Nvidia GeForce RTX 4090 Founders Edition Review: Unrivaled 4K 100 fps behemoth for US$400 less than the RTX 3090 Ti

In this review, we take a good look at the Ada flagship, the Nvidia GeForce RTX 4090 Founders Edition, and see how much of a gen-to-gen performance uplift the new architecture offers over Ampere and AMD RDNA 2 flagships.

The thermal subsystem also sports a familiar look and functionality as the previous generation — one fan sucks air from the bottom while the other sucks this air from the card and pushes it on the opposite end, usually towards the top of the case. This means that if you place the card conventionally, the radiator fans or "pull" exhaust fans should be ideally located on the top of your case to prevent potential heat buildup.

Nvidia first introduced the PCIe Gen 5 power connector with the RTX 3090 Ti and mandated its inclusion even on add-in board (AiB) partner models. This year, the PCIe Gen 5 connector, dubbed the 12VHPWR adapter, has drawn a lot of flak for increasing number of melting reports.

In order to enable the RTX 4090 to fully flex its muscles, you will need a modern CPU of the likes of AMD Zen 4 or Intel Alder Lake/Raptor Lake processors — the AMD Ryzen 9 7900X/7950X or the Intel Core i9-13900K are worthy companions to the RTX 4090.

The Ryzen 7 5800X3D can be a good option as well if you wish to stick with the AM4 platform for now. However, other AM4 CPUs, including the Ryzen 9 5950X, can severely bottleneck the RTX 4090, especially at resolutions below QHD.

In our initial testing with the Ryzen 9 5950X, we found that the RTX 4090's performance at 1080p was so bottlenecked that some of the scores at various settings this resolution were often not that different from an RTX 3090 Ti. Therefore, the current review relies on a Core i9-13900K platform.

The following components were used in our build:

• Cooler Master MasterFrame 700 test bench
• Intel Core i9-13900K
• Gigabyte Z790 Aorus Master motherboard
• Gigabyte Aorus Gen 4 2 TB NVMe SSD for benchmarks and games
• 2x 16 GB Kingston Fury Beast DDR5-6000 RAM with timings 36-38-38-80 at 1.35 V
• Gigabyte M28U 4K 144 Hz monitor
• Cooler Master MasterLiquid ML360L V2 ARGB AIO cooler
• Cooler Master MVE Gold 1250 V2 fully modular PSU

A fresh installation of Windows 11 22H2 with the latest patches was used together with Game Ready driver version 526.47.

We would like to thank Cooler Master for supplying the test bench, AIO, and PSU, Gigabyte for the Core i9-13900K, Z790 motherboard, M28U monitor, and NVMe SSD, and AMD for the Kingston DDR5 memory.

Combining all synthetic benchmarks together, we find that the RTX 4090 FE puts up a significant 51% lead over the RTX 3090 Ti and a 58% lead over the RTX 3090 FE and also with the Radeon RX 6950 XT. 

The Radeon RX 6950 XT does have a perceivable advantage over the RTX 3090 Ti in synthetic 3DMark graphics tests. It remains to be seen how the new RDNA 3 cards would fare once they become available next month, but the RTX 4090 FE is able to get itself a big enough lead for now. Interestingly, the RX 6950 XT is only 9% behind the 4090 FE in Fire Strike Graphics while it trails by up to 40% in other tests such as Time Spy Graphics.

Unigine tests show Nvidia cards leading the way in general and the RTX 4090 FE in particular. We get to see 50%+ improvements in both the Heaven 4.0 tests and also in the Valley 1.0 OpenGL test compared to the RTX 3090 Ti.

Superposition, particularly at higher resolutions, shows the RTX 4090 FE completely decimating the RX 6950 XT by as much as 93% while also showing leads up to 79% over the RTX 3090 Ti.

The RTX 4090 FE's hegemony continues in compute benchmarks as well. In ComputeMark overall, we find Nvidia's Ada flagship attaining a 93% lead over the RX 6950XT and a highly significant 66% gain over the RTX 3090 Ti.

LuxMark presents an interesting paradigm, however. While the RTX 4090 FE leads the RX 6950XT in the Room test by a whopping 150%, AMD cards generally show better performance in Sala.

The RTX 4090 is able to offer significantly better performance in pro apps than other last gen consumer GPUs solely due to its brute hardware power. It is still a GeForce card at the end of the day, so performance in these applications is artificially limited by the driver. 

The GA102-based RTX A6000 has a 109% gain in SPECviewperf tests solely due to driver optimizations. It is for the same reason that even the Turing-based mobile Quadro RTX 5000 is able to offer much better performance in Solidworks and Siemens NX, though they trail behind in workloads such as 3Ds Max and Maya.

Nvidia allows changing the VRAM ECC state even with regular Game Ready drivers. This option is of no use to gamers since it can reduce memory speeds. However, it can come in handy for professionals running long simulations that can make use of the RTX 4090's massive VRAM but do not want to shell out heaps of cash for the Quadros.