Samsung Electronics has started a new phase in the evolution of graphics memories by starting tests of its GDDR7 modules with speeds of up to 36 Gbps. Amostras of these ultra-high-performance components have already been shipped to strategic partners, including NVIDIA, fueling expectations for a significant upgrade to the highly anticipated GeForce RTX 50 SUPER lineup of graphics cards. Este move positions the South Korean company at the forefront of memory technology, promising leaps in performance for games, artificial intelligence and professional content creation.
The company confirmed that mass production of the 28 Gbps version is already underway, while the faster variants, 32 Gbps and 36 Gbps, undergo rigorous validation processes. The new generation of memory not only offers more speed, but also improved power efficiency by up to 20% compared to previous technology, a crucial factor in developing more powerful and thermally stable hardware.
This technological evolution comes at a time of high demand for bandwidth, driven by increasingly complex applications. NVIDIA already uses initial versions of GDDR7 in its professional GPUs with the Blackwell architecture, such as the RTX PRO 6000, signaling an imminent transition also to the consumer segment. The introduction of chips with higher density, 3 GB per module, is another difference that will allow for more robust VRAM configurations.
Technical details of the new memory generation
The Samsung’s GDDR7 technology represents a substantial advancement over its predecessors, maintaining physical compatibility while using the same 266-pin BGA package format, which simplifies integration into existing printed circuit board (PCB) designs. The major technical difference lies in the adoption of PAM3 signaling (Pulse-Amplitude Modulation with 3 levels), which allows the transmission of more data per clock cycle compared to the NRZ (Non-Return-to-Zero) standard used in GDDR6. Essa innovation is responsible for an increase of up to 30% in data transfer speed without a proportional increase in energy consumption. Além Furthermore, Samsung developed a molding compound with low thermal resistance, which reduces operating temperature by up to 70%, ensuring greater stability and longevity for components even under intense and continuous workloads, such as long gaming sessions or rendering complex projects. The capacity of each chip has been increased to 24Gb (3GB), a 50% increase over the 16Gb (2GB) modules, allowing GPU manufacturers to achieve greater VRAM capacities with fewer components, optimizing space and design complexity.
What changes for the design of video cards
The higher density of the new GDDR7 memory modules, now at 3 GB per chip, offers unprecedented flexibility for hardware engineers. With this change, it is possible to design video cards with greater VRAM capacity while taking up less physical space on the PCB. For example, a GPU with a 256-bit memory interface can be configured with 24 GB of VRAM using just eight modules, a configuration that would previously require more components and a more complex design. Essa optimization not only reduces manufacturing costs, but also improves signal integrity and thermal efficiency.
For the high-performance segment, the benefits are even more significant. A card with a 512-bit bus, equipped with 36 Gbps chips, could achieve a theoretical bandwidth of 1,728 GB/s, a value that redefines performance limits for professional applications and games at 8K resolutions. Essa scalability allows the same technology to be applied from mid-range consumer boards to high-end workstations, which can be configured with up to 96 GB of VRAM to handle scientific simulations and large AI model training.
The long-awaited update to the RTX 50 SUPER line
The hardware market eagerly awaits the official announcement of the RTX 50 SUPER series, which traditionally represents a mid-lifecycle upgrade for NVIDIA architectures. Especulações indicate that the launch could take place during CES 2026, with the introduction of models such as the RTX 5070 SUPER, RTX 5070 Ti SUPER and RTX 5080 SUPER.
These new GPUs should be the first to directly benefit from the Samsung’s GDDR7 memory advancements. The expectation is that NVIDIA will use the 28 Gbps modules as standard for the majority of the line, reserving the 32 Gbps and 36
The adoption of 3GB chips will allow a significant increase in VRAM capacity. A future RTX 5070 SUPER, for example, could jump to 18GB of memory, a 50% increase over the standard version’s 12GB, offering a considerable advantage for games with high-resolution textures.
However, NVIDIA remains cautious and has not yet passed on the final specifications to its manufacturing partners, such as Gigabyte and MSI. Essa stance reflects the complexity of validating new components and the need to ensure stability and compatibility across the production chain before a full-scale launch.
Improved gaming and 4K performance
For gamers, the arrival of high-speed GDDR7 memory translates into a more fluid and visually rich gaming experience. Higher bandwidth is critical to eliminating memory bottlenecks, especially at high resolutions like 4K and 8K where the amount of texture and geometry data processed by the GPU is massive.
With a hypothetical RTX 5080 SUPER equipped with 32Gbps GDDR7 on a 256-bit interface, total bandwidth would reach 1,024 GB/s. Esse increase, even if it seems modest in percentage, makes a notable difference in scenarios that use intensive ray tracing, where every fraction of a second in communication between the GPU and VRAM counts.
The increased memory capacity also sets the stage for the next generation of games. Títulos Future AAAs will require more VRAM to store high-quality assets, and cards with 18 GB or 24 GB will become the new standard for those looking for the best visual experience without compromising frames per second.
Competitive landscape and production challenges
Samsung is not alone in the race for leadership in the graphics memory market. Concorrentes and SK Hynix have already announced plans to develop GDDR7 modules that can reach speeds of up to 48 Gbps, promising to further intensify technological competition. Micron, in turn, continues to invest in intermediate solutions such as GDDR6X, but the market shows a clear migration trend to the new standard due to its efficiency and compatibility with new GPU architectures.
One of the biggest challenges to the mass adoption of GDDR7 is the current semiconductor supply chain landscape. The explosion in demand for AI accelerators has put severe pressure on the production of high-performance DRAM memories, driving up prices and limiting availability. Essa situation may influence the launch schedule of NVIDIA and other manufacturers, which depend on a stable supply and competitive prices to make their new product lines viable.
Applications beyond games
Although the gaming market is one of the main drivers of innovation in GPUs, 36Gbps GDDR7 memories have transformative potential in several other areas. In the field of artificial intelligence, high bandwidth is essential to accelerate the training of neural networks and the processing of large volumes of data in real time, enabling faster and more accurate inference applications.
Content creation professionals will also benefit greatly. The ability to handle multiple video streams at 8K resolution without latency and perform complex 3D renderings in less time represents a significant productivity gain. Improved energy efficiency also makes this technology viable for portable workstations and high-performance laptops, which will be able to offer processing power previously restricted to desktops.
Versatility for workstations
The scalability of GDDR7 memory allows for extremely high VRAM configurations, ideal for the workstation segment. The ability to equip a single GPU with up to 96 GB of ultra-fast memory opens up new possibilities for scientific simulations, complex data analysis and engineering projects that previously relied on multiple systems or cloud computing solutions, making on-premises hardware even more powerful and versatile.