Introduction

NVIDIA only just kicked off its GeForce RTX 50-series Blackwell generation last week, and we already have our second GPU from series, the new GeForce RTX 5080. We have many of these cards with us to review, but today we bring you the NVIDIA GeForce RTX 5080 Founders Edition, their de facto reference design. The RTX 5080 is being launched as the generation's second-best graphics card, and does many of the same things as the RTX 5090, but at a much lower price-point. NVIDIA recommends the RTX 5080 for maxed out gaming at 4K Ultra HD—the same use-case its predecessor the RTX 4080 was launched at. You get half the amount of memory as the flagship RTX 5090 at 16 GB, even the memory bus width is half of it, at 256-bit, and yet there's plenty of muscle in this card for gamers to play today's and tomorrow's games at 4K UHD.



The GeForce RTX 5080 debuts NVIDIA's second silicon from the GeForce Blackwell generation, the GB203. This is a physically smaller chip than the GB202 powering the RTX 5090—in fact it's half its size. What's interesting though, is that the RTX 5080 maxes out all available SM on the GB203. If NVIDIA has to release an "RTX 5080 SUPER" next year, it would have to tap into the much larger GB202. In case you missed it, NVIDIA is building Blackwell on the same exact foundry node as the RTX 40-series Ada from October 2022, the TSMC 4N, which is a specialized variant of the 5 nm EUV node that the foundry co-developed with NVIDIA. The GB203 has similar transistor counts and die-size to the AD103 silicon powering the RTX 4080.

The new GB203 silicon physically has 84 Blackwell streaming multiprocessors (SM) across 7 GPCs, and the RTX 5080 has all of them enabled. This works out to 10,752 CUDA cores, more than the 10,240 that the RTX 4080 SUPER comes with. It also comes with 336 Tensor cores, and 84 RT cores. The TMU count is 336, while the chip has 112 ROPs. The shared L2 cache size is unchanged from the previous generation at 64 MB. Memory bandwidth is where the RTX 5080 gets a significant upgrade over the RTX 4080. While the bus width is the same 256-bit, it drives newer GDDR7 memory chips, and NVIDIA picked memory speeds of 30 Gbps, yielding 960 GB/s of memory bandwidth (a 34% increase over the RTX 4080). The media and display engines of the RTX 5080 see generational updates over the RTX 4080, too, you get two each of the latest Blackwell NVDEC and NVENC accelerators. The display engine supports DisplayPort 2.1b with UHBR20, and HDMI 2.1a.

The new GeForce Blackwell graphics architecture lays the hardware-level groundwork for neural rendering, a revolutionary new concept in consumer 3D graphics, where generative AI plays a participatory role in the core rendering stack, and not just part of the DLSS Super Resolution feature, where it helps reconstruct details in upscaled frames. Just as NVIDIA discovered a way to combine certain real-time ray traced elements with classic raster 3D, it found a way to combine objects created by a generative AI with raster 3D. The company worked with Microsoft to standardize this in the DirectX 12 API, letting 3D applications directly access Tensor cores. The GPU also runs generative AI and 3D graphics rendering workloads in tandem thanks to a new hardware scheduler component called the AI Management Processor (AMP).

Blackwell also introduces DLSS 4 and Multi Frame Generation. DLSS 4 replaces the CNN (convoluted neural network) based AI models driving Super Resolution, Ray Reconstruction, and Frame Generation, with a new transformer based AI model that improves image quality at every performance tier. The updated AI models are available to even the RTX 40-series, RTX 30-series and RTX 20-series GPUs in games that implement DLSS 4, however Multi Frame Generation (MFG) is exclusive to the RTX 50-series. MFG is a technology that leverages AI to draw not just the frame after a conventionally rendered frame, but up to three succeeding frames. It requires the new Flip Metering hardware of the Blackwell display engine, which is what makes MFG exclusive to this generation.

The NVIDIA GeForce RTX 5080 Founders Edition features an identical-looking design to the RTX 5090 Founders Edition we reviewed last week, down to the same length, height, and 2-slot thickness. You wouldn't be able to tell the two apart installed until you pay attention to the decals on the backplate. This card features the same Double Flow Through thermal solution where the PCB is shrunk down to the bare minimum size and relocated to the center of the card, relying on breakaway PCBs for the display and PCIe I/O. The design allows for airflow from both fans to flow through heatsink fins of the cooler, venting out the back of the card. This cooler did wonders with the RTX 5090, which has a TGP of 575 W, and so it's only going to do better with the RTX 5080 and its 360 W TGP. NVIDIA is pricing the RTX 5080 Founders Edition at $999, which is also the starting price for this SKU.