Over a few months, two nearly identical Gigabyte RX 9070 XT Gaming OC cards passed through my hands. They use GDDR6 chips from different manufacturers. The older card had Hynix; the newer uses Samsung GDDR6, known for lower power draw and temperatures. We’ll examine how this change affects performance, memory temperatures and power, GPU temperatures and clock speeds, and overall card power consumption in our tests.
Cyberpunk 2077, BIOS Performance
The first test of operating characteristics was done in the usual way with the Cyberpunk 2077 benchmark, using the RT Medium preset at a resolution of 3840 × 2160. It consists of eight runs back-to-back, with average values taken from the last run.
You should see the results in the interactive graphs that change when you hover over them with the cursor. If they don’t work, the same graphs are also available as expandable images.
In the first graph you can see a slight change in performance, but it’s not possible to say with certainty how much of it is measurement error and how much comes from different memory or slightly different GPU clocks.
The most obvious difference is in memory temperatures. In the graphs this is shown in orange. This matches the experience of other users, but—we still only have two samples to compare. It should represent hotspot memory temperature, but it’s impossible to say exactly how much of the difference is due to chip variance, cooler variance, or the contact quality between cooler and memory, which can also differ between cards. Judging by GPU temperatures, the coolers of both cards themselves don’t seem significantly different. The GPU hotspot temperature on the Samsung-equipped card is higher, but this is also influenced higher power draw of the GPU.
On the newer card, the fans ran at higher RPM, likely due to the higher GPU hotspot temperature.
From the detailed monitoring data you can also read the voltage and current of the GPU and memory power supply, from which the monitoring then calculates the power draw. Samsung memory consumes a few watts less. But whatever is saved on the memory, the controller redirects to the GPU, which is constrained by its power limit.
This can be seen both in the GPU power draw from monitoring and in the overall system consumption, which despite the more efficient memory remains the same.
⠀
