Introduction

NVIDIA GeForce RTX 4080 "Ada" is the the latest high-end graphics card from team-green after the sheer domination it brought to the high-end segment last month, with the RTX 4090. The new GeForce RTX 4080 SKU in this review succeeds the RTX 3080, which NVIDIA continued to maintain as its flagship product in the RTX 30-series till the end, treating everything higher as "halo" products. The RTX 4080 is designed with many of the same goals as the RTX 4090—to plow through games at maxed out settings at 4K, including ray tracing, but at a slightly more acceptable price point of $1,200, or 25% less. It is still designed to be faster than anything from the previous-generation.



NVIDIA had originally planned to fork the successor of the RTX 3080, the RTX 4080, into a sub-series of two SKUs, the RTX 4080 16 GB (covered in this review) and the now-cancelled RTX 4080 12 GB at $900—a price closer to the original RTX 3080 10 GB MSRP. This caused some controversy as the 12 GB variant was a vastly different product besides the memory size, and heaped bad press for NVIDIA in the social media. The RTX 4080 16 GB is hence now simply called the RTX 4080. We don't know for sure what NVIDIA will do with all those RTX 4080 12 GB that are already manufactured, probably re-brand them as some other SKU, we'll have to see.

With the RTX 40-series "Ada" generation, NVIDIA achieved a minor course-correction in the way the RTX 4080 is created. Since it is bound to sell in greater volumes than the RTX 4090, the company isn't carving the SKU out by basing it on the same large AD102 silicon and disabling a ton of CUDA cores. There was a large gap in the CUDA core counts between the previous-generation GA104 and GA102, which allowed AMD to slip in competitive products such as the RX 6800, which couldn't be beaten by any GA104-based SKU, including the RTX 3070 Ti. Therefore, NVIDIA developed a new in-between silicon between the two, to better optimize manufacturing costs, the AD103. There is such a silicon even in past generations, such as GA103, but those were confined to the mobile platforms.

The new 5 nm AD103 silicon only has two-thirds the die area of the AD102, which amounts to great manufacturing cost savings. Its memory interface is slashed by a third, too, with a 256-bit GDDR6X memory interface. NVIDIA has increased the memory size generation-over-generation, to 16 GB (up from 12 GB or 10 GB of the RTX 3080), however, the memory bus is narrowed, and while the memory speed is only marginally higher at 22.4 Gbps (versus 19.5 Gbps), the overall memory bandwidth is lesser, at 716 GB/s, compared to 760 GB/s of the RTX 3080 10 GB. This shortfall in memory bandwidth is claimed to be compensated for with architectural improvements, such as larger on-die caches.

The GeForce RTX 4080 is endowed with 9,728 CUDA cores across 76 streaming multiprocessors (SMs) out of the 80 physically present on the silicon. it gets 76 RT cores, 304 Tensor cores, 304 TMUs, and 112 ROPs. NVIDIA has generationally increased GPU clock speeds, and the RTX 4080 ticks at 2.25 GHz, with a memory speed of 22.5 Gbps (GDDR6X effective). The typical power of the card is 320 W, nearly 25% lower than that of the RTX 4090, and many of the board partners appear to be using cooling solutions they originally designed for the RTX 4090, meaning the custom-design RTX 4080 cards should run quieter.

NVIDIA GeForce RTX 4080 Founders Edition is the company's take on a premium custom-design card of its own. This isn't a "reference design" by any stretch, but an attempt by NVIDIA engineers to establish a high benchmark for product design for its board partners to beat with their custom-designs. The RTX 4080 Founders Edition (FE) card looks practically identical in shape and dimensions to that of the RTX 4090 FE, which means the GPU will benefit from a cooling solution that was designed to deal with a more power-hungry GPU. The most interesting aspect of the FE package is the included adapter that converts 8-pin PCIe into 16-pin 12VHPWR—it only has three 8-pin connectors, compared to four on the one NVIDIA includes with the RTX 4090 cards. The adapter is keyed for 450 W power delivery, identical to the one included with the RTX 3090 Ti.