Introduction

MSI debuts its new Vanguard line of graphics cards with the GeForce RTX 50-series, and we have with us the MSI RTX 5080 Vanguard SOC graphics card, in its Launch Edition collector's package. The new Vanguard line of graphics cards are positioned between MSI's SUPRIM line of halo-segment custom design graphics cards, and its more popular Gaming X series. The company retired the Gaming Z series with the advent of SUPRIM, so I guess you could call Vanguard the spiritual successor to the Gaming Z. The MSI Vanguard series combines aspects of both SUPRIM and Gaming X. From the SUPRIM you see hints of titanium gray metal alloy embellishments on the cooler shroud. The backplate has a matching brushed-aluminium appearance, and from the Gaming X series, it has certain sporty elements, such as carbon fiber surfaces and acrylic RGB LED lighting diffusers. Under the shroud, the Vanguard features many of the cooler innovations MSI introduced with the SUPRIM series this generation.



The GeForce RTX 5080 is NVIDIA's latest enthusiast-segment graphics card, designed for maxed out gaming at 4K Ultra HD, with ray tracing enabled. The flagship RTX 5090 from last week does the same thing, but faster, and has other use cases such as AI development and content creation, thanks to its much larger video memory and memory bus. The RTX 5080 squarely targets gamers looking for a 4K gaming experience, with a few allied workloads on the side, such as game streaming. If compare the specs sheets of the RTX 5090 and RTX 5080, you'll find that the latter is almost half the card the RTX 5090 is, and is yet squarely an enthusiast-segment GPU. The RTX 5080 debuts the new GB203 silicon, which is the second largest to implement the GeForce Blackwell graphics architecture. With this generation, NVIDIA is holding on to the process node, so these chips are built on the same NVIDIA 4N silicon fabrication node that the company codeveloped with TSMC, based on the foundry's 5 nm EUV node.

The GeForce Blackwell graphics architecture introduces a revolutionary new concept in consumer 3D graphics, called Neural Rendering. The company figured out a way to combine objects created by a generative AI with conventional raster 3D graphics, just like it discovered a way to combine real time ray traced objects with RTX late last decade. You already know the ability of popular generative AI models to create photorealistic images and video, so you know this technology has immense potential. GeForce Blackwell GPUs have the ability to accelerate generative AI and render 3D graphics in tandem, thanks to a new hardware scheduling component called the AI management processor. The architecture also introduces the ability for 3D applications to directly access the Tensor cores, and NVIDIA worked with Microsoft to standardize this at the DirectX API level.

The new Blackwell SM features concurrent FP32 and INT32 capability on all 128 CUDA cores, the previous generation Ada SM only has INT32 capability on half its cores. The 5th Generation Tensor core puts on FP4 data format capability to step up throughput by trading in precision. The 4th Generation RT core has the hardware groundwork for Mega Geometry, letting ray traced objects have exponentially higher polygon counts. DLSS 4 and Multi Frame Generation are this generation's other major highlights. DLSS 4 introduces a new Transformer based AI model replacing the CNN-based one, for all its subcomponents, namely, super resolution, ray reconstruction, and frame generation. This model is more accurate, and hence steps up image quality in every performance tier. Multi Frame Generation gives the GPU the ability to use AI to draw not just one succeeding frame to a conventionally rendered one, but four such succeeding frames, effectively quadrupling frame rate. It relies on the new Flip Metering component of the Blackwell Display Engine to achieve this.

Blackwell implements many technological firsts for the company. It's the first gaming GPU to use the PCI-Express 5.0 x16 host interface, the 12V-2x6 power connector, and feature an ATX 3.1 + PCIe 5 CEM power architecture. It's also the first generation to implement the new GDDR7 memory standard, which doubles efficiency over GDDR6. The Blackwell display engine implements DisplayPort 2.1b with UHBR20 on all ports. When used with DSC, it enables 8K high refresh-rates with a single cable. You can see where this is going with DLSS 4 MFG.

The GeForce RTX 5080 maxes out the GB203 silicon, enabling all 84 SM, which works out to 10,752 CUDA cores, 336 Tensor cores, 84 RT cores, 336 TMUs, and 112 ROPs. It also features all 64 MB of the on-die L2 cache, and two sets of NVDEC and NVENC video accelerators that support 4:2:2 formats for AV1 and HEVC. The chip features a 256-bit wide GDDR7 memory bus, which drives 16 GB of memory. NVIDIA is running this memory at 30 Gbps on the RTX 5080, yielding 960 GB/s of bandwidth—a 34% increase over that of the RTX 4080.

The MSI GeForce RTX 5080 Vanguard SOC, as we said, combines the cooling innovations of the SUPRIM series, with the aesthetics of both the SUPRIM and Gaming X series. The new Hyper Frozr cooling solution comes with a vapor chamber plate making contact with the GPU and memory, which conveys heat to a cluster of heat pipes that have been flattened to a cuboidal shape for better contact with the vapor chamber. From here, the heat is spread across a large aluminium fin-stack with wavy edges that increase turbulence for better heat dissipation. The cooler relies on a trio of MSI's StormForce fans, the same exact ones found in the company's RTX 5080 SUPRIM graphics card which we are also reviewing today. There are two sub-variants of the RTX 5080 Vanguard, SOC and OC, we're reviewing the SOC, which comes with a higher factory overclock of 2730 MHz on both the Gaming and Silent BIOSes, compared to 2617 MHz reference. The Launch Edition is a special package that includes a puzzle toy based on Lucky, MSI's dragon mascot. MSI is pricing this card at $1230, a $230 premium over the $999 baseline.