Manufacturers ship RX 9070 XT cards with GDDR6 memory from either Hynix or Samsung. Clocks are the same, yet performance and thermals can differ slightly—and the same goes for overclocking. Samsung chips tend to reach higher memory clocks, which suggests better performance, but it isn’t that simple. We will compare the practical differences between both variants using the same card: Gigabyte’s Radeon RX 9070 XT Gaming OC.
The Radeon RX 9070 XT is the most powerful model in the current generation of cards based on the RDNA4 architecture. It uses the Navi48 chip as its foundation. The RX 9070 XT features a fully unlocked chip with 64 CUs, 4096 stream processors, 128 raster operators, and 256 texture units. Below is a reminder of the detailed specifications:
A slightly different silicon lottery
Graphics card manufacturers equip Radeons with GDDR6 chips from either Hynix or Samsung. Even when running at the same clocks, their other characteristics can differ. Chips from Samsung likely have slightly lower power consumption and are also somewhat better at dissipating heat from within the chip. This has led to the prevailing notion that Radeons with Samsung memory have lower power consumption, but it cannot be stated as a universal truth.
We dedicated our first article to a detailed comparison of the power consumption and other operating characteristics of cards with Hynix versus Samsung chips. We tested this on the same Radeon model from Gigabyte—the Radeon RX 9070 XT Gaming OC—comparing the properties of a card with Hynix chips that I tested at launch with a later-released unit featuring Samsung chips.
However, the two tests were several months apart, and although I used the same drivers for both units, performance differed due to other updates (the BIOS and AGESA for the processor, the operating system, and the actual applications used for testing were updated in the interim).
Therefore, for a retest, I sourced an older unit with Hynix chips and compared the performance of both cards under identical conditions. The card with Hynix chips, despite having identical memory clocks and similar GPU clocks, proved to be slightly more performant.
In the initial test comparing operating characteristics, not only the memory temperatures differed, but also the graphics chip temperatures. A significant difference was observed in the GPU hotspot temperature. Due to higher temperatures, the automatic fan control on the card with Samsung chips set higher fan speeds, which further amplified the difference in measured memory temperatures. Therefore, in the third test, I used another unit with Hynix chips and remeasured memory temperatures at identical fan speed settings. And to reduce the likelihood that the lower temperatures were merely a result of monitoring calibration variances, I also measured the temperatures on the chip packages for both cards using external temperature sensors.
The final aspect often discussed regarding the differences between Samsung and Hynix memory is their overclocking potential. And that is what we will focus on today—how the same card model with chips from different manufacturers differs in this regard.
Just as in the previous test—we’re using the black RX 9070 XT Gaming OC with Samsung memory and the white RX 9070 XT Gaming OC Ice, which we reviewed a few days ago.
The Radeon RX 9070 XT Gaming OC 16G and the Radeon RX 9070 XT Gaming OC 16G Ice have their boost clock increased from the reference 2970 MHz to up to 3060 MHz. The stated typical game clock is raised from 2400 MHz to up to 2520 MHz. In practice, the operating clocks are usually higher. Details can be found in the chapters with tests of operating characteristics, which also include graphs showing clock behavior.
It is equipped with a dual BIOS. The Performance mode is active by default, in which the card has a more aggressive fan control curve. Clock speeds and power limit settings do not differ between the two BIOS versions, and furthermore, since we will be using manual fan speed settings during testing (to achieve lower chip temperatures), I will stick with the Silent BIOS as usual.
For completeness, here are the detailed information from the GPU-Z diagnostics. The value from the Silent BIOS will suffice; you can find complete info from both BIOS versions in the earlier review. First, a set of images for the card with Hynix chips, followed by the card with Samsung chips.
The HWiNFO application can retrieve the most information from the monitoring systems of most cards. The list of sensors available for the tested card in the latest version of the application is shown in the image below. On the left is the Performance BIOS, on the right is the Silent BIOS; both show the power limit set at 330 W. This is an increase compared to the reference value of 304 W.
And here are some pictures of the Samsung chip cards:
The test bench in the next chapter is the same as usual, and after that we’ll jump straight into the results.
