RTX 5070 Ti: Memory Lock Bypassed, GDDR7 Exceeds 36 Gbps

Enthusiasts have pushed the memory clock speed of an NVIDIA GeForce RTX 5070 Ti graphics card well beyond its official specification of 28 Gbps with GDDR7 memory, achieving over 36 Gbps. Despite this significant overclock, standard monitoring tools continue to report the original clock speed. The Brazilian overclocking team, Teclab, accomplished this feat by modifying the clock reference used by overclocking software, effectively bypassing a software lock imposed on the RTX 50 series and revealing substantial headroom in the GDDR7 memory subsystem of NVIDIA’s Blackwell architecture.

Teclab Circumvents GDDR7 Memory Limit

Teclab asserts they’ve bypassed the software offset limits on GeForce RTX 50 memory using a method that operates below conventional overclocking layers. Working with a GALAX GeForce RTX 5070 Ti 1-Click OC (16 GB GDDR7, 256-bit bus) with a stock speed of 28 Gbps, the team achieved an effective speed exceeding 36 Gbps although monitoring software still displayed the default value. This suggests the card is operating at a significantly higher memory clock without the operating system or drivers recognizing the change.

The typical software lock of +3000 MHz on the memory offset was bypassed by altering the clock management reference. The GPU effectively “thinks” it’s running at a lower clock speed while the memory is pushed significantly higher. The power limit of the test card was fully unlocked, without shunt modification, explaining the reported 0W power draw during the run. Shunt mods are physical alterations to the card’s power delivery system to allow for higher current flow.

In Unigine Superposition, performance increased from 9,922 points (stock frequencies, unlocked power limit) to 11,722 points with a +500 MHz core overclock and +3000 MHz memory overclock. With the bypass active, the card reached 11,993 points, with Teclab reporting an effective memory clock exceeding 18,000 MHz – over 36 Gbps – despite the monitoring software still displaying 28 Gbps.

Prior Work and Current Limitations

Teclab previously modified NVIDIA memory systems, including pushing a GeForce RTX 4070 Ti SUPER to 26 Gbps with GDDR6X in 2024. They also explored using “donor” GDDR7 modules with the RTX 50 series. A separate project involved rebuilding an RTX 5070 Ti on older PCBs, continuing their Blackwell experimentation. This earlier work demonstrates a pattern of pushing NVIDIA hardware beyond its intended limits.

This manipulation remains confined to a laboratory setting. Replicating it with consumer software like MSI Afterburner or a simple BIOS flash is not currently possible. Teclab plans to apply the same approach to a higher-binned GALAX RTX 5070 Ti HOF (Hall of Fame) card to assess scalability with better silicon and increased thermal headroom. The HOF series represents NVIDIA’s highest-quality, factory-overclocked cards.

If the method proves consistent on a HOF card, it will provide concrete insight into the headroom available in GDDR7 memory on the Blackwell architecture when the software lock is removed. This would also raise questions about how offsets are managed by overclocking tools and drivers, and the potential of high-conclude custom models with extreme cooling.

Source: VideoCardz

Further investigation will be needed to determine the long-term stability and reliability of running GDDR7 memory at these elevated speeds. The increased power draw and heat generation also present challenges for cooling solutions. The team’s next steps involve testing with a more robust card and exploring the limits of thermal management.