# Analysis of 3DMark Time Spy Extreme GPU Performance Across Profiles

## Benchmark Scores Overview

We tested four GPU profiles – **Default (Stock)**, **OC Mode**, **Silent Mode**, and a **Custom Undervolt/OC** (945 mV, 2950 MHz core, 31.5 GHz memory) – using 3DMark _Time Spy Extreme_. The overall benchmark scores and sub-scores for each profile are summarized below:

| **Profile** | **Overall Score** | **Graphics Score** | **CPU Score** | **Graphics Test 1 FPS** | **Graphics Test 2 FPS** |
| --- | --- | --- | --- | --- | --- |
| **Default** (Stock) | 18,946 | 26,338 | 7,314 | 161.5 FPS | 159.9 FPS |
| **OC Mode** (Factory OC) | 18,918 | 26,213 | 7,342 | 160.5 FPS | 159.3 FPS |
| **Silent Mode** (Quiet BIOS) | 18,727 | 25,642 | 7,408 | 158.8 FPS | 154.1 FPS |
| **Custom** (945 mV Undervolt @ 2950 MHz) | 19,209 | 26,692 | 7,421 | 163.4 FPS | 162.3 FPS |

**Key observations:** The Custom undervolt/OC profile achieved the highest overall **Time Spy Extreme score** (19,209) and highest **Graphics score** (26,692), slightly outperforming the stock and OC modes in GPU performance. The Silent profile trailed with the lowest graphics score (≈2.6% lower than Default). CPU scores remained very similar across runs (around 7,3xx–7,4xx), with minor variability likely within run-to-run noise. Graphics Test framerates followed the same pattern: the Custom profile led with ~163 FPS in Test 1 and ~162 FPS in Test 2, slightly ahead of Default/OC (~160–161 FPS), while Silent mode was about 1–3% lower. These results indicate the undervolt + memory OC provided a small uplift in GPU performance over stock, whereas the factory OC yielded virtually no net gain (stock and OC scores are within ~0.1%), and the Silent BIOS incurred a modest performance penalty to reduce noise.

## GPU Power Draw Comparison

_GPU power draw (watts) over time during the Time Spy Extreme benchmark, for each profile._ The **power draw** profiles reveal clear differences in how each mode utilizes the GPU’s power budget. Both the Default and OC Mode run at very high power levels, frequently hitting ~550–600 W during the graphics tests. In fact, all profiles showed spikes up to their power limit (the Stock/OC profiles peaked around 600 W). The **Silent** mode consistently drew less power, generally around ~500–550 W, reflecting its conservative tuning. The Custom undervolt profile had the lowest power consumption, mostly in the ~450–520 W range during load. Notably, the undervolted GPU never draws as much power as the stock card despite achieving higher performance – a clear efficiency gain. All profiles show two major power “spikes” corresponding to the two GPU test phases, after which power drops near idle during the CPU test segment (e.g. after ~180 s in the timeline). Throughout the heavy GPU loads, the Default and OC BIOS run the card at the brink of its power limit (power limit throttling was indicated in the HWiNFO logs, as the “Performance Limit – Power” flag toggled on). By contrast, the undervolted profile uses significantly less power for the same workload, indicating fewer instances of hitting the power ceiling. In summary, the Custom profile consumed the least power for slightly higher performance, whereas the Silent BIOS saved some power at the cost of performance, and the OC BIOS did not materially increase power beyond stock.

## GPU Clock Frequency Behavior

_GPU core clock frequency (MHz) over time during Time Spy Extreme._ Examining the **GPU core clocks**, we see the Default and OC Mode profiles sustaining the highest frequencies during the graphics tests. The stock card hovered around ~2800–2890 MHz under load, while the OC Mode achieved marginally higher peaks (up to ~2915 MHz) – only ~30 MHz above stock, reflecting the small factory OC. The **Silent** profile ran the GPU at a notably lower clock: roughly ~2750–2820 MHz during the tests. This ~100 MHz drop explains its reduced graphics score. The Custom 945 mV undervolt showed clocks around ~2770–2820 MHz, very similar to Silent’s frequency range. Despite the undervolt, the GPU still boosts fairly high, but it tops out around 2820 MHz at 0.94 V, instead of ~2880+ MHz at ~1.06–1.08 V on the stock/OC BIOS. The plotted clock curves for all profiles exhibit momentary dips between 0–1000 MHz at certain points – these correspond to scene transitions or the idle periods between the two graphics tests and before the CPU test, when GPU load briefly drops and the clock throttles down. Importantly, none of the profiles show any erratic clock fluctuations during the sustained load periods – each profile’s clock is relatively flat when the GPU is fully engaged, aside from minor oscillations (likely due to hitting power or voltage limits). The Custom undervolt in particular maintained a very steady ~2.8 GHz core during the tests, indicating no instability or excessive clock jitter from the undervolt. In contrast, the Stock/OC profiles exhibit slight clock tapering at times, consistent with hitting the power limit and voltage reliability limit (the logs indicated occasional “Voltage Reliability” limiting on stock/OC, meaning the GPU couldn’t maintain higher boost due to voltage/frequency curve constraints).

## GPU Temperature Results

_GPU core temperature (°C) over time in the benchmark._ All four profiles kept GPU core temperatures well under control – **none exceeded ~60 °C** during _Time Spy Extreme_. The Default and OC runs peaked around 58–60 °C. The Silent and Custom profiles show similar peak temps (~59–60 °C as well). It’s interesting that Silent mode, despite drawing less power, reached roughly the same max temperature; this is likely because Silent BIOS favors lower fan speeds, allowing a bit more heat for quieteroperation. The Custom undervolt, on the other hand, used normal fan behavior but much less power, resulting in a very cool run (mostly mid-50s °C, peaking ~60 °C). During each run, the GPU warmed up quickly during the first graphics test and then plateaued around the high-50s °C. There is no sign of thermal throttling in any profile – the temperature remained far below any thermal limit (the “Performance Limit – Thermal” flag was not triggered in the logs). In summary, all profiles kept the GPU core around 55–60 °C under load, with no meaningful thermal differences except slightly lower average temps for the undervolted profile (due to its reduced power) and the Silent profile running a tad warmer initially (likely due to its relaxed cooling).

## GPU Voltage and Utilization

_GPU core voltage (V) over time during the benchmark._ The applied core voltage differed dramatically between the stock/OC versus the undervolt profile. **Default and OC Mode** allowed the GPU to reach around **1.06–1.08 V** at peak boost. In the plot, their voltage line hovers just above 1.05 V during the GPU tests. The **Silent** profile also operated at high voltage (up to ~1.07 V) but generally a touch lower on average (~1.04–1.05 V under load) due to the lower clocks and power draw. In stark contrast, the Custom undervolted GPU was locked at ~0.93–0.94 V throughout the tests – its voltage trace is flat at the enforced 945 mV level. This reduced voltage is the reason its power draw was so much lower, yet it still delivered comparable performance. The GPU utilization was effectively 100% during both graphics tests for all profiles (the GPU “Core Load” stayed at 98–100% whenever a graphics test was running). Even the Silent and undervolt modes kept the GPU fully busy; they simply achieved that load at lower clock speeds or voltage. One minor note: the stock/OC profiles did hit the silicon’s voltage-frequency limit – the logs show “Yes” for Reliability Voltage limit, meaning they couldn’t increase voltage further for higher clocks. The undervolted profile, by definition, never hit that particular limit (it stayed well below max voltage), but it did occasionally register “Utilization” limit when the GPU wasn’t fully utilized (e.g. during the CPU test). Overall, the undervolt profile ran at a significantly lower voltage (~0.94 V vs ~1.06+ V) while still keeping the GPU 100% utilized, indicating a more power-efficient operating point without under-utilizing the hardware.

## Performance per Watt Efficiency

A crucial benefit of undervolting is improved **performance per watt**. To quantify this, we compare the GPU performance to the power draw for each profile. One simple metric is **frames rendered per second per watt** during the GPU tests. We calculated the average FPS across the two graphics tests divided by the average GPU power draw in those tests.

The **Custom 945 mV profile delivered the highest efficiency** by a wide margin – about _33.0 FPS per 100 W_, which is ~15% more output per watt than the stock configuration (≈28.7 FPS/100 W). The Silent mode also showed a modest efficiency improvement (≈30.0 FPS/100 W, about 4–5% better than Default). Interestingly, the **OC Mode provided the lowest efficiency** (about 28.5 FPS/100 W, essentially on par with stock). In summary, **undervolting + memory OC yielded the best performance-per-Watt** of all tested profiles, while the stock and factory OC profiles were the least efficient. Silent mode fell in between, slightly boosting efficiency by capping power and clocks. These results underscore that pushing the GPU to maximum clocks (as stock/OC do) has diminishing returns in performance for the power used, whereas undervolting can improve efficiency without sacrificing – and even slightly improving – performance.

## Time Spy Extreme Stress Test (Custom Profile)

Finally, we ran the _Time Spy Extreme Stress Test_ (20 consecutive loops) on the **Custom undervolt/OC profile** to verify its stability and thermal behavior under extended load. The stress test was passed with **98.6% frame rate stability**, well above the 97% passing threshold. This means the **performance only varied by ~1.4% between the fastest and slowest loop**, indicating very consistent output – an excellent stability result.

Over ~20 minutes of continuous load, the undervolted GPU did not exhibit any thermal buildup or throttling. GPU core temperatures leveled off in the high fifty-degree range (around 57–59 °C after warm-up) and even tended to stay a bit lower than in the shorter one-shot benchmark (peaking ~57.5 °C in the log). The core clock also remained stable: it mostly hovered ~2775–2820 MHz throughout the loops, with only minor fluctuations (at times boosting briefly up to ~2940 MHz in lighter segments, and dipping to ~2745 MHz at the lowest). These small variations are normal and likely due to the power target being hit occasionally at the very start of a loop, but importantly no progressive downclocking was observed as the card heated up. Power draw in the stress test stabilized at roughly ~460–470 W average during loops – slightly lower than the ~495 W in the initial benchmark run, which makes sense as the card found a steady thermal-state efficiency. The fact that frame rates stayed extremely consistent (98.6% stability) confirms the undervolt profile was fully stable (no performance degradation or throttling over time). There were also no signs of any erratic behavior like clock or utilization drop-outs. In short, the custom profile handled the extended 100% load test gracefully, maintaining high clocks and performance over 20 loops with temperatures in check and stable output.

## Conclusion

Comparing all four GPU profiles, we can highlight the following points:

* **Best Performance:** The Custom undervolt/overclock profile slightly edged out the others in raw performance, topping the charts in graphics test FPS and overall Time Spy Extreme score. The Stock and OC Mode profiles were effectively tied just a hair behind, while Silent mode lagged by ~2–3% due to its lower clock speeds. So, if absolute performance is the goal, the undervolt+OC profile proved capable of beating even the factory OC (thanks in part to its memory overclock and ability to avoid heavy power throttling).
* **Best Efficiency:** The undervolt profile was by far the most power-efficient. It delivered higher performance while consuming significantly less power than the default configuration. Its performance-per-watt was ~15% better than stock. Silent mode also improved efficiency a bit (about ~5% better than stock) but at the cost of some performance. The factory OC provided no efficiency gain – it drew just as much power as stock for negligible performance difference. This shows that undervolting (with a mild OC) hits a sweet spot: it reduces wasted energy (lower voltage and power) yet retains performance, even improving it slightly.
* **Thermal and Acoustic Trade-offs:** All profiles kept GPU temperatures around ~60 °C or below, thanks to a robust cooler. The Silent BIOS likely ran the fans slower, meaning it might have been quieter, but we saw no dangerous heat increase – just a minor performance loss and slightly higher average temps during the run-up. The stock and OC profiles push the card hardest, which presumably also means louder fan noise under load. The undervolted profile, using default fan policy, would have similar acoustics to stock but ran cooler on average due to lower power. In other words, the custom profile achieved both lower temps and lower noise (vs. stock) by virtue of drawing ~50–100 W less under load, all while delivering top performance.
* **Stability:** None of the profiles showed instability – there were no sudden frame rate drops or erratic clock behavior in the logs. Even the aggressive undervolt was stable through 20-loop stress testing with ~98.6% frame rate stability (a passing result), indicating no throttling or degradation over time. The lack of any thermal throttling or thermal limit flags in logs confirms the cooling was adequate for all modes. The only throttling observed was power-related: the stock, OC, and silent profiles all hit the GPU’s power limit at times (as expected given their 100% usage and high TDP), and the stock/OC also bumped against voltage/frequency limit. The undervolted profile avoided the voltage limit and was less constrained by power, which helped it sustain a consistent clock.

**Overall, the Custom 945 mV Undervolt + OC emerges as the best-balanced profile** – it managed to slightly increase performance above stock while substantially improving power efficiency, keeping the card cool and stable. The Stock and OC Mode profiles delivered the expected top performance as well, but very inefficiently, hitting ~600 W peaks for only marginal gains. The Silent mode significantly reduced power and noise, with a small performance sacrifice, making it a good option if one prioritizes acoustics/efficiency and doesn’t mind ~3% lower FPS. But if one can fine-tune the GPU, the undervolt profile clearly offers **the highest performance per watt with virtually no downsides** in our testing. It provides near-OC performance at a power cost closer to Silent mode – a win-win scenario with lower heat output and possibly more headroom for sustained workloads.