The Nvidia GeForce RTX 4090 hype practice has been constructing for many of 2022. After greater than a 12 months of maximum GPU costs and shortages, CEO Jensen Huang revealed key particulars at GTC 2022, with a worth positive to make many cry out in despair. $1,599 for the highest providing from Nvidia’s Ada Lovelace structure? Truly, that is solely $100 greater than the RTX 3090 at launch, and if the cardboard can come wherever close to Nvidia’s claims of 2x–4x the efficiency of an RTX 3090 Ti, there’ll undoubtedly be folks keen to pay it. The RTX 4090 now sits atop the GPU benchmarks hierarchy throne, at the very least at 1440p and 4K. For anybody who’s after the quickest potential GPU, by no means thoughts the worth, it now ranks among the many greatest graphics playing cards.
That is to not say the RTX 4090 represents worth, although that may get a bit subjective. Trying simply on the FPS delivered by the assorted GPUs per greenback spent, it ranks lifeless final out of 68 GPUs from the previous decade. Besides our customary rating makes use of 1080p extremely efficiency, and the 4090 most decidedly shouldn’t be a card designed to excel at 1080p. The truth is, it is so quick that CPU bottlenecks are nonetheless a priority even when gaming at 1440p extremely. Take a look at 4K efficiency and think about ray tracing, and you can argue it is presumably one of many greatest values — see what we imply about worth being subjective?
Once more, you may pay dearly for the privilege of proudly owning an RTX 4090 card, as the bottom mannequin RTX 4090 Founders Version prices $1,599 and associate playing cards can push the worth as much as $1,999. However for individuals who need the perfect, or anybody with deep sufficient pockets that $2,000 is not an enormous deal, that is the cardboard you may need to get proper now, and we would be shocked to see something surpass it on this era, wanting a future RTX 4090 Ti.
|Graphics Card||RTX 4090||RTX 3090 Ti||RTX 3090||RTX 3080 Ti||RX 6950 XT||Arc A770 16GB|
|Course of Know-how||TSMC 4N||Samsung 8N||Samsung 8N||Samsung 8N||TSMC N7||TSMC N6|
|Die measurement (mm^2)||608.4||628.4||628.4||628.4||519||406|
|SMs / CUs / Xe-Cores||128||84||82||80||80||32|
|Ray Tracing “Cores”||128||84||82||80||80||32|
|Enhance Clock (MHz)||2520||1860||1695||1665||2310||2100|
|VRAM Velocity (Gbps)||21||21||19.5||19||18||17.5|
|VRAM Bus Width||384||384||384||384||256||256|
|L2 / Infinity Cache||72||6||6||6||128||16|
|TFLOPS FP16 (FP8/INT8)||661 (1321)||160 (320)||142 (285)||136 (273)||47.4||138 (275)|
|Launch Date||Oct 2022||Mar 2022||Sep 2020||Jun 2021||Could 2022||Oct 2022|
Here is a have a look at the who’s who of the acute efficiency graphics card world, with the quickest playing cards from Nvidia, AMD, and now Intel. Clearly, Intel’s Arc A770 competes on a totally totally different enjoying area, but it surely’s nonetheless attention-grabbing to indicate the way it stacks up on paper.
We will merely refer you to our Nvidia Ada Lovelace Architectural deep dive if you wish to find out about all the brand new applied sciences and adjustments made with the RTX 40-series. The above specs desk tells a variety of what it is advisable to know. Transistor counts have practically tripled in comparison with Ampere; core counts on the RTX 4090 are 52% greater than the RTX 3090 Ti; GPU clock speeds are 35% sooner, and the GDDR6X reminiscence? It is nonetheless largely unchanged, besides there’s now 12x extra L2 cache to maintain the GPU from having to request knowledge from reminiscence as typically.
On paper, that provides the RTX 4090 simply over double the compute efficiency of the RTX 3090 Ti, and there are undoubtedly workloads the place you may see precisely these types of positive factors. However below the hood, there are different adjustments that may additional widen the hole.
Ray tracing as soon as once more will get an enormous emphasis, and three new applied sciences — Shader Execution Reordering (SER), Opacity Micro-Maps (OMM) and Displaced Micro-Meshes (DMM) — all provide potential enhancements. Nonetheless, in addition they require builders to make use of them, which suggests present video games and engines will not profit.
Deep studying and AI workloads additionally stand to see huge generational enhancements. Ada contains the FP8 Transformer Engine from Hopper H100, together with FP8 quantity format assist. Meaning double the compute per Tensor core, for algorithms that may use FP8 as a substitute of FP16, and as much as 4 occasions the number-crunching prowess of the 3090 Ti.
One algorithm that may make the most of the brand new Tensor cores — together with an improved Optical Movement Accelerator (OFA) — is DLSS 3. The truth is, DLSS 3 requires an RTX 40-series graphics card, so earlier RTX playing cards will not profit. What does DLSS 3 do? It takes the present and beforehand rendered frames and generates an additional in-between body to fill the hole. In some instances, it could actually practically double the efficiency of DLSS 2. We’ll take a more in-depth have a look at DLSS 3 later on this assessment.
From an expert perspective, significantly for anybody involved in deep studying, you may simply justify the price of the RTX 4090 — time is cash, and doubling or quadrupling throughput will certainly save time. Content material creators will discover so much to love and it is a fast and straightforward improve from a 3090 or 3090 Ti to the 4090. We’ll have a look at ProViz efficiency as effectively.
However what about avid gamers? Not like the RTX 3090 and 3090 Ti, Nvidia is not happening about how the RTX 4090 is designed for professionals. Sure, it can work nice for such folks, but it surely’s additionally a part of the GeForce household, and Nvidia is not holding again on its gaming efficiency claims and comparisons. Possibly the previous two years of cryptocurrency mining are guilty, although GPU mining is now unprofitable so at the very least avid gamers will not must struggle miners for playing cards this spherical.