It has been two years since we last saw a new gaming GPU from NVIDIA as last year’s Super GPUs were just speed bumps of the RTX 2000 series from 2018. Gamers have long awaited the arrival of NVIDIA’s RTX 3000 series, especially after seeing the countless rumors and performance expectations after the announcement of the Ampere architecture. The RTX 3000 series builds on NVIDIA’s Ampere architecture announced earlier this year and inside of the A100 supercomputing and AI graphics card. NVIDIA has continually shown graphics leadership in virtually every segment and, without a doubt, in discrete gaming performance. Everyone expects AMD to mount a competitive high-end offering this year, so naturally, one would expect that NVIDIA has come to market with a competitive offering. However, NVIDIA doesn’t plan to compete on raw GPU rasterized performance alone; the company is also increasingly leaning into the RT and Tensor cores it has added to the RTX series of GPUs to differentiate from the competition. Finally, NVIDIA’s GeForce ecosystem is gaining strength and stickiness with some of these new features and applications.
The RTX 3000 Series GPUs
As launched, the RTX 3000 series consists of three GPUs, with the RTX 3080 being the focus and center of the lineup. This is consistent with NVIDIA’s past strategies with virtually every GPU the company has launched in the past decade. All three RTX 3000 series GPUs are manufactured using Samsung’s 8nm custom process node, which is how NVIDIA can build this 28 billion transistor GPU with so many cores and still fit it within a graphics card form factor. These new GPUs feature NVIDIA’s new shader cores, 2ndgeneration Ray Tracing cores, and 3rd generation Tensor cores, and that’s just in the chip itself. The entire RTX 3000 series will also feature NVIDIA’s new RTX IO, which integrates multiple storage technologies, including Microsoft’s new DirectStorage API. DirectStorage leverages the benefits of super-fast PCIe Gen4 NVMe storage to load game assets directly to GPU memory from the system’s SSD storage reducing load times and improving latency. The RTX 3000 series will also feature GDDR6X from Micron, which the company already confirmed earlier this month. The inclusion of this new memory translates to nearly 1GB/s of memory bandwidth on the top-end RTX 3000 series card, the RTX 3090, with 24GB of VRAM. These new GPUs also feature support for HDMI 2.1, DisplayPort 1.4a, and the first discrete GPU to support the AV1 codec.
The RTX 3080 GPU, much like the RTX 2080 before it comes with three different types of cores inside of the GPU; Shader, Ray Tracing, and Tensor. NVIDIA upgraded each core in terms of performance, with the RTX 3080 having a massive 30 TFLOPS of shader performance compared to the 11 TFLOPS in the RTX 2080. This shader core improvement is almost triple the raw shading performance of the RTX 2080. The Ray Tracing cores are also being upgraded to 58 TFLOPS of Ray Tracing performance, according to NVIDIA, compared to 34 TFLOPS of performance in the RTX 2080. The 3rdgeneration Tensor cores are also being upgraded to 238 TFLOPS of performance from 89 TFLOPS from the RTX 2080. These represent massive performance improvements generation on generation. The RTX 3080 also sports 10GB of GDDR6X, which is 2GB more than the RTX 2080, but 1GB less than the RTX 2080 Ti. All this performance is possible with a 320W TDP, which is about 100W or almost 50% higher than the RTX 2080. So, you are paying for some of that increased performance with a more power-hungry GPU. NVIDIA says that this GPU will have a starting price of $699, which is the same MSRP as the RTX 2080 at launch, even though the RTX 2080 rarely sold for that price. We will have to wait until the RTX 3080 launches on September 17th to see real market prices.
The RTX 3090 is NVIDIA’s “BFGPU,” which NVIDIA designed to help the company retain its performance crown. The RTX 3090 is a slightly more power-hungry GPU than the RTX 3080 with a 350W TDP but also 24GB of GDDR6X instead of 10GB. The RTX 3090 is effectively an RTX 3000 series equivalent for NVIDIA’s Titan series, putting out the full-bore GPU for those that need the performance and are willing to pay for it. However, this time NVIDIA is making it more accessible and affordable at $1,499. However, this GPU really is not meant for most gamers as they’ll never fully make use of 24GB of GDDR6X or need the additional performance it affords. NVIDIA says that this is the first GPU to enable 8K 60 gaming, which makes sense because you would need 24GB to be able to handle that much resolution. The RTX 3090, according to NVIDIA, is 50% faster than the Titan RTX, which is why it commands such a premium. The RTX 3090 will be available starting September 24th, but like the RTX 3080, we will have to see where market prices settle considering Covid-19 and the run on computer components.
Judging by the above table, the RTX 3090 is about 20% faster in almost every way than the RTX 3080, which puts it into an entirely different class of its own. Therefore NVIDIA designed the RTX 3090 to be a stand-in for a potential 3080 Ti or a Titan Ampere card. We could still probably see a faster version of the Ampere cards as yields and cooling improve. Since the RTX 3090 is a triple-slot cooler, some considerable space savings are going to be possible with a liquid cooling solution from the usual suspects like EK or Bitspower.
There is also the RTX 3070, which provides the best value of the RTX 3000 family of GPUs at 220W and a retail price of $499. The RTX 3070, according to NVIDIA, is faster than an RTX 2080 Ti at less than half the price and is equipped with 8GB of VRAM. This value play from NVIDIA, combined with the absolute performance of the RTX 3080, will prove to be an exciting challenge for the competition to live up to. That said, AMD’s RX 5700XT did give NVIDIA’s RTX 2070 a run for its money and shifted the entire landscape below $499. NVIDIA says that the RTX 3070 will be available later than the RTX 3080 and RTX 3090 and will be available in October.
NVIDIA has invested considerable time and money into making the RT and Tensor cores in the RTX series of GPUs successful. When the RTX series launched two years ago, those benefits were not immediately recognizable when few games supported Ray Tracing or took advantage of the GPU’s tensor cores. However, now NVIDIA has worked with developers to bring Ray Tracing support to titles like Minecraft and now Fortnite. These games are incredibly popular with a younger audience who is always looking for ways to improve their gaming experience, budget, and parents allowing. Even wildly popular titles like Call of Duty Modern Warfare and Control have adopted NVIDIA RTX, and the hotly anticipated CyberPunk 2077 will when it launches later this year.
In addition to significantly improving ray tracing support, NVIDIA is also introducing applications like NVIDIA Reflex, which optimizes the rendering pipeline to improve reaction times and lower input to display latency. NVIDIA Reflex will be available in September, showing NVIDIA’s commitment to gamers and ensuring that the GPU can deliver the low enough latency responsiveness to match the 360 Hz displays that NVIDIA is helping to bring to the market. This fall, these ‘eSports’ displays are coming from partners like Acer, Alienware, ASUS, and MSI. These ‘eSports’ G-Sync monitors are for a smaller audience of gamers that wish to compete professionally and need the absolute highest frames and lowest latencies.
Another critical piece of software NVIDIA is announcing is RTX Broadcast, which will run on all RTX GPUs when it comes out in September. RTX Broadcast leverages the tensor cores and AI capabilities of RTX GPUs to do audio noise removal, which comes from NVIDIA’s current RTX Voice Beta, which Anshel has been using for months with very high praise. In addition to noise-canceling, however, RTX Broadcast also does background detection and allows you to insert any background you want for video broadcasts or streaming games. In addition to doing those things, it also enables webcam auto-framing, which automatically follows a user’s face. None of these features is necessarily novel to NVIDIA. Still, most of the time, these features are limited to specific applications, and now NVIDIA is making them available to virtually any application that will work with NVIDIA. If NVIDIA’s RTX Voice beta was an indication, RTX Broadcast should be a great piece of software for the millions of gamers out there who stream, but also for the millions of gamers who are currently using their gaming rigs for work-from-home during these COVID times as well.
This software experience all adds up to NVIDIA, adding stickier experiences that make the gaming experience better. Stickier experiences mean a lower likelihood of switching, and that’s a good thing.
The new RTX 3000 series is a massive improvement over the RTX 2000 series in nearly every way and leans more heavily into NVIDIA’s software and peripherals ecosystem.
At the top-end, you have the RTX 3090, which chases absolute performance and brings 8K60 gaming to the PC and will likely be the GPU that Microsoft Flight Simulator 2020 players opt for with their HP Reverb G2 VR headsets.
Then you have the flagship RTX 3080, which virtually doubles the performance of the RTX 2080 in nearly every way and sets a new bar for the discrete gaming GPU.
I believe that RTX IO and RTX Broadcast are also both great ways that NVIDIA can leverage the power of the RTX 3000 series to better compete with AMD on next-gen consoles and graphics. With people playing games more than ever, the attention is on NVIDIA to deliver new gaming innovations and enable higher performance and higher quality experiences as people play some of the most anticipated games of the year.
We eagerly await the availability of these new GPUs as they hit the shelves and will hopefully have a review to you soon.
Note: Moor Insights & Strategy analyst Anshel Sag contributed to this analysis.