You read the title right… this is not a drill: hot on the heels of a 21 day celebration of 21 years of GeForce GPUs, the NVIDIA RTX 3090 Founders Edition is here in all of its rechonkulous glory.
Within this article, we are going to break down the features set of the 2nd Generation RTX line-up, putting it through a variety of gaming and creative benchmarks, and discovering just who benefits from the release of this TITANic beast.
Strap in, people, because we’re about to see what 21 years of experience has led to!
Specifications
- MSRP: $1499.99
- Core Clock: 1.4 GHz (Base) 1.7 GHz (Boost)
- CUDA Cores: 10496
- Memory: 24 GB GDDR6X
- Memory Bus: 384-bit
- Bus: PCI-e 4.0
- Ports: HDMI 2.1, 3x DisplayPort 1.4a
- Ray Tracing Cores: 2nd Generation
- Tensor Cores: 3rd Generation
- TGP: 350W
- Supports 8K Gaming, NVIDIA G-Sync, NVIDIA DLSS, G-Sync HDR
Editor's Note: Due to allocation and customs issues, our RTX 3080 samples were delayed until after receiving the RTX 3090. That review and those comparisons are on the way, but we decided to move ahead with this review first for the sake of timeliness and answering the questions many gamers have about this card. We'll be out of order for the launch timelines, but those reviews are in process now.
Before we get too far into this review, there is something that we need to get out of the way as we digest all of this information.
The RTX 3090 is marketed as delivering “the ultimate gaming experience,” but the RTX 3090 is a TITAN-class GPU in everything but name. The TITAN series has served as the “prosumer” GPU in between the workstation class QUADRO-series and RTX/GTX-series, respectively. This class of GPU has historically only been available from NVIDIA, but the company is now making it available in both a Founders Edition (which we are reviewing today) and from board partners.
The conclusions we draw will be based on this information along with our discoveries from testing.
Design
With the RTX 30-series, NVIDIA has taken an avant-garde approach to their design. Instead of looking at what has been done previously and repeating it with a shiny coat of paint, the design team took an approach to thermal management to see just how much heat they could take out of the equation. What the team came up with was a radical redesign, mixing up the airflow with a dual-axial thermal solution that pulls air into one side and pushes it out on the extreme opposite end while also pushing it out the back. We will talk more about how well this works later on, but we have to talk about getting this beast out of the box first.
There is no gentle way of putting this: the RTX 3090 is a chonkie GPU, especially in comparison to the RTX-20 series Founders Edition cards. Coming in at around 7 lbs., this thing was quite intimidating to behold, especially when considering that it was going into a mid tower case (Corsair iCUE 4000X) with a front mounted radiator! While the RTX 3090 does take up three PCIe mounting slots, the Founders Edition card is just slightly bigger than MSI’s RX 5700XT Gaming X. That being said, if you are considering one, check your case to make sure it will fit!
From an aesthetic standpoint, the Founders Edition RTX 3090 is stunning. The GPU is encased in a gunmetal grey möbius strip that accents the black thermal fins and fans. The NVIDIA logo and series designation are etched tastefully on opposite sides of the card and a white LED RTX 3090 logo lights up on the side when the card is receiving power.
For connectivity, the RTX 3090 has one HDMI 2.1 port that is capable of connecting to 8K HDR TVs along with three DisplayPort 1.4a connections. That is an omission from this generation of GPUs is the USB-C VirtualLink port for VR headsets found on the RTX 20-series. However, this isn’t necessarily a loss for what this GPUs is made for
The RTX 30-series does have a proprietary connection for these GPUs that is mounted near the center of the card. The included adapter does allow users to plug in their existing 8(6+2)-pin connections, but it is a bit unsightly - especially if you are using color-themed cables.
It isn’t all external redesign with the RTX 30-series, NVIDIA’s Ampere architecture delivers a whole new level of processing to the table. Leveraging a triple-process system, the RTX 30-series divides up real-time ray tracing to be handled by the RT cores, the Tensor Core handles AI-related talks such as deep learning super samples (DLSS) or the new RTX Broadcast, and programmable shaders for handling rasterization (or typical game rendering). While we see these features at work in the 20-series, Ampere turns up the specs in each of their iterations.
To further solidify the RTX 3090’s position, here is a breakdown of specs between it and the TITAN RTX:
- Programmable Shaders: 35.6 Shader-TFLOPs (up from 16.3 TFLOPs)
- Tensor Core: 284.7 Tensor-TFLOPs (up from 130.5 Tensor-TFLOPs)
- RT Cores: 69.5 RT-TFLOPs (up from 49.2 RT-TLOPs)
- Memory Clock: 19.5 Gbps (up from 14 Gbps)
- Memory Bandwidth: 936 GB/sec (up from 672 GB/sec)
- Memory Interface: 384-bit (same as TITAN RTX)
Now, it is important to point out that a lot of comparisons are being drawn between the RTX 30-series and the new console generations. Because these are two different architectures, we will hold off on the comparisons until we can see Big Navi and its kin in action.
Features Set
RTX I/O, NVIDIA Reflex, NVIDIA Broadcast
NVIDIA has taken a measure of pride in their ability to leverage AI in everyday computing tasks through the RTX GPUs, pushing the boundaries of just how a GPU can enhance the gaming experience. With the announcement of the RTX 30-series came a suite of new offerings for RTX-enabled GPUs that benefit in different use-cases.
The first among those was the eSports-focused NVIDIA Reflex. This feature offers a reduction in input (or mouse-to-display) latency while running games at high resolutions. With titles like Apex Legends, Valorant, Fortnite, and Call of Duty: Modern Warfare listed among the titles that support this feature, NVIDIA aims to help players clinch their wins with a bit of help. While available with RTX-base GPUs, this feature will also be integrated into a series of high refresh rate monitors.
RTX I/O offers a GPU-assistance to reduce the bottleneck found between the CPU and storage drives. With the fourth generation of PCIe becoming the new standard, the bandwidth needed to access information at those high speeds is becoming more demanding. By leveraging GPU memory, CPU utilization can be dropped by a significant amount. At the time of testing, we did not have a PCIe 4.0 NVMe drive to test, but we are looking forward to seeing this in action.
For content creators, NVIDIA Broadcast is going to be a game-changer. This tool puts the control in the hands of the user to do everything from mitigate background noise to creating a bokeh effect on video recording. I say this is a game-changer because some of these features - which have been staples for high-quality production - can be removed or controlled by equipment that the user already has: a GPU. NVIDIA Broadcast may just remove the financial barrier of entry to allow some new stars of streaming rise to the top within heavy investment outside of typical gaming gear.
DLSS 2.0
Outside of these new additions to the GPU flexibility, we also have DLSS 2.0, the latest edition of Nvidia's RTX-based scaling technology. As mentioned earlier, DLSS, short for deep learning super sampling, uses the architecture to render in-game graphics at a lower resolution while enabling the AI-power of Tensor cores to upscale the image to a higher resolution. For example, using DLSS enables users to render a scene at 1440p while displaying that scene at 4K, achieving the performance of the lower resolution, displayed at the higher.
Since initially launched, we have seen the technology get better and better as NVIDIA fine-tunes it and as the hardware behind the technology improves as well. In many cases, the usage of DLSS improved the quality of the gaming experience, in both frame-per-second and in the overall visual quality. In some cases, which you will see here in a moment, DLSS 2.0 radically changed the performance and clarity… and that was WITH RTX on.
While DLSS may have started out in a rough place out of the gate, NVIDIA’s dedication to continual improvement has made and will continue to make DLSS a go-to feature, pushing the limits of how games perform. And for the 30-series, especially, using DLSS, when available, is a solid choice.
Synthetics, Benchmarks, and Thermal Performance
Ok, we have talked about the size and power of the RTX 3090, this is the part where we’re going to see just how well it works. But, before we get into the numbers, here are the system specifications for our test bench:
- CPU: AMD Ryzen 7 3800X
- Cooler: Corsair Hydro Series H150i ELITE CAPELLIX (Closed loop cooler)
- RAM: Zadak SPARK 32 GB 3200MHz DDR4
- Motherboard: ASUS Prime X570-Pro
- GPU: NVIDIA RTX 3090 Founders Edition
- Storage: 250GB Patriot Scorch, 1TB WD_Black SN750, 1 TB Seagate FireCuda
- PSU: NZXT E850
- Case: Corsair 4000X
We have broken our testing up into a few different portions to see how the RTX 3090 performs in normal rasterization situations, how is handles real time raytracing, the impact of DLSS, and, because this is a Titan-class GPU, how it handles high memory tasks and workflow for content creation. Because of the shortage of the RTX 3080, we, too, have yet to get our hands on one, so a comparison between the two is forthcoming. Instead, we are going to look at some comparative numbers with the RTX 2080 Ti and RTX 2080, both Founders Editions, to show performance uplift between generations, specifically in RTX applications.
Let’s begin with rasterization.
Rasterization
It is not unexpected that we saw, on average, a ~39.667% uplift in performance between the 2080 Ti and the 3090 in normal rasterization situations, but an important data point in that number is that the average is brought down significantly by the difference in performance delta in Doom Eternal at ~12%. That is a staggering generational uplift in just rasterization.
Moving on to our synthetic tests, In 3DMark TimeSpy we saw a 46% performance uplift at 4K graphics scores. In order to completely strip the CPU completely out of the equation, we tested at 8K and what we saw was mind-blowing: a 68.76% difference in performance between the 2080 Ti and the 3090.
These scores are purely based on rasterization tests and represent a leap in both generation and class of GPU, so let’s see what happens when you add real-time raytracing to the equation.
Ray Tracing Performance
Within our tests, we saw an average uplift in RTX performance of 40.33% of the previous generation while testing in 4K. To take things a bit further, we also tested this in 1440p to see just how much of a performance gain there would be between generations where the CPU would have also had some impact on performance. Even here, we saw an average uplift of 29%. It is important to note a missing number from this chart: Battlefield V would not permit the 2080 Ti to use DLSS in 1440p.
We also ran a few synthetic benchmarks specifically aimed at heavy RTX performance. Both Bright Memory and Boundary show off gaming performance with heavy shadows, reflective surfaces, and use of lighting both with and without DLSS. As you can see from the chart, while running on the performance mode of DLSS, the RTX 3090 pushes past 4K60 in Bright Memory while falling just behind with the deficit from DSR.
At its inception, real-time ray tracing came out of the gate with a pretty rocky start. While it was visually stunning to look at, the practical application was very limited, especially without DLSS. However, as both DLSS and RTX have matured in the last two years the combination of features and performance take the technology out of being buzzwords and into being viable, high-performance options for gaming.
Future Feature: 8K Gaming
Forgive my exposition here for a moment, but I remember when the idea of 4K gaming was being bandied about. It became the aspirational “next step,” but 4K gaming is still a bit of an outlier in the average gaming situation considering the expense of technology to support it.
We have a bit of a similar story here, as the RTX 3090 is not the “everyman’s GPU,” nor are 8K displays with high refresh rates commonplace. We did run some tests, thanks to DSR and our numbers we received are interesting.
In games without the benefit of DLSS, such as Warhammer: Vermintide II, Far Cry 5, and Final Fantasy XIV, the RTX 3090 struggled to keep up. However, with the DLSS uplift, Wolfenstein: Youngblood and Control do the unthinkable: they actually average above 60 frames per second.
While we wait for this technology to be more widespread, it is fascinating to see these kinds of advancements.
Rendering Tests
To collect this data, ran a series of rendering benchmarks in Blender and in OTOY’s OctaneRender to show just how the RTX GPUs stack up to each other in project rendering. As impressive as the gaming scores are, we must remember: we have been told that this is a TITAN-class GPU. So, can it handle TITAN-class work?
The short answer is this: like a champ.
As we evaluate these numbers, keep in mind that we are evaluating a TITAN-class GPU against an enthusiast’s consumer-grade GPU. The 2080 Ti is no slouch, but these are memory intensive tests that we are running; it is at a clear disadvantage to the expanded memory, that memory’s clock speed, interface, and bandwidth, CUDA cores, and more that set the RTX 3090 firmly in the Titan camp.
As we ran our tests, we utilized NVIDIA’s ray tracing API, OptiX, to keep the tests as evenly matched as possible. While using OptiX, we saw a remarkable difference in performance in the RTX lineup across the board, but it even using the same API in rendering, there was a clear advantage that the 3090 had. At its most extreme case, in OctaneRender, our final render was 33-seconds on the 3090 as compared to the 2080 Ti’s 7-minutes and 40 seconds.
For a point of reference on the difference OptiX makes within these processes: the RTX 3090 took ~6:10 to complete Blender’s BMW GPU benchmark without OptiX. However, with OptiX, the same render took 12-seconds with the RTX 3090 and 22-seconds with the 2080 Ti. This type of performance carried through some of our other benchmarks in Blender as well.
In DiVinci Resolve, we noticed an improvement in the overall editing process. While the 3090 is capable of crunching through 8K R3D RAW footage with ease, it absolutely crushed the ten minutes of 4K 60FPS gameplay footage I used to put it through its paces. There was a smoothness to the monitoring when applying extra processes like color correction or applying transitions. While it didn’t cut down on the finish export of time, the process of getting there was great!
Thermal Performance and Acoustics
While talking about design earlier in this review, I talked about the design of the cooling array on this GPU as a push-pull system that marries the best parts of a blower-style GPU with that of multi-fan setups. The result is a GPU that is quiet under load while keeping cool.
In the tests that I ran, I noticed a few consistent details: the first is that the GPU stayed in the low registers of 70C while performing, pushing heat far away from the core and into the heatsink fins. The second detail that I noticed is that the RTX 3090 Founders Edition seems to have a long fan curve. Where some GPUs will throttle fans to decrease heat, causing spikes in fan noise and performance. That did not occur during my testing. The cooling system worked steadily with the temperature to dissipate them over time without the heat getting out of control or being immediately slammed down.
Final Thoughts
While we have had “Ti” versions of GTX and RTX GPUs from NVIDIA that have danced around the line between consumer use and profession application, what we see with the performance of the RTX 3090 is the People’s TITAN. This means that, yes, it is going to have unrivaled performance in gaming, but it is meant to crush workloads that demand significant computational power.
With that considered, the value proposition of price-to-performance between the RTX 3080 and RTX 3090 is skewed, from all reports that I have seen, toward the RTX 3080 being the better value for dollar for the average user. But, the Titan-series was never the card for the average user; the TITAN has always been the GPU for professional application with the added benefit of being the pinnacle example of that generation’s architecture as gaming is concerned.
So, NVIDIA was correct in saying that the RTX 3090, in all of its rechonkulous glory, delivers the “ultimate gaming experience” from a performance standpoint. Is it outside the reach of the average gamer? Definitely. However, when considering value we have to look at the whole package.
From a pricing standpoint, if we were only looking at gaming performance, the difference between the RTX 3090 and the RTX 2080 Ti is just as significant as the reported differences between the RTX 2080 and the RTX 3080 in third party reviews. Add on top of that the TITAN-class memory bandwidth, clock, and interface, the 3rd generation Tensor core improvements, increased RT cores, as well as the sheer amount of GDDR6X… and so much more, what you end up with is a GPU that wildly outperform anything on the market while landing around $300 above the previous generation’s enthusiast grade GPU. If we were to compare the pricing TITAN to TITAN-in-all-but-name for perspective’s sake, the TITAN RTX costs $2499…
If you are gamer whose battle station doubles as your workstation and you need girthy computational power for video editing, rendering, scientific computation, or you just have a stupid amount of money laying around burning a hole in your pocket, the RTX 3090 is an unparalleled GPU in a class all by itself as the People’s TITAN by blurring the lines between enthusiast gamer and prosumer needs.
Just make sure it will fit in your PC case first!
