Modifiers upgraded the GTX 1650 video memory to 8GB and nearly doubled the running score by simply changing the core.

A hardware modification player recently successfully upgraded an NVIDIA GeForce GTX 1650 originally equipped with 4GB GDDR6 video memory to 8GB by replacing the video memory chip, and achieved nearly twice the performance improvement in synthetic benchmark tests, attracting attention. It should be emphasized that this modification is only applicable to a specific version of GTX 1650 based on the TU106 core and using 1GB GDDR6 video memory particles. Not all GTX 1650 graphics cards can be copied.

In the graphics card modification circle, it is not uncommon to improve performance by modifying shunt resistors (shunt) or replacing/re-soldering video memory modules, but this usually requires firmware modification, and whether it can be properly recognized by the GPU also depends on the support of the core itself. In comparison, this transformation is described as "simple" because the GDDR6 memory particles themselves come in both 1GB and 2GB specifications. For products that originally used 1GB particles, the capacity can be doubled by directly replacing them with larger-capacity particles of the same frequency.

The report pointed out that there are multiple versions of GeForce GTX 1650 on the market, including designs based on different GPU cores such as TU117, TU116 and TU106. Only a few models using the TU106 core and paired with 1GB GDDR6 memory particles have the hardware foundation for "painless replacement" with higher-capacity particles. TU106 itself has been used in mid-range RTX 20 series graphics cards, and on this GTX 1650, it also allows manufacturers to use four 1GB GDDR6 graphics memory to form a 4GB capacity configuration.

This transformation comes from YouTube creator Paulo Gomes, who shows in the video how to replace the original 4 1GB GDDR6 video memory on this TU106-based GTX 1650 with 4 2GB GDDR6 video memory particles from the Samsung HC16 series, increasing the video memory capacity of the graphics card from 4GB to 8GB. What’s more noteworthy is that the modification process did not involve VBIOS or other firmware-level modifications, but he encountered two faulty video memory particles during actual operation and had to check and replace them first. This also suggests that similar welding operations require higher manual skills and equipment.

After completing the graphics memory upgrade, Paulo Gomes conducted a performance test using Unigine Superposition to compare the score changes of the same graphics card in this benchmark before and after the modification. Tests show that the original GTX 1650 with 4GB of video memory scored 624 points in Superposition. After upgrading to 8GB of video memory, the score jumped to 1,245 points, nearly doubling the score and achieving a performance improvement of about 200%.

However, the report also pointed out that inferring "performance doubling" from benchmark test results alone is not completely consistent with the actual gaming experience, because simply replacing the video memory particles will not directly increase the core computing power, and it is difficult to obtain the same gain in all game scenarios. The author emphasizes that it is a pity that the video does not show the actual frame rate test of the game, otherwise it would be more reference value for ordinary players to evaluate the significance of such modifications in real games.

Nonetheless, against the background that many new AAA games have increasing requirements for video memory capacity, the increase in video memory from 4GB to 8GB is expected to alleviate problems such as high-resolution material loading, frame rate jitters or freezes to a certain extent. In particular, perceivable improvements are more likely to occur in scenes originally restricted by video memory bottlenecks. The report believes that for players who own a GTX 1650 corresponding to the TU106 version and have related welding equipment and experience, this type of video memory upgrade provides a low-cost technical path to extend the life of the old card. However, for most ordinary users, this high-risk and high-threshold operation is still not widely feasible.