Samsung advances testing of 36 Gbps GDDR7 memories for NVIDIA RTX 50 SUPER GPUs

Samsung Electronics announced the start of mass production of GDDR7 memories with a speed of 28 Gbps and a capacity of 3 GB per module. The South Korean company also took initial steps in testing faster variants, 32 Gbps and 36 Gbps, sending samples to hardware manufacturers. Essas news, revealed in updates on the semiconductor division’s official website, comes amid speculation about the impact on NVIDIA graphics cards, especially the GeForce RTX 50 SUPER line. The development aims to meet growing demands for greater bandwidth in gaming and artificial intelligence applications.
The Samsung initiative comes at a time of high demand for graphics memory components. GDDR7 modules represent an evolution over previous generations, with improved power efficiency by up to 20% and PAM3 signaling for faster transfers. Fabricantes like NVIDIA already incorporate initial versions in professional products, such as the RTX PRO 6000 Blackwell, which uses 3 GB chips for configurations of up to 96 GB of VRAM.
- Main speeds tested: 28 Gbps (mass production), 32 Gbps and 36 Gbps (sampling).
- Capacity per chip: 24 Gb, equivalent to 3 GB.
- Initial applications: High-performance GPUs and workstations.
These advances could change the release schedule in the graphics hardware sector.
Evolution of technical specifications
Samsung’s new GDDR7 modules maintain the 266 FBGA form factor, facilitating integration into existing graphics card designs. The 28Gbps speed is now available in volume, enabling GPUs to achieve bandwidths in excess of 1TB/s on 384-bit interfaces.
The transition to 3GB chips per module represents a 50% increase in density compared to the 2GB chips used in the first RTX 50. Essa change optimizes PCB space, reducing the number of components needed to achieve high VRAM capacities.
Supply chain challenges
Mass production coincides with a global DRAM crisis, driven by exploding demand in AI accelerators. High Preços may delay line updates, such as the RTX 50 SUPER, scheduled for the first or second quarter of 2026.

GPU manufacturers face constraints on final specifications, with NVIDIA not yet sharing full details with partners like Gigabyte and MSI. Essa caution reflects dependence on component validations such as GDDR7 modules.
The Samsung, however, keeps pace, with samples of 32 Gbps and 36 Gbps already distributed. Esses preliminary testing aims to ensure compatibility in high-load scenarios, such as 3D rendering and machine learning model training.
Implications for gaming performance
A hypothetical RTX 5080 SUPER with a 256-bit interface and 32Gbps GDDR7 would increase bandwidth from 960 GB/s to 1,024 GB/s, a 6.7% gain. Essa improvement would benefit 4K resolutions with ray tracing enabled, reducing memory bottlenecks.
Models like the RTX 5070 SUPER could jump to 18 GB of VRAM, compared to 12 GB for the standard version, supporting denser textures in AAA games. The adoption of 36
Rumors about the RTX 50 SUPER line
Speculation points to three main SKUs in the SUPER update: RTX 5070 SUPER, RTX 5070 Ti SUPER and RTX 5080 SUPER. Cada would bring increases of up to 50% in memory capacity, prioritizing efficiency over higher clocks.
NVIDIA plans announcements possibly at CES 2026, aligning the refresh with the mid-life cycle of the Blackwell series. Parceiros are awaiting confirmation of speeds, with indications that 28 Gbps modules will be the initial standard, reserving 36 Gbps modules for professional editions.
Advantages of PAM3 technology
PAM3 signaling on GDDR7 chips allows transfers 30% faster than previous generation NRZ, without increasing power consumption. Essa efficiency reduces heat generated, with molding compounds that decrease thermal resistance by 70%.
In internal tests, the 36 Gbps modules demonstrate stability under continuous loads, ideal for prolonged gaming or video editing sessions. The Samsung highlights applications beyond GPUs, such as next-generation consoles and automotive systems.
Optimized VRAM Settings
GPUs with 256-bit interfaces can reach 24GB using eight 3GB modules while maintaining compact designs. Para 512 bits, 32 chips chegam a 96 GB, configurável para workstations como a RTX PRO 5000, atualizada para 72 GB de GDDR7 ECC.
This scalability supports varied workloads, from AI inference to scientific simulations. Higher density minimizes interference, improving reliability in multi-monitor environments.
Samsung positions GDDR7 as a versatile solution for the high-performance ecosystem, with scalable production for future demands.
Competition in the memory market
Rivals like SK Essa dispute accelerates innovations, benefiting hardware manufacturers with diversified speed and cost options.
Micron, in turn, focuses on GDDR6X for intermediate transitions, but GDDR7 gains traction due to its compatibility with Blackwell architectures. Samsung leads in sampling, with shipments to potential customers since November 2025.
The market reacts with optimism, predicting that these speeds will balance the growth in resolutions and graphical complexity.
Applications in AI and content creation
The 36 Gbps modules support real-time AI acceleration, with bandwidth to process large datasets without latency. In 8K video editing, the 3GB capacity per chip makes it easier to handle multiple frames.
Software developers are already testing integrations, exploring gains in neural upscaling. Energy efficiency attracts portable designs like RTX 50 series laptop GPUs.






