Okay, this convention floor is just too loud. I’m trying to get this out, but the noise is… it’s a lot. Right, NVIDIA. The RTX 5070 Ti SUPER.

So, the big story, the big rumor, is this card is coming, right? It’s a mid-cycle Blackwell refresh, that’s the word. And it’s going to be interesting because of the memory. The core thing here is the memory configuration. The RTX 5070 Ti SUPER is expected to jump from 16 GB to 24 GB of GDDR7 memory.

That’s a 50% increase in VRAM capacity over the standard RTX 5070 Ti. That’s a big number, a really big number for VRAM. But the speed, the actual memory speed, is rumored to stay at 28 Gbps GDDR7, with the same 256-bit memory bus. So, bandwidth is probably staying around 896 GB/s, same as the non-SUPER version.

It’s about capacity, not raw bandwidth, mostly. The Blackwell architecture itself, it’s already out there in some forms. NVIDIA officially announced Blackwell at GTC 2024, back on March 18, 2024. They focused on data center accelerators, the B100 and B200, and the Grace Blackwell platform. Jensen Huang, the CEO, said it’s a processor for the generative AI era.

Gaming wasn’t really mentioned then. But for consumer cards, the RTX 50 series, they debuted with the RTX 5070, RTX 5080, and RTX 5090 in January 2025. These cards use TSMC’s 4N process node. Blackwell brings fourth-generation RT cores for ray tracing and fifth-generation Tensor Cores for AI compute.

The Tensor Cores support FP4 precision, which is faster for AI workloads, like 3x higher performance compared to the previous generation. Power consumption is another point. The RTX 5070 Ti SUPER is listed at 350W TBP, that’s a 50W increase, 17% more than the RTX 5070 Ti’s 300W. This extra power budget could mean higher core clocks, but no concrete clock speed info yet. The 12VHPWR connector is likely for this kind of power level. Performance expectations?

Not a huge generational leap in raw gaming performance from the RTX 5070 Ti SUPER. If the specs are right, it uses the same GB203 die configuration as the RTX 5070 Ti. So, modest uplift in general gaming. But, that 24 GB VRAM will make a difference in memory-limited scenarios.

Some leaks suggest the RTX 5070 Super, not the Ti SUPER, would be the only model getting more CUDA cores, swapping to a full-fat GB205 GPU. But for the Ti SUPER, it’s mostly about the memory. The GPU market is interesting right now. In Q1 2026, about 11.8 million GPUs shipped. NVIDIA still has a huge share, around 90% in Q1 2026 for desktop GPUs. AMD was at 8%, Intel at 1%.

This is according to Jon Peddie Research. The overall graphics card market size was $23.57 billion in 2025 and is projected to grow to $27.64 billion in 2026. It’s expected to hit $97.4 billion by 2034, a 17.0% CAGR. AI computing demand is a big driver, also gaming ecosystem expansion. NVIDIA’s financial results, they are just massive.

For Q1 fiscal 2026, revenue was $81.6 billion, up 85% from a year ago. Data Center revenue was $75.2 billion, up 92% year-over-year. This AI boom is just insane. NVIDIA announced the BioNeMo Agent Toolkit on June 23, 2026, that’s for life sciences, accelerating scientific discovery.

It’s all about agents now. Pricing for the RTX 5070 Ti SUPER, no official MSRP yet. The standard RTX 5070 Ti launched at $749. With 24 GB GDDR7 and current DRAM prices, it’s hard to imagine the SUPER version staying at $749. It’ll probably be higher.

Some earlier rumors for the RTX 5070 Super suggested it might launch at $549, matching the non-Super, but that was before the GDDR7 price increases really hit. My own trade, I bought NVDA at $176.08 on December 15, 2025. I’m holding that until it hits $250 or if the data center growth shows any sign of slowing down, which, let’s be real, doesn’t seem likely anytime soon, does it? The stock closed at $200.04 on June 23, 2026, the day the BioNeMo Agent Toolkit was announced. It’s been a good ride. The RTX 50 SUPER lineup, including the 5070 Ti SUPER, is rumored for Q4 2026 or Q1 2027.

This is a mid-generation refresh, focused on memory upgrades and modest clock speed increases, not huge architectural shifts. NVIDIA is replacing 2GB GDDR7 chips with 3GB chips, giving most Super models 50% more VRAM. This is really important for lower-end cards where memory can be a bottleneck in games. The Blackwell architecture is designed for neural networks and AI workloads. It has an AI Management Processor, a dedicated scheduler chip on the GPU.

This offloads scheduling from the CPU, helping the GPU manage its own resources better. This is all part of the push for AI everywhere, not just data centers. The consumer cards benefit from this too, like with DLSS 4, which uses AI models retrained on Blackwell-based data center clusters. This improves temporal stability and motion clarity.

Ray tracing is also getting better, with fourth-generation RT Cores delivering roughly twice the ray-triangle intersection throughput of Ada Lovelace. So, full ray tracing at 1080p and 1440p is becoming more practical on laptops and desktops.