Two long years after the GeForce RTX 20-series kicked off ray tracing in earnest, Nvidia revealed their successors on Tuesday, and the wait was worthwhile: The GeForce RTX 30-series graphics cards look monstrous, spearheaded by a fully loaded, 8K-ready $1,400 GeForce RTX 3090 “BFGPU” that revives the dormant xx90-class brand previously reserved for dual-chip beasts. The $700 GeForce RTX 3080 and $500 GeForce RTX 3070 were also announced, and all of them will include a radical new RTX IO technology designed to rival the ultra-fast storage technologies coming to the next-gen consoles.
CEO Jensen Huang calls this the “greatest generational leap ever” for Nvidia’s GeForce cards. Hot damn.
Oh, and ray tracing is coming to Fortnite. And new Nvidia Reflex software will make you much more deadly in esports games. And Nvidia Broadcast will use the AI tensor cores inside these new GPUs to spruce up the audios and visuals of your game streams in real time.
Phew. Let’s dig in.
GeForce RTX 3090, RTX 3080, and RTX 3070 specs and features
The GeForce RTX 3090 is so brimming with new technology that Nvidia need to create a new 12-pin power connector and a radically overhauled cooling system just to stuff it all in. The GeForce RTX 3080 and GeForce RTX 3070 aren’t quite as ludicrous, but they nevertheless look poised to push frame rates far beyond their predecessors in both traditional and ray traced games—quelling a key criticism of the older RTX 20-series GPUs, which focused more on introducing ray tracing than advancing pure gaming performance.
Bottom line: These suckers will be fast.
Let’s take a look at the light, high-level specifications that Nvidia revealed before diving into the nitty-gritty. Take the performance comparisons with the caveat that they likely cite best-case performance scenarios with ray tracing and DLSS enabled, rather than just traditional gaming frame rates.
GeForce RTX 3090 specs:
- 8K 60fps gameplay with DLSS
- 24 GDDR6X memory
- 3-slot dual axle through-flow design
- 30 degrees cooler than RTX Titan design
- 36 shader teraflops
- 285 tensor TFLOPS
- Launching September 24
GeForce RTX 3080 specs:
- 2X performance of RTX 2080
- 10GB GDDR6x memory
- 30 shader TFLOPS
- 58 ray tracing TFLOPS
- 238 tensor TFLOPS
- Launching September 17
GeForce RTX 3070 specs:
- Faster than RTX 2080 Ti
- 8GB GDDR6 (not GDDR6x)
- 20 shader TFLOPS
- 40 RT TFLOPS
- 163 tensor TFLOPS
- Launching October
Onto the nitty-gritty.
The GeForce RTX 30-series builds on the foundation set by its predecessors with more advanced RT cores and third-generation tensor cores inside of its new-look “Ampere” GPU architecture. That should improve ray tracing and Deep Learning Super Sampling performance, respectively. But Nvidia’s new chips also make the jump to 8nm process technology, moving up from the 12nm process used in the RTX 20-series “Turing” GPUs, which itself was an optimized iteration of the 16nm process used in the GTX 10-series.
Without getting too technical, this is the first major process node leap in a while, and it should provide a big uplift in traditional gaming performance. The last time Nvidia moved up a node, the GTX 10-series blew away its predecessors, and leaping to the 8nm process gave AMD’s rival Radeon RX 5700 graphics cards a huge boost in both performance and power efficiency.
Nvidia says its new Ampere GPUs offer twice the energy efficiency of Turing, with shader cores that are 2.7X faster than the ones inside the GeForce RTX 20-series. The revamped RT and tensor cores will be 1.7X and 2.7X faster than their predecessors, respectively, which should help deliver huge performance boosts in games with ray tracing and/or DLSS 2.0 enabled, as shown in the chart below.
Nvidia didn’t give any concrete numbers for traditional gaming performance, but the company says the GeForce RTX 3070 will spit out frames faster than the previous RTX 2080 Ti flagship at 4K in “average performance across multiple popular graphics intensive games,” while the GeForce RTX 3080 will be significantly faster. It’s unknown whether this comparison includes games with ray tracing and DLSS enabled, but we’d guess so. The RTX 3090, meanwhile, was billed as the world’s first 8K graphics card with DLSS enabled, so it’s sure to be a monster.
The GeForce RTX 30-series’ secret sauce isn’t limited to GPU improvements alone. The GeForce RTX 3080 and RTX 3090 will include cutting-edge GDDR6x memory that Nvidia co-developed with Micron, and Huang claims it can transmits twice as much data as traditional GDDR6 memory in the same amount of time. (The GeForce RTX 3070 still uses GDDR6.) The GeForce RTX 3090 packs a whopping 24GB of GDDR6x, while the RTX 3080 includes 10GB, and Nvidia plans to put all those gigs to a newfound use: Improving game loading times.
Nvidia’s new RTX IO technology puts the speed of PCIe 4.0 storage to more efficient use, routing some loading data over the PCIe 4.0 connection of the GeForce RTX 30-series GPUs to tap into their blazing-fast onboard memory, rather than routing all that information through the CPU first. Huang didn’t go into much detail, but claims the technology will offer 100X faster throughput and 20X lower CPU utilization than traditional methods, which—if true—will no doubt be a massive boon when you’re loading worlds using massive 4K or 8K textures especially.
Whether or not developers need to explicitly support RTX IO remains to be seen, but the technology leverages the Microsoft DirectStorage API for Windows—the same technology underpinning the Xbox Series X’s ultra-fast Velocity Architecture. Fingers crossed this is as awesome as it seems, as it could be just as much as a game changer for enthusiast-class PC gamers as ray tracing itself if it is.
All three of the graphics cards include an “ultra-compact” PCB with 18 phases of power, along with Nvidia’s redesigned cooling solution. The new cooler still packs dual fans, but in independent push/pull configurations with different aims. The one closer to the rear of the system sucks in air and blows it out of your graphics card’s I/O bracket, following traditional methods. But the one in the front leverages the shorter PCB to intake air and then expel it right back out the top of the graphics card, letting your case’s rear fan outtake handle more of the thermal load.
Nvidia didn’t specific power requirements for these cards, but given all the engineering efforts required to invent this cooler, they must be hefty
Nvidia Reflex and Nvidia Broadcast
The new graphics cards are being bolstered by several gameplay-enhancing software improvements.
Nvidia CEO Jensen Huang kicked off the event by announcing that real-time ray tracing and DLSS 2.0 is coming to Fortnite. Those technologies debuted in Minecraft earlier this year to startling effect, so now Nvidia has the two most-played games in the world onboard the RTX bandwagon.
But looks aren’t everything. In the blink-and-you’ll miss it world of esports, every frame matters, and a delay of even a few milliseconds can mean the difference between death and victory. To help you in your battles, the newly revealed Nvidia Reflex software optimizes the rendering pipeline in games to improve latency by up to a whopping 50 percent. If it works as promised, this could give you a massive advantage versus your competition. Call of Duty Warzone, Valorant, Fortnite, and Destiny 2 will be among the first games to support Nvidia Reflex when it launches this month.
Finally, this streaming thing isn’t going away, and Nvidia Broadcast brings the power of the GeForce RTX 30-series GPU’s advanced tensor cores to bear on real-time audio and video processing. The software works on any RTX graphics card—not just these new ones—and includes the magical RTX Voice noise-cancellation feature, along with adjustable background blur, greenscreen-style background replacement, and automatic head-tracking effects for your video feed. Even the greenest Twitch streamers can get a professional-seeming setup up and running quickly with that toolkit in their arsenal. It’s also launching in September.