Galax RTX 5070 Ti HoF (Hall of Fame) review – To the top with a monster board?

1 - Introduction, overview and technical specifications 2 - Test system and equipment 3 - Teardown: PCB and components 4 - Teardown: Cooling system 5 - Teardown: Material analysis and TIM 6 - Benchmarks: gaming performance 7 - Power consumption, transients, PSU recommendation 8 - Clock rates and overclocking 9 - Temperatures and thermal imaging 10 - Fan curves and noise with audio samples 11 - Summary and conclusion

Summary and overview of all games

Calculating an average value from the FPS results of several games is an obvious approach to gain a rough overview of a system’s gaming performance. To do this, the measured frame rates of all tested titles are added together and then divided by the number of games. This produces a figure that appears to provide a generally valid statement on performance. In practice, however, this value can only be used to a limited extent. The individual games differ massively in their hardware requirements. Some titles primarily use the graphics card, while others place a greater load on the CPU or react sensitively to memory bandwidth and latencies. The fluctuations in the measured values are correspondingly large. Demanding AAA titles can pull the calculated average down significantly, while older or less demanding games distort it upwards. In addition, the weighting of all games is the same on average, regardless of whether it is a graphically complex reference game or a comparatively light title. For a meaningful assessment of gaming performance, it is therefore better to present the results in a differentiated manner, i.e. per game and, if possible, in different resolutions and quality levels. A simple average value can serve as a guide, but is no substitute for a detailed analysis of the individual results.

The normalization of FPS values is a much more precise method of comparing the actual hardware performance. Since every game has different priorities – sometimes GPU-limited, sometimes CPU-limited, sometimes heavily dependent on memory throughput or API efficiency – raw frame rates are only comparable to a limited extent. By converting to a common scale, these differences are smoothed out so that individual extreme values do not distort the overall picture.

The advantage is that this results in a balanced performance display. Particularly demanding titles no longer push the average down disproportionately, while easier games do not artificially push the result up. Instead, the result is a homogeneous profile that objectively describes the relative performance of a graphics card or system over the entire test course. You apply this method consistently to both the average FPS and the P1 Low. The latter is of particular value because it records the weakest one percent of frame times and thus provides information about micro-stutters and minimum frame rates. Especially in practical gaming scenarios, this parameter is almost more important than the pure average frame rate, as it directly reflects the perceived fluidity of the gameplay.

In combination, normalized average values and normalized P1 Lows enable a very robust, comparable assessment of gaming performance that goes beyond the mere addition of raw data and highlights the differences between different hardware classes more clearly. The term P1 Low describes a benchmark value that specifically depicts the weakest one percent of the recorded frame rates. While the average value only provides an average performance figure, the focus here is on possible downward outliers, which are often more noticeable in practice than a slightly low average value.

To determine this, all FPS values measured during the test are sorted in order. The lowest percentage is regarded as the critical range and the highest value within this range is defined as P1 Low. This value thus marks the threshold below which only one percent of all frames lie. P1 Low is particularly suitable for making short but disruptive frame rate dips visible. While a high average can still be convincing even with occasional frame drops, the P1 Low shows whether the game is actually running stably and smoothly or whether the image output is noticeably stuttering. This measured value therefore complements the classic average to provide a much more meaningful assessment of the real gaming experience.

Choosing Full HD as the test resolution is a tried and tested approach for looking at the performance of a CPU in isolation. As the graphics card is only put under limited strain at this comparatively low resolution, the bottleneck is shifted more towards the processor. This makes it much easier to identify bottlenecks that would be lost in the GPU limit at higher resolutions. This is particularly revealing for the Ryzen 7 9800X3D with its large additional cache, as it shows in which games the architecture actually has advantages and where the limitations lie in the front end or in the memory connection.

In addition, the measurements of the power consumption of the CPU and GPU provide important information on which component dominates in the respective scenarios. If the CPU is working under full load while the GPU remains largely idle, this clearly indicates a CPU limit. Conversely, high GPU power consumption at low CPU load indicates that the game or scene is primarily graphically bound. In combination, this creates an overall picture that makes both the efficiency and the balance between processor and graphics performance transparent and allows a realistic classification of the system’s performance.

The CPU load also changes noticeably with increasing resolution. While it is still clearly required in Full HD because the graphics card is not fully utilized and the frame rate is heavily dependent on the processor performance, its power consumption gradually decreases at higher resolutions. The reason for this is that the GPU takes over the main part of the work as the number of pixels increases and the CPU only supplies the data stream without becoming a limiting factor itself. This effect can be clearly seen in the measurements: At 1440p and especially in 4K, the CPU power consumption drops, while the GPU works continuously towards full load. This shifts the energy profile of the system, which in practice means that processors such as the Ryzen 7 9800X3D primarily show their strengths in low resolutions, but hardly play a role in the actual performance in higher resolutions.

In principle, efficiency is calculated by reversing the power consumption. To do this, the average power consumption, measured over the entire benchmark run of a game, is compared with the FPS determined. The result is a value that expresses how many frames per second are achieved per watt used. In this way, not only the pure speed can be compared, but also how effectively the energy used is utilized.