Design and Features
The RTX 3080 packs an impressive spec-sheet. Compared to last generation’s RTX 2080 Super, it features more than double the CUDA core count with a whopping 8704 cores and 10GB of 320-bit GDDR6X VRAM. GDDR6X is the latest and fastest video RAM available today with a maximum bandwidth of 19 Gbps, which is a substantial bump from Turing’s 14 Gbps. The card is fast with a boost clock of 1.71 GHz, though Nvidia is claiming it’s also power-efficient, offering 1.9x performance per watt over the 20-series.
It would be easy to draw a line to the expanded CUDA core count and Nvidia’s claim of “up to 2x performance,” but things aren’t quite so simple. With Ampere, CUDA core count doesn’t scale 1:1 with performance, as roughly half of those cores can shift between FP32 shader cores and INT32 integer cores depending on what’s needed at the time. How game engines and applications utilize these cores vary, so while it’s safe to expect improved performance, every game won’t suddenly run twice as fast.
The Founders Edition also features Nvidia’s latest iteration of the RTX triple-processor system. The RTX engine is composed of three parts: the programmable shader, which is responsible for standard rasterization duties (normal game rendering); the RT Core, which handles ray tracing; and the Tensor Core, which powers the AI side of RTX, enabling features like DLSS and RTX Voice.
Each of these systems has received massive throughout boosts with 2.7x improvement to the shader, 1.7x to the RT Core, and 2.7x to the tensor core. The sheer teraflop count compared to the RTX 20-series is staggering:
- Programmable Shader: 30 TFLOPs (previously 11 TFLOPs)
- RT Core: 58 RT-TFLOPs (previously 34 RT-TFLOPs)
- Tensor Core: 238 Tensor-TFLOPs (previously 89 Tensor-TFLOPs)
As a quick point of comparison, the upcoming Xbox Series X will feature 12 TFLOPs of performance. Compared to the 30 TFLOPs offered by the RTX 3080’s Programmable Shader, the gap is vast. (That said, the Series X is using a different graphics architecture, so comparisons aren’t necessarily one to one.)
The RTX 3080 Founders Edition also brings with it Nvidia’s new dual-axial cooler. Compared to the last generation of Founders Edition coolers, it feels like a giant heatsink where the last one felt more or less like a polished metal block. The PCB has been shrunk down while the overall size has remained similar, allowing for a greater surface area for heat to disperse. Embedded in the cooler are two fans which work in tandem with the natural airflow path for most mid-tower cases. The fan closest to the front of the card pulls air through its exposed fin-stack and ejects it out the back of the case like a blower-style cooler. The rear fan is positioned on the other side of the graphics card, pulling air up and through to be exhausted through the path of the CPU. Nvidia claims this design is up to 20C cooler and 3x quieter than last generation, and while it is cooler and quieter, I didn’t quite see that caliber of result. Still, it’s a unique design that works well.
Positioned at an angle on the side of the cooler is Nvidia’s new 12-pin power connector. Thankfully, you won’t need to buy a new PSU just to accommodate the new connector: Nvidia includes an adapter in the box to convert two 8-pin headers into the single, smaller connection. That said, I would much rather Nvidia had stayed with the normal, dual 8-pin connectors as the adapter just doesn’t look very good dangling from the side of the card, especially if you have custom cables.
If you’re running even a mid-tier power supply, though, you may still want to upgrade. The RTX 3080 has a thirst for power with a rated TDP of 320 watts. That’s up from 250 watts on the 2080 Super and 2080 Ti. Along with that bump comes the recommendation for a minimum 750 watt power supply and that each 8-pin connector should be a separate wire coming from the PSU – no split ends.
When it comes to connectivity, the card features three DisplayPort 1.4 connections and an HDMI 2.1 port. Like last generation, the RTX 3080 FE is capable of outputting in wide-gamut HDR content and supports Display Stream Compression for high bandwidth throughput to supported monitors. Those outputs can push a maximum resolution of 7680×4320 across four monitors.
RTX IO, RTX Reflex, and RTX Broadcast
Along with the card itself, the announcement of the RTX 30-series brought with it several additional announcements. The foremost among these is RTX IO, which allows GPU-based decompression directly from the SSD, which will dramatically expand its bandwidth. Like the PlayStation 5, this has the promise of near-instantaneous load times in games, but since it requires developer integration, I wasn’t able to test it for this review.
Other core features announced were RTX Reflex and RTX Broadcast. RTX Reflex is targeted squarely at esports gamers, promising greatly reduced system latency when running games at 4K. In the past, running games at high resolution, especially at high graphics settings would result in higher system latency, making the game feel less responsive and potentially impacting your aim. RTX Reflex monitors and optimizes frame rendering to reduce latency, particularly in GPU bound scenarios.
To test this program in a GPU bound scenario, I swapped out the RTX 3080 for the GTX 1660 Super I used earlier in testing and loaded up Valorant. On my 4K monitor, I was able to reduce my system latency by nearly 50% by enabling the Reflex+Boost setting. I’m not a competitive esports player, so I expect much of this improvement is likely lost on me (we’re sub 15 ms here), but it did feel slightly faster even to me. If you’re the kind of gamer who needs every edge, there is no reason not to take advantage of RTX Reflex, especially since it’s either currently available in or planned for major first-person shooters like Call of Duty: Modern Warfare, Fortnite, and Apex Legends.
RTX Broadcast, on the other hand, is targeted at creatives and streamers versus competitive gamers. Using the AI functionality built into each RTX card, you can apply settings like background noise removal and virtual background effects. Of course, we’ve already seen this functionality in other software, but the AI enhancement offers each a much higher fidelity than I’ve seen before. Nvidia demonstrated the AI noise removal filtering out the sound of a vacuum in the same room as a streamer while having minimal impact on the quality of his voice. Likewise, the AI-powered background effects seemed equally high-fidelity, effectively cutting the streamer out from their environment with only minor fringing. This is especially exciting if your streaming space is too constrained for a real green screen.
But enough with the background, let’s get into the benchmarks.
The technology is impressive, but what really matters most is how it performs under pressure. To test the card, I ran it through a series of synthetic and in-game benchmarks. All tests were performed at ultra settings to see how well it handles the most stressful scenarios. Following these tests, I also ran a number of specific tests to see what kind of ray tracing improvements the 3080 offers over its predecessor.
Do note that we chose to include the RTX 2080 Super over the original RTX 2080 since it is the most recent and relevant model. Given that the Super already offers an FPS advantage over the original, the lead over the OG model would be even greater.
Let’s start with the synthetic benchmarks, focusing on rasterization:
Nvidia GeForce RTX 3080 Synthetic Benchmarks
The RTX 3080 offers a significant lead over each other card we tested against, peaking with a 56% improvement over the 2080 Super in 3DMark Fire Strike Ultra and 47% in Unigine Heaven. These are big jumps, but what’s even more striking is the lead over the 2080 Ti, which topped out at 28% in Fire Strike and 18% in Heaven.
On to game performance:
Nvidia GeForce RTX 3080 Gaming Benchmarks
Simply put: the RTX 3080 Founders Edition is a rasterization powerhouse. It clearly led every other card I tested against – which isn’t a surprise. What did surprise me was by how much. Compared to the 2080 Super, it delivered 29% to 51% higher FPS at 1080p, for an average 44% FPS gain. The improvements are even more striking at 4K, which ranged from 53% to 67% improvement, for an average of 57% higher FPS.
Even pitted against last generation’s flagship, the RTX 2080 Ti, the RTX 3080 won out, though the margins were expectedly slimmer. FPS gains ranged from 6% to 24% at 1080p and 17% to 34% at 4K – averaging to FPS boosts of 18% and 26%, respectively. Given the relative pricing of the two cards, $699 versus $1199, the RTX 3080 feels like an incredible value.
I also spent some time testing other games specifically in comparison to the 2080 Super. Since it’s such a direct counterpart to the RTX 3080, it makes a great measure to illustrate the performance gains with this new generation. These tests were run at 4K on ultra settings with ray tracing and DLSS enabled:
These are impressive results that illustrate the leap the RTX 3080 brings to the table. This is a 4K card that absolutely delivers on the promise of 4K gaming, ray tracing included. It also marks a substantial improvement over the uplift we found in the jump from the 10- to 20-series.
It’s important to note here that the gains over last-gen will vary depending on the game you’re playing, the settings used, and how well optimized the game engine is to leverage the RTX 3080’s improved processing capabilities. The sampling of titles and tests I conducted show a meaningful leap over the 20-series, especially if you’re playing at 4K.
Ray Tracing Performance
Rasterization is only one piece of the equation with the RTX series: the other element is, of course, ray tracing performance. With both next-gen consoles offering hardware-level ray tracing, I was curious what kind of improvements the RTX 3080 would bring in anticipation of games to come. Like with the normal rendering tests, I began with a pair of synthetics, specifically targeting ray tracing performance. Since I was looking for a clear generational improvement, I focused on the RTX 3080 and RTX 2080 Super Founders Editions.
The tests I used were Boundary, from Surgical Scalpels, and Bright Memory, from FYQD Studio – both extremely demanding system crushers, even with DLSS enabled. Each test ran at 4K resolution. In Boundary, the RTX 3080 offered an 81% performance improvement over the RTX 2080 Super. In Bright Memory, the uplift was 67%.
When it came to testing improvements in actual games, I was less worried about overall FPS and more concerned with how RTX/DLSS impacted FPS. To accomplish this, I tested four games at 4K with RTX and DLSS (Quality Preset) on and off and recorded the FPS over multiple runs to verify my results. Wolfenstein: Youngblood, Metro Exodus, and Minecraft RTX each had built-in benchmarks (the latter created by Nvidia in the Portal Pioneers world). For Control, I used a repeatable path through the Bureau with plentiful ray traced reflections, taking care to make as identical a run as possible with each pass, and recorded the results with Nvidia FrameView.
In the chart below, you can see the performance of each game, first in FPS and then in percentages to clearly see the impact of enabling RTX and DLSS.
As you can tell from this chart, it’s clear that the processing enhancements on the RTX 3080 mean that 4K, ray-traced gaming at 60+ FPS is a reality. The combination of improved rasterization and enhancements to the RT and Tensor cores give it a significant advantage over the RTX 2080 Super in virtually every way. But is ray tracing less impactful? Here are the performance hits in percentage form:
With RTX and DLSS enabled, each game experienced a performance drop, but the impact is generally smaller with the RTX 3080. The exception here is Metro Exodus, which showed less of a drop than the RTX 2080 Super. This limited sampling is difficult to draw hard conclusions from, but it does look like, generally, the improved RT and Tensor cores are boosting ray tracing performance in particular. Interestingly, Control performs better with RTX and DLSS on than with both disabled across multiple test passes.
That said, the ray tracing improvement on both Wolfenstein and Control is smaller than I hoped to find and Metro went the opposite direction with the 2080 Super maintaining a slight edge. The RTX 3080 is fully capable of running each with RTX on at high frame rates (Youngblood: 126 FPS Control: 69 FPS, Metro: 64 FPS), but it’s going to take more testing with a wider array of games to draw any hard conclusions on the overall percentage improvement the RTX 3080 offers over the 2080 Super.
Within this sample, Minecraft is the outlier with 30.4% better performance while Control and Wolfenstein: Youngblood are only improved by 4.5% and 3.1% respectively. Will more games perform closer to Minecraft or Metro? Time will have to tell.
Since its inception, DLSS, short for Deep Learning Super Sampling, has evolved substantially. The technology leverages the power of AI to intelligently upscale games, allowing an RTX graphics card to render games at a lower resolution and then use its tensor cores to upscale, theoretically, without a decrease in clarity. Gaming at 4K with DLSS might only require the processing power of 1440p, for example, giving you 1440p-like frame rates with a 4K image. When the technology debuted, there were notable problems with the softness of upscaled images and much of the excitement was quelled out of the gate. Since then, Nvidia has been quietly refining the technology and it is now able to deliver DLSS graphics that are crisper than native 4K.
This screenshot from the beginning of Wolfenstein: Youngblood demonstrates exactly that. Looking closely at the text and lines on the box, as well as the textures on the sides of the ammo crates, DLSS provides a slightly sharper image while running a full 39 FPS faster. This is “better than 4K” fidelity with 1440p processing. It’s incredibly impressive. Simply put, there is no reason not to use DLSS.
Thermal and Acoustic Performance
Nvidia made big claims about its new cooler design. In its marketing material, the company noted that it was “up to” three times quieter and 20C cooler. For gaming and benchmarking, the card was both cool and quiet, but didn’t quite live up to those numbers. Inside my Lian Li Lan Cool 2 case with a closed glass side, temperatures peaked at 75C on a hot day where the ambient temperature was ~78F/26C and 69C on a cooler day with ambient temps of 70F/21C. This is quite a difference and highlights how dependent the RTX 3080’s thermal solution is on the presence of cool air. The RTX 2080 Super, by comparison, peaked at 79C and the RTX 2080 Ti at 82C on the same warm day. It’s also worth noting that our test system featured the radiator mounted on the front of the case with fans blowing in through the radiator. The amount of cool air entering the case is less than if the radiator were top mounted (which, unfortunately, the Lan Cool 2 would not allow). With a custom fan curve and top-mounted radiator, I’m confident these temps could be lowered even further.
The card was also very quiet. Without a decibel meter, I couldn’t test the exact noise reduction, but more than once, I opened the side panel to make sure everything was running correctly. After years of noisey stock coolers, the dual-axial design is a major improvement for noise, even under load.
The Nvidia GeForce RTX 3080 Founders Edition is available direct from Nvidia with an MSRP of $699, starting on September 17.