Local LLM Users Eye RTX 5070 Ti SUPER and RTX 5080 SUPER After Price Leak Challenges RTX 5090

New pricing information for NVIDIA’s upcoming RTX 50 SUPER series has surfaced, suggesting a significant shift in the value proposition for local large language model (LLM) enthusiasts. According to the leak, the new SUPER cards will launch at the same Manufacturer’s Suggested Retail Price (MSRP) as their non-SUPER predecessors. For anyone building a system for local inference, this news is important as it directly addresses the VRAM capacity limitations of the initial RTX 50 series lineup without increasing the price.

The Leaked Prices and What They Mean for LLM Hardware

The core of the leak indicates that despite offering more VRAM and, in some cases, higher memory bandwidth, the new SUPER models will not carry a price premium. This is a direct response to lower-than-expected demand for the current generation, aiming to make the new cards more appealing.

For the local LLM user, the most interesting models are the RTX 5080 Super and the RTX 5070 Ti Super, both of which are rumored to feature 24GB of VRAM. This capacity is a key target for running popular 30-billion to 40-billion parameter models or even 70B models at effective quantization levels without needing to offload to slower system RAM. The RTX 5070 Super, with its 18GB of VRAM, presents a unique option for multi-GPU setups or as a step up from 16GB cards.

Comparison Against RTX 3090, 4090 and 5090 GPUs

To understand the potential value of these new cards, we need to compare them to the current hardware options used for LLM inference. The primary benchmark for value has been the second-hand RTX 3090, which provides 24GB of VRAM for around $800. High-end options like the RTX 4090 and RTX 5090 offer more performance but at a substantially higher cost.

The leaked pricing places the new SUPER cards in a very competitive position. The RTX 5070 Ti Super, with a leaked MSRP of $749, directly targets the used RTX 3090. It would offer the same 24GB VRAM capacity with a new architecture, better power efficiency, and a full warranty for a similar price. Its memory bandwidth of 896 GB/s is slightly lower than the RTX 3090’s 936 GB/s, but the architectural improvements may balance this for many inference tasks.

The RTX 5080 Super, at a leaked price of $999, offers a clear performance step-up. With 24GB of VRAM and a higher memory bandwidth of 1024 GB/s, it surpasses the RTX 3090 and gets close to the RTX 4090’s bandwidth. For builders looking for the best single-card performance for 70B models without moving into the $2000+ price bracket, this could be the new standard.

VRAM, Bandwidth, and Price Comparison Table

Here is a direct comparison of the specifications that matter most for local LLM performance, based on current market data and the recent leaks for August 2025.

The Reality of Street Prices and Market Impact

While these leaked MSRPs are promising, we must remain practical. It is unlikely these cards will be widely available at their stated prices upon release, especially if demand from the LLM community is high. Initial availability will likely be limited, with prices inflated by retailers.

However, the introduction of these cards will inevitably pressure the second-hand market. If a new RTX 5070 Ti Super can eventually be bought for around $800, the price of a used RTX 3090 will have to decrease to remain a viable option. This could push used 3090s into the 600 – 700 range, further solidifying its position as a budget-friendly, high-VRAM solution. For the price-conscious builder, the release of the SUPER series may be a good time to watch the used market for deals.

Ultimately, this pricing leak signals a positive trend for anyone building a dedicated LLM machine. The RTX 5070 Ti Super, if it lands near its MSRP, is positioned to become the new value leader for 24GB cards. The final decision will depend on actual retailer pricing, but the landscape for affordable, high-VRAM inference hardware is about to become much more competitive.

Trending Stories