Thief

Resolution Scaling

The measured FPS data for Thief shows a clear and consistent scaling pattern as resolution increases. At 1080p with Low settings, the AMD Ryzen AI Max+ 395 and NVIDIA GeForce RTX 5090 combination delivers 204 FPS average. Moving to 1440p Low drops that to 180 FPS, and at 4K Low the average falls to 108 FPS. The percentage drop from 1080p to 4K at Low settings is approximately 47%, indicating that the GPU becomes significantly more constrained as pixel count rises.

For Medium settings, the scaling is equally pronounced. The 1080p Medium average of 184 FPS falls to 139 FPS at 1440p, then to 82 FPS at 4K. That represents a 55% reduction from 1080p to 4K. The High preset tells a similar story: 154 FPS at 1080p, 106 FPS at 1440p, and 62 FPS at 4K — a 60% drop across the resolution ladder. Ultra settings show the tightest scaling curve with 147 FPS at 1080p, 101 FPS at 1440p, and 59 FPS at 4K, a 60% decline.

The data suggests that at 1080p, the system is not fully GPU-bound at lower presets. The jump from Low to Ultra at 1080p only costs 57 FPS (204 down to 147), which is a modest 28% penalty. This relatively small spread indicates that the CPU and memory subsystem are capable of feeding the GPU adequately at this resolution, but the RTX 5090's raw throughput is still the dominant factor. At 4K, however, the GPU is clearly the limiting component — the spread from Low to Ultra narrows to just 49 FPS (108 down to 59), and even the lowest preset cannot break past 108 FPS average.

The 4K Ultra result of 59 FPS sits just below the 60 FPS threshold commonly targeted for smooth gameplay. The 4K High result of 62 FPS barely clears it. Meanwhile, 1440p delivers a much more comfortable margin: every preset except Ultra stays above 100 FPS, with Low reaching 180 FPS. The 1080p results are uniformly high, with the lowest measured average being 147 FPS at Ultra.

This scaling behavior points to a system where the RTX 5090's 32 GB of GDDR7 memory and 1.79 TB/s bandwidth are sufficient to handle Thief's demands at any resolution, but the GPU's compute throughput becomes the bottleneck at 4K. The 82 FPS figure at 4K Medium versus 108 FPS at 4K Low shows that even a single preset step carries a meaningful performance cost at this resolution, reinforcing that pixel fill rate and shading workload are the primary constraints.

GPU Role

The NVIDIA GeForce RTX 5090 is built on the Blackwell 2.0 architecture with a GB202 chip fabricated on a 5 nm process from TSMC. The GPU contains 92,200 million transistors on a 750 mm² die, yielding a transistor density of 122.9M per mm². Its 21760 shading units, 680 texture mapping units, and 176 ROPs provide substantial raw compute capacity. The FP32 throughput is rated at 104.8 TFLOPS, with FP16 matching that at 104.8 TFLOPS (1:1).

Clock speeds are relatively modest for a flagship: a base clock of 2017 MHz and a boost clock of 2407 MHz. The memory operates at 1750 MHz with 28 Gbps effective speed. The 512-bit memory bus combined with 32 GB of GDDR7 yields a bandwidth of 1.79 TB/s, which is more than sufficient for Thief's texture and geometry demands even at 4K Ultra.

The RTX 5090's benchmark results place it in the 92nd percentile among all GPUs, with an average benchmark score of 79842. Its nearest rivals according to the data are the NVIDIA Tesla P100 PCIe 16 GB (average score 79605, deltaPct 0.3) and the Tesla P100 PCIe 12 GB (79605, deltaPct 0.6). The AMD Radeon RX 6850M XT trails by 1.1%, while the AMD Radeon Pro Vega 64X leads by 1.4%. These narrow deltas suggest that the RTX 5090's synthetic benchmark scores are competitive but not overwhelmingly dominant over these specific workstation-class competitors.

In Thief specifically, the GPU's performance is characterized by the 4K results. At 4K Ultra, the average FPS is 59, which is the lowest measured figure in the entire dataset. The 4K High average of 62 FPS and 4K Medium average of 82 FPS show that the GPU has headroom to spare when settings are reduced. The 4K Low result of 108 FPS demonstrates that the RTX 5090 can deliver high frame rates even at 4K when the workload is lightened.

The VRAM capacity of 32 GB is never stressed by Thief, a game from 2014 that predates modern texture-heavy titles. The GPU's memory bandwidth, however, plays a role in the scaling behavior observed. The 1.79 TB/s bandwidth ensures that even at 4K, memory throughput is not a limiting factor; the bottleneck instead lies in the shading and rasterization workload. The 423.6 GPixel/s pixel rate and 1,636.8 GTexel/s texture rate are the specifications most relevant to Thief's rendering demands.

The RTX 5090's TDP is 575 W, with a suggested PSU of 950 W. Its dual-slot design and 304 mm length make it a large card, but these physical characteristics do not directly impact the measured FPS data. What matters for Thief is that the GPU can sustain its boost clock across the benchmark runs, and the data suggests it does — the 4K Medium min FPS of 70 and max FPS of 94 bracket the average of 82 closely, indicating stable performance without thermal throttling artifacts.

FAQ

*Q: What is the highest average FPS recorded for this CPU-GPU combination in Thief?*

A: The highest average FPS is 204, achieved at 1920x1080 resolution with Low settings.

Q: Can this system maintain 60 FPS at 4K with Ultra settings?

A: No. The 4K Ultra average is 59 FPS, which falls just below the 60 FPS threshold. The 4K High preset averages 62 FPS, which does clear it.

Q: How does the RTX 5090 compare to its nearest rival in synthetic benchmarks?

A: The RTX 5090's average benchmark score is 79842, which is 0.3% ahead of the NVIDIA Tesla P100 PCIe 16 GB (79605) and 1.1% ahead of the AMD Radeon RX 6850M XT (78940). The AMD Radeon Pro Vega 64X leads by 1.4% with a score of 80959.

Q: What is the FPS penalty for moving from 1440p to 4K at Medium settings?

A: At 1440p Medium, the average FPS is 139. At 4K Medium, it drops to 82 FPS. That is a 57 FPS reduction, or approximately a 41% performance loss.

Q: Does the CPU or GPU become the limiting factor at 1080p?

A: The data suggests the GPU remains the primary constraint. The spread between Low and Ultra at 1080p is 57 FPS (204 down to 147), which is a 28% penalty. If the CPU were the bottleneck, this spread would be smaller, as the GPU would have headroom to render higher settings without significant FPS loss.

*Q: What is the combo's rank among all tested configurations in Thief?*

A: This CPU-GPU combination ranks 473 out of 3720 tested combos, placing it in the top 12.7% of all configurations.

Settings Recommendations

Based on the measured FPS data, the optimal settings depend heavily on the target resolution and desired frame rate. For 1080p displays, every preset delivers playable frame rates. The Low preset at 204 FPS is excessive for most displays, and even Ultra at 147 FPS provides a smooth experience. The Medium preset at 184 FPS with a min of 157 and max of 212 offers the best balance of visual quality and performance headroom at this resolution.

At 1440p, the High preset at 106 FPS is the recommended choice for most users. It provides a substantial visual upgrade over Medium (139 FPS) while still maintaining a comfortable margin above 60 FPS. The Ultra preset at 101 FPS is also viable, with only a 5 FPS penalty over High. For users with high-refresh-rate 1440p monitors, the Low preset at 180 FPS is the only option that fully exploits such displays, but the visual degradation is significant.

At 4K, the situation is more constrained. The Ultra preset at 59 FPS is borderline — it misses the 60 FPS target by a single frame. The High preset at 62 FPS clears that threshold but with minimal margin. For users prioritizing frame rate stability, the Medium preset at 82 FPS is the safest recommendation. It provides a 20 FPS buffer over High while still offering better visuals than Low. The Low preset at 108 FPS is the only 4K option that provides consistently high frame rates, but it sacrifices nearly all graphical enhancements.

The data indicates that the step from Medium to High costs roughly 20 FPS at 4K (82 down to 62), while the step from High to Ultra costs only 3 FPS (62 down to 59). This nonlinear scaling suggests that High is the efficiency sweet spot at 4K — it captures most of the visual benefit of Ultra while maintaining a playable frame rate. At 1440p, the jump from High to Ultra costs only 5 FPS (106 down to 101), making Ultra a reasonable choice for users who want maximum quality. At 1080p, the same jump costs 7 FPS (154 down to 147), again making Ultra viable.

For users with variable refresh rate displays, the measured min FPS figures provide additional guidance. The 1440p Medium result shows a min of 118 and max of 160, a 42 FPS range that is manageable for adaptive sync. The 1080p Medium min of 157 and max of 212 spans 55 FPS, which is wider but still falls within the range of most high-refresh monitors. The 4K Medium min of 70 and max of 94 is the tightest range in the dataset, indicating stable frame pacing at that setting.

Measured FPS Breakdown

1920x1080:

• Low: 204 FPS average
• Medium: 184 FPS average, 157 min, 212 max
• High: 154 FPS average
• Ultra: 147 FPS average

At 1080p, the system delivers its strongest performance. The Low preset at 204 FPS is the single highest average in the entire dataset. The Medium preset provides a 30 FPS penalty over Low but still exceeds 180 FPS. High and Ultra are closely matched, with only a 7 FPS difference between them (154 vs 147). The min FPS of 157 at Medium suggests that even under demanding scenes, the frame rate stays well above 144 FPS.

2560x1440:

• Low: 180 FPS average
• Medium: 139 FPS average, 118 min, 160 max
• High: 106 FPS average
• Ultra: 101 FPS average

The 1440p results show a more pronounced separation between presets. Low retains a high average of 180 FPS, but Medium drops to 139 FPS. High and Ultra cluster together at 106 and 101 FPS respectively, a mere 5 FPS apart. The Medium min of 118 FPS indicates that even the lowest frames remain comfortably above 60 FPS. The progression from Low to Ultra represents a 44% performance reduction, from 180 down to 101 FPS.

3840x2160:

• Low: 108 FPS average
• Medium: 82 FPS average, 70 min, 94 max
• High: 62 FPS average
• Ultra: 59 FPS average

At 4K, the performance curve steepens. Low still manages 108 FPS, but Medium drops to 82 FPS. High falls to 62 FPS, and Ultra bottoms out at 59 FPS. The Medium min of 70 FPS and max of 94 FPS bracket the average tightly, suggesting consistent performance. The gap between High and Ultra is just 3 FPS, making Ultra a marginal upgrade over High. The overall spread from Low to Ultra is 49 FPS, indicating that the GPU is the dominant constraint at this resolution.

The complete dataset shows that the RTX 5090 delivers playable frame rates at all settings and resolutions tested, but the 4K Ultra result of 59 FPS is the only configuration that fails to reach 60 FPS. The 4K High result of 62 FPS is the only other configuration below 100 FPS at the High preset or above.

CPU Role

The AMD Ryzen AI Max+ 395 is a 16-core, 32-thread processor based on the Zen 5 architecture with the Strix Halo codename. It operates with a base clock of 3.00 GHz and a boost clock of 5.10 GHz, with a TDP of 55 W. The CPU is fabricated on a 4 nm process from TSMC and uses the AMD Socket FP11 interface. Its cache hierarchy includes 80 KB of L1 per core, 1 MB of L2 per core, and 64 MB of shared L3 cache.

In synthetic benchmarks, the Ryzen AI Max+ 395 demonstrates strong multi-threaded performance. Its Cinebench R23 multicore score is 35314, while the single-core score is 2044. Geekbench results show 20992 multicore and 2464 single-core. The Passmark multithread score is 54934, with a single-thread score of 4147. The CPU's average benchmark score is 77740, placing it in the 95th percentile among all CPUs.

The nearest rival comparisons show narrow margins. The AMD Ryzen Threadripper PRO 9945WX scores 76513, which is 1.6% behind the Ryzen AI Max+ 395. The AMD EPYC Embedded 8224P scores 76492, also 1.6% behind. The Intel Core i9-14900K leads by 1.7% with a score of 79097. The AMD Ryzen 9 8945HX trails by 2% with 76212. These results indicate that the Ryzen AI Max+ 395 is competitive with high-end desktop and workstation processors despite its mobile market segment designation.

In Thief, the CPU's role is evident from the 1080p performance data. The game is not heavily threaded, but the 16-core, 32-thread configuration ensures that background tasks and game logic never starve the rendering pipeline. The 1080p Low result of 204 FPS suggests that the CPU can feed the RTX 5090 at very high frame rates. The relatively small spread between Low and Ultra at 1080p (57 FPS) indicates that the GPU, not the CPU, is the primary constraint even at this resolution.

The CPU's memory support includes LPDDR5X with a quad-channel bus and 256.0 GB/s bandwidth. This high bandwidth is relevant for Thief because it ensures that texture data and geometry can be streamed to the GPU without stalling. The 64 MB shared L3 cache also helps by keeping frequently accessed data close to the cores. The PCIe Gen 4 interface with 16 lanes provides adequate bandwidth for the RTX 5090, though the GPU itself is PCIe 5.0 capable.

The Passmark physics score of 3481 and data compression score of 690650 provide additional context. These benchmarks suggest strong integer and floating-point performance, which translates to responsive gameplay in Thief's stealth mechanics. The CPU's single-thread score of 4147 is particularly relevant for a game from 2014, which likely relies more heavily on single-threaded performance than modern titles. The 5.10 GHz boost clock ensures that individual cores can reach high frequencies when needed.

The CPU's production status is active, with a release date of 2025-01-05. Its 4 nm process node and 2x 70.6 mm² die size indicate a modern, efficient design. The 55 W TDP is notably low for a 16-core processor, which suggests that thermal throttling is unlikely to be a factor in sustained gaming sessions. The data shows that the Ryzen AI Max+ 395 provides sufficient headroom to keep the RTX 5090 well-fed across all tested resolutions and settings, with the GPU emerging as the limiting factor only at 4K Ultra.