Nvidia introduces GeForce RTX 30 series, headlined by the $699 RTX 3080

[Karlston]

Nvidia introduces GeForce RTX 30 series, headlined by the $699 RTX 3080

After months of leaks and rumors, Nvidia has finally announced the GeForce RTX 30 series of graphics cards, consisting of the GeForce RTX 3090, RTX 3080, and RTX 3070. The star of the show is, of course, the Nvidia GeForce RTX 3090, which is the highest-end card of the bunch.

 

All of the new cards feature a new design, with dual fans that push air out on one side and pull it in from the other side. The cooling system promises 56% more airflow, three times less noise, and 30% more efficiency. In terms of efficiency, Nvidia says that the new cards offer 1.9x the performance per watt compared to the previous generation.

Starting with the RTX 3090, it costs $1,499 and it's touted as a successor to the Titan cards. However, this time around it will be available from partners, in addition to Nvidia. Nvidia says this is the first graphics card that lets you play games at 8K and 60 frames per second. It boasts 36 Shader-TFLOPS, 69 RT-TFLOPS for ray tracing, and 285 Tensor-TFLOPS for AI-related tasks such as deep learning super-sampling (DLSS). It has 24GB of all-new GDDR6X memory, which Nvidia says is the fastest memory out there.

Nvidia compared the RTX 3090 to the Titan RTX more than, promising 50% more performance while being up to 10 times quieter.

 

The GeForce RTX 3080 is the card that's coming the soonest, and it should be the better value in general. It's also the one Nvidia is focusing on the most, since the RTX 3090 is aimed at a small subset of users, just like the Titan cards.

 

It features 30 Shader-TFLOPS, 58 RT-TFLOPS, and 238 Tensor-TFLOPS, and in the end, it promises to double the performance of the RTX 2080, and largely outpace the RTX 2080 Ti. It has the new GDDR6X memory, too, but it settles for 10GB of it. This GPU will be available September 17 starting $699, the same price as the RTX 2080 was at launch.

The last card announced today was the GeForce RTX 3070, and it's still pretty powerful, with 20 Shader-TFLOPS, 40 RT-FLOPS, and 163 Tensor-FLOPS, which means it's still faster than an RTX 2080 Ti. It has slower GDDR6 memory, but it does come significantly cheaper than the RTX 3080, at just $499, which is significantly lower than the similarly-powered RTX 2080 Ti.

just like with the previous generation, the GeForce RTX 3070 is arriving later than the higher-end models, and it's only set for October for now.

 

Aside from the new cards, Nvidia also introduced a few new technologies today. One of them is RTX IO, which aims to speed up data transfers between a computers SSD and the GPU memory. This is done by offloading the necessary work from the CPU cores to the GPU, and Nvidia says it can deliver up to 100 times faster speeds compared to accessing the data traditionally from an HDD, while lowering CPU usage by 20 times. This feature works in tandem with the DirectStorage feature for Windows.

Nvidia also announced a handful of new tools for gamers, streamers, and creators. For e-sports gamers, Nvidia Reflex promises to help reduce system latency in games, so reaction times are shortened to give players an edge. This feature is rolling out with a Game Ready driver update in September.

Also in September, Nvidia Broadcast will be available for streamers. Using AI, the tool will allow streamers to blur their background, replace it, or removing it entirely, among other things. The software can keep people in the frame by recognizing where they are or remove background noise to isolate the streamers voice. The tool will be available for RTX cards.

Nvidia also announced Omniverse Machinima, a tool that makes it easier to import models and motion capture data to create machinima, a type of art that uses game assets to recreate movies or other scenes. This tool will be available in beta in October.

With Nvidia's RTX technology now in its second-generation, the company is unlocking more power from its GPUs, and it'll be interesting to see how they perform in real life. We shouldn't have long to wait, considering the range-topping cards are launching in just a few weeks.

 

 

Nvidia introduces GeForce RTX 30 series, headlined by the $699 RTX 3080

 

[Karlston]

Behold, Nvidia’s 3080 GPU: 2x 2080 power, starting at $699 on Sept. 17

RTX 3070 will exceed RTX 2080 Ti at $499 in October. Three-slot RTX 3090 is $1,499.

First image of article image gallery. Please visit the source link to see all images.

 

Nvidia CEO Jensen Huang announced the RTX 3080 as Nvidia's "new flagship" GPU on Tuesday morning, confirming over a month of leaked rumors and card designs. This was met with a flurry of other hardware, app, and software partnership announcements, but arguably the biggest numbers out of today's event came in the form of performance and price.

 

Huang alleged that the RTX 3080 will "double" the performance of the RTX 2080 GPU while starting at $699, with hardware going on sale September 17. Next to that, Huang announced the RTX 3070, whose power will reportedly exceed the $1,200+ RTX 2080 Ti while starting at only $499.

 

After these, the three-slot RTX 3090 was announced without any aspirations of value, but with a September 24 launch starting at a whopping $1,499. This was followed by a sizzle-reel of fans playing recent RTX-enabled 3D games at 8K resolution and "60 frames per second."

"Pascal friends: It is safe to update"

The announcements came with a flurry of Nvidia-handled specs, particularly those couched in Nvidia's estimations of performance for each card, so you'll see in the above gallery how Huang's performance estimations might break down differently depending on the software in question. But in at least one case, the claim checks out. Though Huang claims that the RTX 3070 will exceed the 2080 Ti, one of the charts slots it as more of a "1440p, 60fps" card with RTX effects enabled top-to-bottom—but that's quite honestly the performance you can expect from a current RTX 2080 Ti. Its 4K/60fps threshold is easier to reach with either RTX effects turned down or DLSS turned way up.

 

The 45-minute presentation saw Huang repeatedly reference Nvidia's two-year publicity campaign about RTX technologies and how they have been a crucial pivot due to what Huang calls "the ultimate limits of rasterization approaching." That sales pitch, combined with the leap in GPU architecture on the jump from Turing to Ampere process, led Huang to make the following bold claim: "To all my Pascal gaming friends: It is safe to update now."

 

First image of article image gallery. Please visit the source link to see all images.

 

The Ampere process ushers in 28 billion transistors on its cards, up from Turing's 19.8 billion, and it also supports what Huang describes as "four-level pulse amplitude modulation G6X memory," which will be found in both the RTX 3080 (10GB) and the RTX 3090 (a whopping 24GB). The RTX 3070, meanwhile, will sport 8GB of GDDR6 memory, no "X" there. The CUDA core counts for each are as follows: 10,496 on the RTX 3090; 8,704 on the RTX 3080; and 5,888 on the RTX 3070. Those are all a significant leap from the 4,352 CUDA cores on the RTX 2080 Ti from late 2018.

 

We'll have to wait longer for a whitepaper breaking down other technical aspects of the RTX 3080 and 3090 ahead of their September launches. While we'd love to point to leaked specs posted on Monday from GPU manufacturer Gainward, these got the CUDA core counts incorrect, so we're going to hold momentarily on those numbers. We're also curious to see exactly how the RTX 3000 line's PCI-e 4.0 support will factor into performance, and whether anyone holding onto high-end Intel CPUs will suffer with a previous generation of motherboards that max out at PCI-e 3.0.

 

The rest of the event's information can be found in the second gallery, and this revolved around software that will leverage RTX-specific technologies. The most interesting of these is Nvidia RTX IO, which promises to streamline delivery of computing assets of all kinds straight to the GPU that might otherwise be routed to the CPU. (Nvidia would love for you to do this, since its Tensor cores sure want to blur the line for what kinds of computing is handled on the GPU side.)

 

Obviously, we want to know more—and to go hands-on with the results. (Or, in the case of VR on high-resolution headsets like Valve Index and HP Reverb G2, to go eyes-on with the results.) Hopefully, we'll have good news to share with Ars Technica readers on that front by the end of September.

 

 

 

Update, 2:15pm ET: Thanks to a deep-dive hands-on video from Digital Foundry's Richard Leadbetter, we can confirm exactly how the RTX 3080's 12-pin socket looks and works. The GPU will ship with a dual 8-pin adapter for those whose PSUs aren't equipped with a smaller, neater 12-pin cable.

 

Listing image by Nvidia

 

 

Behold, Nvidia’s 3080 GPU: 2x 2080 power, starting at $699 on Sept. 17

 

(To view the article's image galleries, please visit the above link)