Queen of the Castle For the majority of 2022, the Nvidia GeForce RTX 4090 hype train has been building. CEO Jensen Huang revealed key details at GTC 2022, including a price that is certain to make many cries of despair after more than a year of extreme GPU prices and shortages. The most expensive product from Nvidia’s Ada Lovelace architecture costs $1,599? In fact, that is only $100 more than the RTX 3090 at launch, and if the card can even come close to Nvidia’s claims that it will perform 2x–4x as well as an RTX 3090 Ti, there will undoubtedly be individuals who are willing to pay for it. The RTX 4090 has risen to the top of GPU benchmarks, at least at 1440p and 4K, and it now ranks among the best graphics cards for anyone looking for the fastest GPU, regardless of price.
This does not mean that the RTX 4090 is a good value, although this can be somewhat subjective. Out of 68 GPUs from the previous decade, it comes in dead last in terms of FPS delivered per dollar spent. Except that 1080p ultra performance is used in our standard ranking, and the 4090 is clearly not a 1080p-specific card. In point of fact, even when gaming at 1440p ultra, CPU bottlenecks remain a concern. You could argue that 4K performance and ray tracing make it one of the best values—get what we mean about value being subjective?
Again, owning an RTX 4090 card will cost you a lot, as the base model, the RTX 4090 Founders Edition, costs $1,599, and partner cards can cost up to $1,999. However, this is the card you should get right now if you want the best or have enough money that spending $2,000 isn’t a big deal. We wouldn’t be surprised to see anything surpass it in this generation, with the exception of a future RTX 4090 Ti.
The fastest graphics cards from Nvidia, AMD, and now Intel are showcased here, along with a look at the who’s who of the extreme performance graphics card industry. Despite the fact that Intel’s Arc A770 competes on a completely different playing field, comparing how it stacks up on paper is still interesting.
If you want to learn about all the new technologies and changes made to the RTX 40-series, we will just direct you to our Nvidia Ada Lovelace Architectural deep dive. Almost everything you need to know can be found in the specs table above. When compared to Ampere, transistor counts have nearly tripled; The RTX 4090 has 52% more cores than the RTX 3090 Ti; The GDDR6X memory and GPU clock speeds are 35% faster. Except that there is now 12 times more L2 cache to prevent the GPU from having to request data from memory as frequently, it has largely remained unchanged.
On paper, that gives the RTX 4090 compute performance that is just over twice that of the RTX 3090 Ti. There are certain workloads where you will definitely see gains of this kind. However, under the hood, there are additional modifications that have the potential to further widen the gap.
Shader Execution Reordering (SER), Opacity Micro-Maps (OMM), and Displaced Micro-Meshes (DMM) are three new technologies that offer potential enhancements to ray tracing. However, existing engines and games won’t benefit because they also require developers to use them.
Workloads for AI and deep learning are also likely to see significant generational advancements. Ada supports the Hopper H100’s FP8 Transformer Engine in addition to the FP8 number format. For algorithms that can use FP8 instead of FP16, this means twice as much compute per Tensor core and up to four times as much ability to crunch numbers as the 3090 Ti.
DLSS 3 is an algorithm that can use the new Tensor cores and a better optical flow accelerator (OFA). In point of fact, an RTX 40-series graphics card is required for DLSS 3, so earlier RTX cards will not benefit. What is DLSS 3 used for? It creates an additional frame between the currently rendered and previously rendered frames to fill the void. It can nearly double DLSS 2’s performance in some situations. Later on in this review, we’ll examine DLSS 3 in greater detail.
The RTX 4090’s price is easy to justify from a professional perspective, especially for deep learning enthusiasts: time is money, and doubling or quadrupling throughput will unquestionably save time. The 4090 is a quick and simple upgrade from a 3090 or 3090 Ti that content creators will enjoy. We will also examine ProViz’s performance.
What about gamers, though? Nvidia isn’t talking about how the RTX 4090 is made for professionals, unlike the RTX 3090 and 3090 Ti. Despite the fact that it is a member of the GeForce family and will work well for such individuals, Nvidia is being forthright about its gaming performance claims and comparisons. Although GPU mining is now unprofitable, gamers won’t have to compete with miners for cards this round, so it’s possible that the past two years of cryptocurrency mining are to blame.
