Nvidia GeForce RTX 5090 SE Specs Leak Points to 32GB VRAM and $1,500 Price Tag

A leaked specification sheet from Russian outlet GameGPU has ignited the PC hardware community, pointing to an unannounced heavyweight contender in Nvidia's Blackwell architecture lineup: the GeForce RTX 5090 SE.[1][2][4]

According to the circulated data, the RTX 5090 SE will utilize a heavily cut-down version of the flagship GB202 silicon. By enabling 110 of the die's 192 streaming multiprocessors, the card would feature 14,080 CUDA cores. This configuration is designed to bridge the massive performance gulf that currently exists in the 50-series stack, offering roughly 31 percent more compute power than the RTX 5080, while remaining safely behind the flagship RTX 5090's 21,760 cores.[1][3][4]

The leaked specifications also point to a 500-watt total graphics power (TGP) rating, stepping down from the 575-watt draw of the full-fat 5090 but still requiring a robust 1,000-watt power supply. More importantly, the leak suggests a target launch price of $1,500. For PC builders who have watched the street price of the RTX 5090 skyrocket past $4,000 due to relentless AI industry demand, a $1,500 alternative offers a rare glimmer of hope for uncompromised 4K gaming.[1][2][6]

However, the rumor has drawn intense scrutiny and skepticism from hardware engineers over its proposed memory configuration. The leak claims the RTX 5090 SE will feature 32GB of GDDR7 memory running on a 384-bit memory bus. Mathematically, pairing a 384-bit bus with standard 2GB memory modules yields 24GB of VRAM, while utilizing newer 3GB modules would result in 36GB.[3][5]

To achieve exactly 32GB on a 384-bit interface, Nvidia would have to resort to a mixed-density memory configuration—combining 16Gb and 24Gb chips on the same printed circuit board. Critics point out that Nvidia has historically avoided this 'clamshell' or mixed approach in its ultra-high-end gaming products due to the signal integrity challenges, validation headaches, and uneven performance profiles it can introduce.[4][5]

The skepticism is compounded by the optics of the current GPU market. Analysts note that memory supply is tight industry-wide, driving up the cost of GDDR7 modules. Allocating 32GB of premium memory to a slightly cheaper enthusiast card while the budget segment remains starved for VRAM would be a controversial allocation of limited resources.[3][6]

Despite the technical doubts, the business case for an RTX 5090 SE remains compelling. Nvidia's current Blackwell gaming lineup features an unprecedented $1,000 gap between the RTX 5080 and the RTX 5090. Furthermore, the manufacturing process for the massive GB202 die inevitably produces chips with minor defects. Fusing off damaged compute units and selling the silicon as a 5090 SE allows Nvidia to monetize yields that would otherwise be discarded.[1][2][6]

The inclusion of 32GB of VRAM, if accurate, also speaks to the shifting demographics of high-end GPU buyers. While 16GB is generally sufficient for modern rasterized gaming, 32GB makes the card highly attractive to local AI developers, machine learning researchers, and 3D rendering professionals who are currently priced out of the flagship tier but need massive memory buffers.[1][3]

If the RTX 5090 SE is indeed in active development, it is unlikely to hit retail shelves imminently. Industry watchers suggest that a formal unveiling would most likely occur at CES 2027, potentially alongside a broader 'Super' refresh of the Blackwell architecture. Until Nvidia officially confirms the silicon, the $1,500 gap in the product stack remains wide open, leaving enthusiasts to debate the validity of the leaked spec sheet.[6]