NVIDIA GeForce RTX 4070 Ti SUPER Gets Memory Tuned To 26 Gbps Speeds, Ends Up Faster Than 4080 SUPER GPU

The folks over at TecLab & Paulo Gomes have pushed the memory on the NVIDIA GeForce RTX 4070 SUPER GPU to 26 Gbps which ends up delivering faster performance than the RTX 4080 SUPER.

The NVIDIA GeForce RTX 4070 Ti GPU saw a memory upgrade when it launched, replacing the 12 GB (192-bit) memory interface with a 16 GB (256-bit) interface using the same 21 Gbps GDDR6X memory modules. These memory modules are known to feature some strong OC capabilities and we have seen in our testing that overclocking them to 24-25 Gbps is fairly easy when the right OC conditions are used.

The TecLab and Paulo Gomes teams put this memory on the NVIDIA GeForce RTX 4070 Ti SUPER GPU to the test. Our friend, Ronaldo from GALAX who is also the chief of the TecLab channel along with his team were the masterminds behind the 26 Gbps overclock.

For the competition, the TecLab team used a GALAX GeForce RTX 4070 Ti SUPER graphics card while the Paulo Gomes team used the Manli GeForce RTX 4070 Ti SUPER graphics card. Both of these cards were swapped out in regards to their memory modules. As we mentioned above, the card features 21 Gbps Micron dies by default but it was decided that to reach the maximum OC potential, the 21 Gbps dies (161-0479-900) had to be swapped with 24 Gbps dies (161-0480-900).

It should be pointed out that swapping memory modules like this isn't something anyone can do at home. It requires soldering expertise and you need to have the right tools and information in regards to the PCB and BIOS for everything to work out. Both overclocking channels had this info so it was a process that they are well aware of but for a PC enthusiast, this whole process is very difficult and can easily void your warranty.

No fancy LN2 cooling was used and the cards were overclocked under their stock custom cooling solutions which include triple-fans and a 2.5-3.0 slot design. The fact that these GPUs ran well within their thermal limits with such extreme overclocks shows just how over-engineered some of these coolers are.

During the OC session, both the GPU and the memory dies were overclocked and the results are going to speak for themselves. The Paulo Gomes team raised the GPU clock to +200 MHz and the memory clock to +2000 MHz which resulted in a score of 8870 points while the TecLab (GALAX) team pushed the GALAX GeForce RTX 4070 Ti SUPER GPU even further with a +271 MHz GPU & a +2850 MHz memory clock which resulted in a score of 9133 points. That's a 2% lead over the GPU score from Paulo Gomes in The Unigine Superposition benchmark.

Videocardz managed to compile the benchmark results as listed below:

This is around 20-30% performance uplift simply by adding faster memory to the GPUs & upping the clock speeds. The next-generation GDDR7 memory standard is going to begin at 28 Gbps die speeds which are a 33% bump over the 21 Gbps GDDR6X modules so that along with added GPU architectural improvements and clock speed bumps with a refined process node can add to some stunning gains on the GeForce RTX 50 "Blackwell" GPU lineup which is expected to debut later this year in the form of the RTX 5090 and RTX 5080.

Overall

News Source: VideoCardz