Recent industry leaks suggest that NVIDIA is poised to launch two new graphics cards: the RTX 5070 SUPER and the RTX 5070 Ti SUPER. Partial specifications have already been obtained.
The RTX 5070 SUPER will feature the GB205-400-A1 GPU chip with 6400 CUDA cores, a 4.1% increase over the standard RTX 5070’s 6144 cores. The streaming multiprocessor (SM) units will increase from 48 to 50. The total board power is 275W, 10% higher than the standard 250W. Its most significant highlight is the VRAM upgrade: it will come with 18GB of GDDR7 VRAM, a 50% increase compared to the RTX 5070’s 12GB. The memory bus width remains at 192-bit, with a speed of 28 Gbps, achieving a bandwidth of 672 GB/s, consistent with the non-SUPER version. The 18GB VRAM breaks the traditional 16GB limit for mid-to-high-end graphics cards, making it suitable for demanding tasks like 4K gaming, video editing, and 3D rendering. Thanks to GDDR7 VRAM’s high transfer efficiency, the RTX 5070 SUPER is expected to perform excellently in high-frame-rate and low-latency scenarios. It is anticipated to be priced similarly to the standard version, offering outstanding value for money.
The RTX 5070 Ti SUPER is positioned higher, utilizing the GB203-350-A1 GPU chip. It boasts 8960 CUDA cores, the same as the standard RTX 5070 Ti, but with an increased clock frequency. Its power consumption is 350W, a 16% increase from the standard 300W. VRAM capacity is boosted from 16GB to 24GB, a 50% increase, paired with a 256-bit bus and 28 Gbps GDDR7 VRAM, delivering a bandwidth of 896 GB/s. The 24GB VRAM is comparable to the RTX 5080 SUPER, supporting 4K gaming, AI training, and professional rendering tasks such as real-time ray tracing and complex model processing. The higher clock frequency further enhances computing performance, meeting the demands of ultimate visual quality and professional needs. Its price is expected to be on par with the non-SUPER version.
Blackwell Architecture and Key Features The RTX 50 SUPER series is based on the Blackwell architecture, bringing optimizations in ray tracing, AI acceleration, and power efficiency compared to the previous Ada Lovelace architecture. Blackwell introduces efficient SM units that support FP8 precision computation, improving AI and deep learning performance. Its ray tracing cores and Tensor Cores are significantly optimized, and paired with DLSS 4 technology, they will deliver stronger performance in real-time ray tracing and high-resolution games.
GDDR7 VRAM is another highlight. Compared to GDDR6X, it offers higher bandwidth and lower power consumption, supporting speeds from 28 Gbps to 32 Gbps. GDDR7 utilizes a 3GB module design, allowing for large VRAM configurations to meet the demands of resolutions above 4K and complex computations. Industry data indicates that GDDR7’s PAM3 signaling modulation technology improves transfer efficiency by approximately 25%, supporting high frame rates and real-time rendering.
Market Positioning and Outlook Currently, the RTX 50 SUPER series includes the RTX 5080 SUPER, RTX 5070 Ti SUPER, and RTX 5070 SUPER, targeting the high-end, enthusiast-entry, and mid-to-high-end markets, respectively. The RTX 5080 SUPER features 24GB of GDDR7 VRAM, 10752 cores, a memory speed of 32 Gbps, a bandwidth of 1024 GB/s, and power consumption exceeding 400W, aimed at top-tier gaming and creative needs. The SUPER series responds to market demand for large VRAM capacities through memory and performance enhancements. In recent years, 4K and 8K gaming have increased VRAM usage, with mainstream AAA games requiring over 12GB of VRAM at high settings, and professional software like Blender and Adobe Premiere Pro demanding even more. The 18GB and 24GB VRAM configurations fill this gap, enhancing competitiveness.
NVIDIA maintains its lead in the GPU market, with the global GPU market size estimated to reach $50 billion in 2025, driven primarily by gaming and AI. DLSS technology, which boosts performance through AI super-resolution, has become a core advantage for NVIDIA graphics cards.
The RTX 50 SUPER series is expected to debut at CES in January 2026. The initial batch may include the RTX 5070 SUPER and RTX 5080 SUPER, with the RTX 5070 Ti SUPER potentially launching later. NVIDIA aims to solidify its market position with the SUPER series, paving the way for next-generation flagships like the RTX 5090 SUPER. The RTX 5070 SUPER and RTX 5070 Ti SUPER, with their balance of performance and price, are set to meet the demands of both gamers and professional users. The 18GB and 24GB VRAM not only address current applications but also provide headroom for high-load tasks over the next 3 to 5 years, making them highly anticipated hardware upgrades.