MSI GeForce RTX 5090 Lightning Z Review – Up to 1000 W

Introduction

The MSI GeForce RTX 5090 Lightning Z is designed to be a paragon of PC hardware for the enthusiast class, attracting overclockers and professional gamers for whom money is no object when they want the best the market has to offer. This card, produced in very limited quantities, comes with one of the most robust graphics card VRM solutions in the industry that can handle continuous loads of up to 1000 W, using two 600 W power inputs; making the GeForce RTX 5090 GPU bound to the top capable of maintaining high clock speeds with minimal effort; and a host of first-class features.

MSI puts the GeForce RTX 5090 Lightning Z a few notches above its current flagship graphics card, the RTX 5090 SUPRIM Liquid SOC. The Lightning Z brand dates back nearly 15 years, when MSI designed special graphics card SKUs targeting professional overclockers. Fast forward to 2026, and the RTX 5090 Lightning Z is trying to do the same thing, but with so much power out of the box, it doesn’t need aftermarket cooling. MSI overclocked the RTX 5090 to 2730 MHz, compared to 2527 MHz for the SUPRIM Liquid SOC, and NVIDIA’s reference of 2407 MHz. That’s not all, while the SUPRIM Liquid SOC comes with a power limit of 575 W that can be expanded to 600 W with software, the new Lightning Z charges this to 800 W, opening the power limit that takes it all the way to 1000 W, provided that both 12V-2×6 connectors are in place and the connectors capable of using Extreme per hour are in the 60 position per option. The 2500 W BIOS pushes the power limits even further, but this may require a cooling modification.

The NVIDIA GeForce RTX 5090 represents the apex of the GeForce RTX Blackwell generation. At its heart is the massive GB202 chip, a 750 mm² slab of silicon with more than 92 billion transistors, and 192 broadcast multiprocessors (SM). The GPU introduces several first-to-market features, including a PCI-Express 5.0 x16 interface, ATX 3.1 + PCIe Gen 5 capabilities, DisplayPort 2.1b UHBR20, and the new GDDR7 memory standard. The GPU has a wide 512-bit GDDR7 interface that drives 32 GB of 28 Gbps GDDR7 memory for a jaw-dropping memory bandwidth of 1,792 GB/s. The GPU needs this for neural, DLSS 4 multi frame generation, and other new technologies to work. The RTX 5090 powers 170 out of 192 SM, and 96 MB of the 128 MB L2 cache physically present on the silicon. It also hosts 3 of 4 NVENC accelerators, and 2 of 4 NNVDEC accelerators. This works out to 21,760 CUDA cores, 680 Tensor cores, and 170 RT cores, across 11 GPCs.

Blackwell Graphic Architecture, now in its second year, presents Neural Rendering. You’ve already seen the incredible power of AI generation in creating photorealistic images and video. NVIDIA has found a way for the GPU to use the AI ​​model that generates and renders images in parallel, thanks to a new component called AMP (AI management processor). The GPU combines 3D objects created by a generating AI model with raster 3D images in much the same way that it combines real-time ray-traced objects. The result is a technological leap in photorealism and geometric detail. We were blown away by the tech demos NVIDIA showed us at CES, and we can’t wait to see game developers take the technology to task. For its part, NVIDIA has collaborated with Microsoft to standardize the technology, making it possible for 3D applications to deal directly with Tensor cores. The shader reprogramming engine supports neural shaders.

The new 4th Generation RT core has added Mega Geometry Hardware, the ability to render a ray traced object with a high poly count, and all those additional areas to precisely match the rays. Then there is DLSS 4. NVIDIA has replaced the convoluted neural networks (CNN) based AI model powering the various parts of DLSS, with a new transformer-based model that is more accurate, and provides higher image quality for all performance presets. This applies not only to the RTX 40-series Ada and RTX 30-series Ampere generations, but even the RTX 20-series Turing cards, exclusive to Blackwell, however, the Multi Frame Generation. NVIDIA has created a way for the AI ​​frame generation model to create not only the second frame that is normally rendered, but up to three AI-generated frames, taking into account motion vectors and other relevant information. Combined with high resolution, it takes the rendering power of 1 pixel to create up to 16 pixels. This feature relies on a key component of the hardware found in Blackwell’s display engine—hardware flip-metering—for frame-rate accuracy, which is why Multi Frame Generation is exclusive to Blackwell.

The MSI GeForce RTX 5090 Lightning Z is part of the latest wave of premium RTX 5090 graphics cards announced between late 2025 and early 2026. In the year since its announcement, the board’s partners have managed to accumulate enough numbers of highly bonded GB202 silicon, which they have combined with their studies on the RTX 5090 during the year, to design these new cards for the enthusiast segment. The Lightning Z is designed to compete fiercely with the likes of the ASUS ROG Matrix RTX 5090 and the GIGABYTE RTX 5090 AORUS Infinity.

The card comes with a customized liquid cooling solution that uses a full copper baseplate, with a cooling channel that draws heat form not only the GPU, but also the VRM and memory. This is then disposed of by a large 360 ​​mm radiator with three efficient fans. There are several touches for the enthusiast class such as dual 12V-2×6 inputs that reduce the load per connector (thus the connection pins run cooler), a powerful VRM solution capable of exceeding 1000 W, and several overclocker-friendly features. There’s a true-color display on the card itself, which when paired with the company’s Afterburner software, can be made to display almost anything, including real-time monitoring stats. MSI didn’t design this card for those who just want RTX 5090, but those who want the best RTX 5090. So it has a price of $5,090.