Nearly two years after bringing ray tracing to the mainstream with its Turing cards, Nvidia is hard at work on the successor to its popular RTX 2080 flagship.
The RTX 3000 series will use a new Ampere architecture that debuted on Nvidia’s DGX A100 supercomputers, marking the first time that the company will use a 7nm manufacturing process for its GPU.
Like Turing before it, the Ampere-powered RTX 3000 card is expected to bring even better graphics performance to gamers and workstations, but the most notable upgrade for PC enthusiasts will be bottleneck-free ray-tracing capabilities. Here’s everything we know about it so far.
Nvidia has been tight-lipped about the launch of its RTX 3080 graphics cards amid extensive leaks recently. The company only recently teased that a big supernova-level announcement is planned for September 1 at 9 a.m. PT. The company revealed its #UltimateCountdown campaign on Twitter, along with the text “21 days. 21 years.”
Given that many industry insiders had already been speculating that the RTX 3080 could launch by as early as fall 2020, Nvidia’s Twitter post — albeit mysterious — does suggest that the launch could happen at this event. With the RTX 3080 launching 21 years after Nvidia debuted the GeForce 256, or what the company dubs as the world’s first consumer graphics cards, the 21 years part of Nvidia’s countdown campaign makes sense.
Previously, it’s been reported that the Ampere cards are currently undergoing a Design Validation Test, or DVT, and mass production could begin in August. If all goes well, an announcement could happen in September, according to TweakTown‘s revised launch timeline estimate.
Historically, Nvidia staggers the launch of a new consumer GPU family, and we expect this will be the case again. This means that the flagship and top-tier cards — such as the RTX 3080 and RTX 3080 Ti, the latter of which potentially could be renamed under the RTX Titan branding — could debut first, with more budget-friendly options, including the RTX 3070 and 3060, arriving at a later date that could stretch into early 2021.
This timeline would match what Nvidia used for the launch of the RTX 2080, which was announced in August two years ago.
Nvidia’s next-generation RTX lineup represents the premium segment of gaming and workstation graphics cards, and the new RTX 3000 series is expected to command a hefty price. The most recent leak suggests that the RTX 3090 Founders Edition — Nvidia’s flagship card for the RTX 3000 range — could be priced as high as $1,500.
For reference, Nvidia’s RTX 2080 Ti retails for $1,199 and the RTX 2080 Super costs $699. The RTX 3080 likely will be priced at the premium end of the spectrum.
If you balk at the price premium, Nvidia will have other cards in the RTX 3000 series that will cost less. The non-Founders Edition, for example, is expected to cost $1,400, or $100 less than the flagship, and it’s widely believed that Nvidia will also debut mid-range and entry-level cards in this range to cover a wide spectrum of price points. It’s unclear if the midrange and entry-level offerings will launch alongside the high-end cards, or if Nvidia will repeat what it had done previously with a staggered launch.
In the past, there was a big price jump moving between Nvidia graphics generations — the Turing cards were more expensive than the Pascal cards that they replaced — and recently leaked pricing fort eh RTX 3090 suggests that this will again be the case for the latest GeForce generation. However, there are rumors that Nvidia may be moving manufacturing from TSMC to Samsung. This move could give Nvidia access to Samsung’s 7nm extreme ultraviolet, or EUV, process, which could reduce manufacturing costs. The accuracy of that rumor is unclear, however.
Gamers with a smaller budget for an upgrade shouldn’t be discouraged. The RTX 2060 currently starts at $299, and the upgraded RTX 3060 likely will be similarly priced. The midrange RTX 2070 is priced at $499.
These lower-priced cards will help Nvidia compete against AMD’s forthcoming lineup of RDNA 2 graphics. Additionally, a leaked benchmark from earlier this year showed an unknown Radeon GPU outperforming Nvidia’s RTX 2080 Ti by 17%. More competitive performance from its rival may force Nvidia to engage in a price war with the release of the RTX 3080.
Nvidia’s Ampere architecture will appear in multiple varieties of graphics cards. The first such card was unveiled in May at Nvidia’s online Graphics Technology Conference in the form of the A100 for data centers and the DGX A100 supercomputer. The company claimed that this is the largest 7nm chip ever built, saying it was designed for better A.I. performance.
On the consumer side, the RTX 3000 series will likely use the same 7nm architecture that debuted on the data center version, which would bring better power efficiency. An earlier Taipei Times report suggested that this new architecture could result in up to a 50% uplift in GPU performance, while at the same time reducing power consumption by one-half compared to Turing.
As we are nearing the launch of Nvidia’s latest flagship GeForce cards, we are seeing more leaks on Ampere. Most recently, a leaked UserBenchmark post dated August 15 revealed that the card should be clocked at 2.1GHz. Prior leaks suggested that the cards will be clocked between 1.7 to 1.8 GHz, and boost speeds can reach 1.9 GHz. Custom cards could even go up to 2.1 GHz with better power delivery or more optimized thermal management.
The benchmark results have since been deleted, but not before Wccftech posted screenshots of the results, revealing that the test was driven by an unreleased 455.90 driver.
The Device ID of the card has been revealed as 2206, and Twitter user Rogame claimed that there should be at minimum three separate IDs for three different cards in this generation. The RTX 3080 was spotted with 10GB of VRAM running 19 Gbps GDDR6X memory at 4750 MHz across a 320-bit bus interface, giving the GPU a 760 GB/s memory bandwidth, or roughly a 53% boost from Nvidia’s current GeForce RTX 2080 Super graphics.
While earlier leaks suggested that the RTX 3080’s memory could be in the 10GB range, newer leaks indicate that Nvidia could double the memory to 20GB of GDDR6X, though none of this has been confirmed. Rogame‘s recent PCI ID leak suggests that memory allocation for the RTX 2080 Ti should be 12GB, with the base RTX 3080 shipping with 10GB. The upgraded RTX 3090 — which could fall under the Titan branding, is expected with 24GB. The RTX 3080 is said to be given a GA102 model number and SKU of GA102-200-KD-A1.
YouTube channel Moore’s Law is Dead reported that the RTX 3080 should put Nvidia in the performance lead once again, as it’s capable of 21 teraflops of performance, or nine more teraflops than what is possible on the Xbox Series X. When it comes to gaming, Nvidia’s RTX 3000 series cards will show that PC gaming still commands a performance lead over consoles.
A separate leak from Twitter user KittyCorgi, however, shows the 3080 series bearing the GA103 model number. Regardless of how they’re labeled, the performance of these cards isn’t contested.
Hardwareleaks posted a leaked 3DMark Time Spy score of what is believed to have been obtained with the RTX 3080 Ti. The score of 18,257 points represents a 30% performance improvement over the RTX 2080 Ti. In a recent tweet, @kkatcorgi suggested a 20% generational performance uplift, though it wasn’t clear which benchmarking method or utility was used to obtain this figure. This could in turn lead to the 3080 Ti gaining 40% better graphics performance than the 2080 Ti, according to Notebookcheck, which could allow the new card to render games in 4K at 60fps with ease.
For comparison, earlier leaked benchmarks for AMD’s Big Navi GPU, which uses the company’s RDNA 2 architecture, suggested that the card performs just 10% to 15% better than Nvidia’s RTX 2080 Ti. If accurate, this would mean that the RTX 3080 Ti could outperform AMD’s next-gen graphics for serious gamers, and place Big Navi’s performance more in line with the non-Ti variant of the RTX 3080.
While the generational improvement in performance is expected for a new GPU family, the RTX 3080 Ti is also expected to perform significantly better at ray tracing, the hallmark feature of the RTX family, and an area of struggle for the current RTX 2000 series.
Ampere’s improvements with artificial intelligence and DLSS should help improve ray tracing performance, especially when games are rendered at higher resolutions. The card is also expected to come with 108 ray tracing acceleration cores, more than double the 48 ray tracing acceleration cores supported by the RTX 2080 Super.
Along with the card’s new double-sided PC board design, Nvidia is rumored to use a new traversal co-processor to help improve ray tracing performance. This co-processor would be the GPU’s ray tracing chip, according to Tweak Town, and could serve to unite the RT cores under one silicon.
It’s unclear how this architecture will play out, and if Nvidia will offload all or some of the ray-tracing demands to this traversal co-processor, but gamers should be able to expect much better performance in games with ray tracing enabled, especially at enhanced game settings in high resolution.
Though there was been conflicting information about clock speeds and memory in recent weeks, an earlier leak showed the card could come with a boost clock of only 1,935MHz. At this point, it’s still unclear if this will change or if Nvidia has made enough improvements in efficiency that a higher clock speed isn’t required.
An earlier leaked specs sheet for the RTX 3080 Ti posted by Twitter user CyberCatPunk revealed that the card will come with 5,376 CUDA cores and support 12GB GDDR6 memory running at 18 bps. The TDP for the card is stated as 320 watts.
The card is said to be PCIe 4.0-complaint and will feature three DisplayPort 2.0 ports, one HDMI 2.1, and one USB-C port.
The base RTX 3080 is expected to feature 4,352 CUDA cores, similar to the RTX 2080 Ti, and feature a total of 68 SMs, according to a recent Wccftech report.
For gamers, though, the killer feature of Ampere-based GPUs will be support for next-generation deep learning super sampling DLSS 3.0 specifications for any game that is rendered using TAA, or temporal anti-aliasing, techniques. The implementation of DLSS 3.0 support still requires Game Ready drivers, but it will still be easier for developers to bring DLSS into their game with some code work involved. For gamers, this will result in increased performance in games, and Nvidia is rumored to enable DLSS 3.0 out of the box by default, according to the Moore’s Law is Dead YouTube channel.
In addition to the flagship 3080 series, a GA104 was also leaked, which could pave the way for the mid-range GeForce RTX 3070 and RTX 3060 series release. According to a performance chart posted by Chiphell, the RTX 3000 series should include the GeForce RTX 3090, GeForce RTX 3080, GeForce RTX 2070 Ti, GeForce RTX 3070, and GeForce RTX 3060.
In terms of the card’s overall package, the RTX 3080 could debut with a new and controversial design. Leaked images depict a new cooling system for the dual-fan design on the card. Rather than having both fans on the same side, the fans are placed on opposite sides. Although the fins that envelope the card likely serve to dissipate heat, online readers noted that this gives the RTX 3000 series a cooler-like appearance.
“In order to achieve this perfect airflow, we had to remove the constraints we had before,” Nvidia thermal architect David Haley said of the new design in a published video ahead of the card’s launch. “We have to change the PCB, we have to move the fans, we have to change the software stack that’s controlling the fans.”
In addition to fan-based cooling methods, Nvidia partners are also working on water cooling mechanism for the RTX 3000 series. Partner EK responded to a user query on Facebook stating that it will have water blocks “at or close to launch” for the new cards, Videocardz reported.
A more recent image leaked by Twitter user @GarnetSunset shows the new RTX 3090 flagship side by side with an older RTX 2080 card. It’s clear from the image that the new RTX 3090 card is significantly larger than the RTX 2080 card, which leaves us to believe that the card will use a large triple-slot design for the Founders Edition variant. This leaked image appears to confirm the new design and cooling mechanism previously leaked for the RTX 3000 series. This flagship card is expected to carry the SKU GA102-300-A1 GPU and come with 5,248 CUDA cores and 82 SMs, or a 20% increase in core counts from the RTX 2080 Ti, according to Wccftech.
Details about the new cards were revealed in a video published by Nvidia that highlights the upcoming RTX 3000 series ahead of the company’s official announcement. In that video, Nvidia revealed that the card will have a stronger mechanical structure and a low-profile spring for a back cover. Structural reinforcements could be essential, given the larger design of the RTX 3090, which could officially succeed the RTX Titan in Nvidia’s current lineup, and the expensive anticipated cost.
More recently leaked images reveal that Nvidia could be moving toward a new 12-pin PCIe power interface system for the high-end graphics card that’s capable of delivering up to 600 watts of power. Early rumors suggest that Nvidia’s flagship cards will require at least 400 watts of power. The new 12-pin interface would be used on Nvidia GPUs with PG142 board numbers, which is speculated for the higher-end 3000 series, including the RTX 3080 Ti/RTX 3090, RTX 3080, RTX 3070 Ti/RTX 3070, according to Wccftech.
A schematic on Chinese tech site Fcpowerup shows the power interface, which looks like a single connector with two stacked six-pin connectors joined together. Ahead of Nvidia’s GeForce RTX 3000 unveiling, the new connector interface was seemingly confirmed by accessory-maker Seasonic with its Nvidia 12-pin PCIe Molex Micro Fit 3.0 Connector.
According to HardwareLuxx, which obtained the Seasonic connector, the 12-pin design occupies roughly the same footprint as an eight-pin connector in a design that is much smaller than previously expected. The connector comes in a configuration with six +12V pins, four grounds, and two Sense pins which can be used to adjust the power delivery. On the box, Seasonic recommends that the connector be used with an 850W or more PSU, suggesting that the new RTX series will require more power.
The new 12-pin connector was also subsequently confirmed by Nvidia in a YouTube video, and the company noted that the shift to the new connector design was made to conserve space on the new board design. The new connector is described to be smaller than two eight-pin connectors stacked together, and Nvidia stated that it will make an adapter to allow owners to use an existing eight-pin cable on the new connector.
Previously, given the configuration of the pins, some in the tech community had speculated that gamers who may not want to shell out extra for the 12-pin connector can still join two six-pin connectors together. However, this solution would require an extra cable, which would not only make the inside of your tower look cluttered but could present new thermal challenges where airflow is concerned, especially given the higher power requirements of the card.
The most optimal solution, it appears, would be for people to also upgrade their power supplies to accommodate the new connector and satisfy the higher power requirements of Nvidia’s flagship cards. This, however, would add to the cost of upgrading an existing PC build. The publication also detailed a separate four-pin interface adjacent to the new 12-pin power connector, though the purpose of that connector is unknown at this time.
And with Nvidia officially launching a countdown clock, we won’t have too much longer to learn more about the oft-rumored and much-hyped RTX 3080. Industry insiders expect Nvidia to unveil three cards at this event, including the GeForce RTX 3080, an upgraded GeForce RTX 3080Ti or RTX 3090, and a next-generation Titan model.