Neat Blog

NVIDIA RTX 3090. Should you?

A little more than three weeks ago the world got to try the first NVIDIA GeForce RTX 3090 cards.

NVIDIA is claiming that this is a “big ferocious GPU (BFGPU) with TITAN class performance”. AND by big they mean BIG. It’s almost two times bigger than RTX 2080 Ti was!

The new card is powered by Ampere — NVIDIA’s 2nd gen RTX architecture. The crazy number of 10496 CUDA cores (aka shaders) packed in this GPU is something that can’t be found in any other consumer-grade GPU at this stage. Like all cards in RTX 30 series, 3090 supports the new generation PCI Express 4.0 bus protocol which is twice as fast as PCI Express 3.0. But the most important characteristics that set this card apart from an also great GPU, RTX 3080, are the bandwidth and the memory.

 RTX 3090RTX 3080 RTX 2080 Ti  
Architecture  Ampere  Ampere  Turing  
CUDA cores  10496   8704   4352  
Internal memory bandwidth   936 GB/s   760 GB/s   616 GB/s  
Memory   24 GB GDDR6X   10 GB GDDR6X  11 GB GDDR6 
Bus protocol  PCI Express 4.0  PCI Express 4.0   PCI Express 3.0  

As we’ve written in the article on NVIDIA GeForce RTX 3080, the real-life video denoising performance is not defined by the raw computing horsepower alone. In many cases, the actual performance of a GPU depends not only on the number of cores, but also on the internal memory bandwidth as cores may be waiting for data to be read from GPU memory into registers or written from registers to GPU memory. Therefore internal memory bandwidth has a big impact on calculation speeds as data is made available to GPU computing cores sooner. This card has come a long way from it’s smaller siblings and is offering the 936 GB/s bandwidth thanks to the wider 384-bit GDDR6X bus.

RTX 3090 has also been packed with massive 24 GB of on-board memory. Why is this important? Because GPU memory has been one of the main bottlenecks on the way to the efficient collaboration of video editing software, plugins, and other applications you might be running on your computer. Video memory is that resource that usually gets exhausted first not allowing a GPU to reach its full potential. With 24 GB memory, you should be able to work comfortably even when “heavy” GPU memory-hungry effects are applied to large video frames in GPU-accelerated video editing applications.

Now let’s see how all those numbers translate into render speed. The table below represents the speed of Neat Video only without video editing application overhead (aka NeatBench speed).  

Frame size Temporal radius
Frames per second
RTX 3090
Frames per second
RTX 3080
Frames per second
RTX 2080 Ti
1920x1080 2 67.1 59.7 45.4
1920x1080 5 48.5 48.5 36.8
3840x2160  20.9  18.1   12.8  
3840x2160   5   15.3   13.5  10.2  


The speed improvement from RTX 2080 Ti and RTX 3090 is huge, however, the performance difference between 3080 and 3090 seems to be not as impressive as the price gap between them. So why you would want to spend extra $$$ for NVIDIA’s flagship? As we've pointed out earlier, RTX 3090 is shipped with 24 GB of memory That should ensure that this GPU will deliver great results not only in isolated tests, but also in real-life scenarios when your editing application, Neat Video, codecs, and other effects are all willing to get their slice of the GPU memory pie to work fast and stably. This is especially important if you are dealing with 4K or larger frames.