's compute capabilities are substantial: 30.87 TFLOPS FP32 performance, a pixel rate of 201.0 GPixel/s, and a texture rate of 482.3 GTexel/s. The 48 RT cores and 192 tensor cores add hardware support for ray tracing and AI acceleration, though Thief from 2014 predates widespread ray tracing adoption. The 80 ROPs handle final pixel output, which becomes a limiting factor at higher resolutions.

In Passmark benchmarks, the RTX 5070 scores 29,137 in G3D and 15,787 in GPU compute, with a 3DMark Steel Nomad DX12 score of 5,077. The Geekbench scores of 172,660 (OpenCL) and 178,923 (Vulkan) show strong compute performance across APIs. The GPU's percentile versus all GPUs is 82, placing it in the upper fifth of the database. Its launch MSRP is 549 USD.

The nearest GPU rivals in benchmark scores are intriguing. The AMD Radeon Pro 580 scores 40,318 average, a 0.1% delta from the RTX 5070's 40,377. The AMD Radeon Pro WX 7100 scores 40,063, 0.8% behind. The AMD Radeon Pro 5300 scores 40,870, 1.2% ahead. The NVIDIA RTX A500 Mobile scores 39,568, 2% behind. These are professional workstation cards, not gaming-focused GPUs, yet their average benchmark scores are remarkably close to the RTX 5070. This suggests that in synthetic workloads, the RTX 5070's real-time gaming advantage may not be fully captured by average scores.

For Thief, the 12 GB VRAM is more than sufficient — the game's textures and geometry from 2014 would fit comfortably in even half that capacity. The 672.0 GB/s bandwidth ensures that even at 4K Ultra settings, texture streaming should not cause hitches. The measured frame rates show the GPU scaling with resolution: at 1080p Ultra, it produces 100 FPS, while at 4K Ultra it drops to 40 FPS. This 60% drop from 1080p to 4K indicates that the GPU is the primary scaling factor, as the CPU's workload does not increase proportionally with resolution.

The RTX 5070's boost clock of 2512 MHz is relatively high for a mobile-capable GPU in this class, though this appears to be a desktop card given the 250 W TDP and dual-slot design. The 1x 16-pin power connector and 600 W suggested PSU confirm this is a full-power desktop SKU, not a laptop part. The display outputs of 1x HDMI 2.1b and 3x DisplayPort 2.1b support modern high-refresh monitors.

# Resolution Scaling — how FPS drops from 1080p to 4K and what it says about the limiting component

The measured FPS data across resolutions reveals a clear scaling pattern. At 1080p Low, the system achieves 181 FPS. Moving to 1440p Low drops this to 124 FPS, a 31.5% reduction. At 4K Low, the frame rate falls to 73 FPS, another 41.1% drop from 1440p. This steep decline indicates that the GPU is the limiting factor at lower settings — the CPU could likely push even higher frame rates, but the GPU's pixel throughput caps performance.

At Medium settings, the trend continues but with a less dramatic slope. 1080p Medium averages 138 FPS, 1440p Medium averages 94 FPS (a 31.9% drop), and 4K Medium averages 56 FPS (a 40.4% drop from 1440p). The minimum FPS at Medium settings shows similar scaling: 117 at 1080p, 80 at 1440p, and 47 at 4K. The maximum FPS values are 158 at 1080p, 109 at 1440p, and 64 at 4K.

High settings show a gentler drop between 1080p and 1440p: 105 FPS to 72 FPS, a 31.4% reduction. From 1440p to 4K, the drop is 72 FPS to 42 FPS, a 41.7% reduction. Ultra settings follow the same pattern: 100 FPS at 1080p, 68 FPS at 1440p (32% drop), and 40 FPS at 4K (41.2% drop from 1440p).

The consistent ~31% drop from 1080p to 1440p and ~41% drop from 1440p to 4K across all settings reveals a uniform scaling factor. This is characteristic of a GPU-bound scenario where pixel count scales nearly linearly with frame time. The number of pixels at 1440p is 1.78 times that of 1080p, and 4K is 2.25 times 1440p. If the GPU were fully saturated, we would expect frame rates to drop roughly in proportion to pixel count increase. The observed drops are slightly less than proportional, which suggests some efficiency gains at lower resolutions where the GPU's fixed overhead (like draw call processing) is spread across fewer pixels.

The fact that Low settings at 1080p produce 181 FPS — nearly double the 4K Ultra result of 40 FPS — confirms that resolution scaling is the dominant factor. The settings level has a smaller effect: at any given resolution, the difference between Low and Ultra is roughly 45-50% (e.g., 181 vs 100 at 1080p, 73 vs 40 at 4K). This means that for this hardware combination, resolution choice matters more than quality preset selection.

Data indicates the CPU is not the bottleneck at any resolution. Even at 1080p Low with 181 FPS, the CPU's 16 cores and 5.40 GHz boost clock provide enough headroom. If the CPU were limiting, we would see frame rates plateau across resolutions at high settings, but instead they scale smoothly with pixel count.

# How This Combo Ranks — use comboRankInGame vs other tested combos

The combination of the AMD Ryzen 9 8945HX and NVIDIA GeForce RTX 5070 achieves a combo rank of 1,453 out of 2,851 tested combos in Thief. This places it at approximately the 51st percentile — almost exactly the median of all tested configurations. This is surprising given the individual component rankings: the CPU is in the 95th percentile and the GPU is in the 82nd percentile. A system with two top-tier components ranking at the median suggests that either the game is highly optimized to run well on older hardware, or that many other combos in the database are extremely capable for this title.

The rank of 1,453 means there are 1,452 combos that perform better, and 1,398 that perform worse. This near-middle placement indicates that Thief's engine, released in 2014, does not scale dramatically with modern high-end hardware. The game's rendering techniques and draw call patterns were designed for the hardware of its era, so even a top-tier modern system may not push frame rates far beyond what a mid-range system could achieve.

Compared to the CPU's nearest rivals, the combo ranking suggests that CPU choice matters little for this game. The Intel Core Ultra 9 285HX, which scores nearly identically on average benchmarks (0.1% delta), would likely produce similar Thief results. The AMD Ryzen 9 9950X3D, with its 0.6% higher CPU benchmark score, might edge out slightly, but the difference would be within measurement noise.

Similarly, the GPU's nearest rivals — the AMD Radeon Pro 580, Pro WX 7100, Pro 5300, and RTX A500 Mobile — are all workstation cards with benchmark scores within 2% of the RTX 5070. These cards would likely produce comparable Thief performance, meaning the combo rank is not unique to this specific pairing.

The data implies that for a 2014 game like Thief, the hardware bottleneck is largely resolved by even modest modern components. The median rank of 1,453 out of 2,851 suggests that the game's performance ceiling is relatively low, and most modern systems hit that ceiling with ease. The 4K Ultra result of 40 FPS indicates that the GPU is still the limiter at extreme settings, but the game's age means that even this "low" frame rate is playable for a stealth title where reaction times are less critical.

# FAQ

Q: What is the highest average FPS achievable with this combo at 1080p?

A: The highest measured average FPS at 1080p is 181 FPS on Low settings. At Ultra settings, the average drops to 100 FPS.

Q: Does the CPU or GPU limit performance at 4K Ultra?

A: The data shows 40 FPS at 4K Ultra, compared to 42 FPS at 4K High. The GPU is clearly the limiting factor, as the CPU's performance headroom at lower resolutions (181 FPS at 1080p Low) far exceeds what the GPU can deliver at 4K.

Q: How does the RTX 5070 compare to its nearest GPU rivals in benchmark scores?

A: The RTX 5070's average benchmark score of 40,377 is within 2% of its nearest rivals. The AMD Radeon Pro 5300 is 1.2% higher, the AMD Radeon Pro 580 is 0.1% higher, the AMD Radeon Pro WX 7100 is 0.8% lower, and the NVIDIA RTX A500 Mobile is 2% lower.

Q: What is the FPS difference between Low and Ultra settings at 1440p?

A: At 1440p, Low settings average 124 FPS, while Ultra settings average 68 FPS — a difference of 56 FPS, or approximately 45% lower performance at Ultra.

Q: Is the Ryzen 9 8945HX's multi-threaded performance significantly better than its nearest rival?

A: No. The Ryzen 9 8945HX has an average benchmark score of 76,212, while the Intel Core Ultra 9 285HX scores 76,155 (0.1% lower) and the AMD Ryzen 9 9950X3D scores 75,779 (0.6% lower). These differences are negligible.

*Q: Does the combo rank in Thief reflect the individual component rankings?*

A: No. The combo ranks 1,453 out of 2,851 (about 51st percentile), while the CPU ranks in the 95th percentile and the GPU in the 82nd percentile. This suggests the game's performance ceiling is reached by many mid-range systems.

# Settings Recommendations

The measured data provides clear guidance for optimal settings. At 1080p, High settings deliver 105 FPS on average, which is well above the 60 FPS threshold for smooth gameplay. Ultra settings only increase fidelity marginally while dropping performance to 100 FPS — a 5% reduction for what is likely a subtle visual improvement. Medium settings at 138 FPS offer a substantial frame rate boost but sacrifice visual quality. For 1080p, High settings provide the best balance of visual fidelity and performance.

At 1440p, High settings average 72 FPS, which remains playable on most displays. Ultra drops to 68 FPS, a 5.6% reduction. Medium at 94 FPS offers smoother motion but with noticeably reduced detail. Given that 72 FPS is comfortably above 60, High settings are recommended for 1440p gaming.

At 4K, the situation changes. High settings average 42 FPS, which is below 60 and may feel less smooth. Medium averages 56 FPS, closer to playable but still not ideal. Low averages 73 FPS, which is playable but sacrifices significant visual quality. Ultra at 40 FPS is borderline for a stealth game, but the lower frame rate may impact the feel of quick movements. For 4K, Medium settings at 56 FPS offer the best compromise between visual quality and playability, though players prioritizing frame rate should consider Low.

The minimum FPS data at Medium settings shows the worst-case frame times: 117 at 1080p, 80 at 1440p, and 47 at 4K. These minimums indicate that even at Medium settings, 4K can dip below 50 FPS in demanding scenes. This reinforces the recommendation for Medium at 4K as the highest playable setting.

# Measured FPS Breakdown — resolution by resolution, settings by settings, exact numbers

1080p (1920x1080):

• Low: 181 FPS average
• Medium: 138 FPS average, 117 FPS minimum, 158 FPS maximum
• High: 105 FPS average
• Ultra: 100 FPS average

1440p (2560x1440):

• Low: 124 FPS average
• Medium: 94 FPS average, 80 FPS minimum, 109 FPS maximum
• High: 72 FPS average
• Ultra: 68 FPS average

4K (3840x2160):

• Low: 73 FPS average
• Medium: 56 FPS average, 47 FPS minimum, 64 FPS maximum
• High: 42 FPS average
• Ultra: 40 FPS average

The pattern across all resolutions is consistent: Low settings provide roughly 80-90% more FPS than Ultra. The Medium settings row is the only one with complete min/max data, showing that frame time consistency scales with resolution — the spread between minimum and maximum FPS narrows as resolution increases (41 FPS spread at 1080p, 29 at 1440p, 17 at 4K). This narrowing suggests that at higher resolutions, the GPU operates closer to its sustained limits, with less burst capacity. At 1080p Medium, the GPU has enough headroom to spike to 158 FPS, while at 4K Medium, it caps at 64 FPS.

The jump from Low to Medium at each resolution costs approximately 23-25% of average FPS (181 to 138 at 1080p, 124 to 94 at 1440p, 73 to 56 at 4K). The jump from Medium to High costs roughly 24-28% (138 to 105 at 1080p, 94 to 72 at 1440p, 56 to 42 at 4K). The jump from High to Ultra is much smaller, at 5-7% (105 to 100 at 1080p, 72 to 68 at 1440p, 42 to 40 at 4K). This diminishing cost of Ultra settings means that players who want maximum fidelity can enable Ultra with minimal FPS penalty compared to High.