Introduction

Welcome back to a new one-on-one graphics card comparison article in a series we call TPU50. in each of these, we compare two graphics cards from a comparable market segment across an enormous selection of 50 games, each spanning the three most popular resolutions for a total of 600 benchmark runs. For this review, we are pitting NVIDIA's GeForce RTX 3070 Ti against AMD's Radeon RX 6800.



Both the RTX 3070 Ti and RX 6800 represent an "aspirational" market segment which would have under normal circumstances been around $600. A more realistic price for these cards in these times is $1200 for the RTX 3070 Ti and $1500 for the RX 6800. The RTX 3070 Ti possibly came into existence only because of the RX 6800. AMD's newest Radeon RX 6000 graphics card series, which is based on the 7 nanometer RDNA 2 graphics architecture, propelled the company back into the high-end market segment for the first time in eight years. The Radeon RX 6800 XT, RX 6800, and RX 6900 XT lead the top-end and are very competitive with NVIDIA's offerings.

The Radeon RX 6800 XT ended up crossing swords with the GeForce RTX 3080, which was the posterboy of NVIDIA's Ampere architecture, while the RX 6800 non-XT came in significantly faster than the RTX 3070 at only a slight price premium. This prompted NVIDIA to make a number of changes to the lineup, including introducing the RTX 3070 Ti to better compete with the RX 6800, and the RTX 3080 Ti to fill the huge performance gap that existed between the RTX 3080 and RTX 3090.



The RX 6800 is based on the same "Navi 21" silicon as the RX 6900 XT, but is hugely cut down—60 out of 80 compute units are enabled, which works out to 3,840 stream processors. The ace up its sleeve is the memory subsystem, which is exactly the same as the RX 6900 XT. You get 16 GB of GDDR6 memory that ticks at a data rate of 16 Gbps across a 256-bit wide memory bus cushioned by the full 128 MB Infinity Cache memory physically present on the 7 nanometer silicon, which is manufactured at TSMC.

The RTX 3070 Ti, on the other hand, maxes out the second-largest GeForce Ampere silicon available, the GA104, by enabling all 48 streaming multiprocessors available on the die. The RTX 3070 only had 46 of these. NVIDIA also bolstered the memory subsystem, giving the card 8 GB of 19 Gbps GDDR6X memory, which resulted in a massive 35% increase in memory bandwidth. The GA104 GPU is made on a 8 nanometer process at Samsung.

This article should help you make a much more informed decision on which card to buy as we're testing the two across a much broader selection of games than our launch-day reviews, and with much newer drivers.

Benchmarks

Among our games, you'll find titles that have been included in TechPowerUp graphics card reviews over the past years, as well as some of the newer ones joining our bench soon. Going forward, we will of course make changes to the game selection for TPU50. Our goal is to have a rich diversity of game genres, engines, and 3D graphics APIs. Equal settings were used for both cards, and proprietary features disabled.

All games are tested in custom bench scenes as the integrated benchmarks often paint a completely inaccurate picture compared to actual gameplay. Also, the GPU vendors actively optimize their drivers to achieve good results in integrated benchmarks.

Test System