NVIDIA officially announces the GeForce RTX 2080 Ti, RTX 2080 and RTX 2070



quadro RTXNow that almost all the details of the new GeForce generation based on the Turing architecture were known in advance, NVIDIA officially unveiled the new maps as part of a gambling event. At the beginning of the event NVIDIA referred again to the development stages of the past decades – from two line-connected pixels to the graphics engines we know from many AAA titles.

For the CEO and founder Jen-Hsun Huang, ray tracing is one of the most important development steps of recent years. The holy grail of graphic technology is designed to raise the lighting of objects to a level. Ray Tracing is about simulating the behavior of as many bundles as possible. Reflections, scattering, shading, absorption and (semi) transparent materials are just a few terms that describe the mechanics that are modeled by ray tracing.

20 years ago a supercomputer was needed to calculate an image of 512 x 512 pixels by means of ray tracing. Multiple PFLOPS of computing power would be needed today to apply ray tracing to an entire scene. It would take another ten years to get this computing power at the current pace of GPU development. That's why NVIDIA has developed RTX as a software interface that brings together classic screening and ray tracing.

NVIDIA calls this RTX hybrid representation. It is a combination of classic rationalization for representing the objects, where the lighting has to be recalculated every time. Global Illumination is a tool that has used various tricks so far to make the calculation of lighting as realistic as possible.

Per Ray Tracing calculates only the light rays that touch our eyes. NVIDIA uses deep learning networks and trains the driver or software interface with data from the cloud. This way a still impossible Ray Tracing is made possible again.

Turing as architecture and hardware

Thousands of engineers each spent hundreds of hours developing the Turing architecture. On the whole, NVIDIA has worked on it for ten years. According to NVIDIA, this is the biggest technological leap forward since the introduction of the CUDA core, which we always refer to as shader units.

The most important technical data for ray tracing can be summarized as follows:

  • 14 TFLOPS (floating-point calculations)
  • 14 TIPS (integer calculations)
  • RT cores with 10 Gigarays per second (1080 Ti makes 1.21 Gigarays per second)

This is the computing power required for the current approach of NVIDIA. With hardware based on the Pascal and Volta architecture, such calculations would not have been possible. NVIDIA provides some data for calculating the Star Wars demo:

  • 4x Tesla V100: 55 ms / 18 FPS
  • 1x GeForce RTX 2080 Ti: 45 ms / 22 FPS
  • 1x GeForce GTX 1080 Ti: 308 ms / 3 FPS

We will add this information in the coming hours for more information.

With the Turing architecture, NVIDIA also changes the behavior of the GPU when creating and calculating a frame. In addition to classical INT32 and FP32 shading, ray tracing and DNN (Deep Neural Network) calculations are now used. For a GeForce RTX 2080 Ti NVIDIA is delivered with a combined processing power of 78 TRTX-OPS (Tera RTX operations). NVIDIA Titan X should come on only 12 TRTX OPS.

NVIDIA must use different schedules for the different calculations to make optimal use of all areas of the GPU. As part of a deeper presentation of the Turing architecture, we expect further insights here.

The Tensor cores are also important. These also play a role in ray tracing, because the distracting performance of the Tensor Cores can be used to remove noise through a thorough algorithm. The calculation of several thousand or even millions of light rays takes time, even on modern hardware. The pending areas of the calculation are represented by a noise. But because the renderer does not know which objects are most important to a scene, he simply calculates the rays in a logical order. To reduce the waiting time for the finished image and remove noise more quickly, deep learning algorithms calculate the important parts of the image and calculate them faster.

The third component is the new RT cores. These can calculate the so-called Ray-Triangle Intersection, as well as a Bounding Volume Hierarchy (BVH). The BVH consists of a tree structure in which the geometric objects are arranged. Based on these assignments, collision detection or ray tracing can take place. Up to 10 Gigarays per second must be able to calculate the 576 RT cores of the Turing architecture.

Calculations are partially executed in the cloud

Despite all shaders, tensor and RT cores, NVIDIA can not perform a complete Ray Tracing calculation with the Turing hardware. The noise reduction on the RT cores that we have already mentioned, but the noise is only possible through a Neural Graphics Framework. The NVIDIA NGX trains various image processes on in-depth learning networks and sends the results to the driver of the graphics card.

This makes new anti-aliasing modes possible. Instead of a TAA on 4K, a DLSS (Deep Learning Super Sampling) can now be used. Of course, NVIDIA uses its own hardware in the form of the DGX servers to train hundreds of thousands or even thousands of Tesla V100s to train in-depth learning networks. Inferencing then takes place on the Turing hardware in the slot machine and is supported by the Tensor cores.

GeForce RTX 2080 Ti, RTX 2080 and RTX 2070

The technical data of the GeForce RTX 2080 Ti, RTX 2080 and RTX 2070 in the overview
fashion model GeForce RTX 2080 Ti GeForce RTX 2080 GeForce RTX 2070
price 1,259 euros 849 euros 639 euros
Technical data
architecture Turing Turing Turing
GPU TU102 TU104 TU104
Shadereinheiten 4352 2944 2304
GPU clock (base) 1,350 MHz 1,515 MHz 1.410 MHz
GPU clock (boost) 1,635 MHz 1,800 MHz 1.710 MHz
RTX-OPS 78 TRTX-OPS 60 TRTX-OPS 45 TRTX-OPS
Gigarays / s 10 grades / s 8 Grays / s 6 Grays / s

memory clock

1.750 MHz 1.750 MHz 1.750 MHz
memory type GDDR6 GDDR6 GDDR6
memory 11 GB 8 GB 8 GB
memory interface 352 bits 256 bits 256 bits
bandwidth 616 GB / s 448 GB / s 448 GB / s
TDP 250 W 215 W 185 W
delivery 2x 8-pin 1x 8-pin + 1x 6-pin 1x 8-pin + 1x 6-pin
SLI / NVLink NVLink 2-way NVLink 2-way NVLink 2-way

The GeForce RTX 2080 Ti is in some ways the titanium model of the GeForce RTX 20 series. It therefore comes with 4,352 shader units. The basic clock is set at 1,350 MHz, boost must be possible up to 1,635 MHz. Storage capacity is identical to the GeForce GTX 1080 Ti in 11 GB capacity, but NVIDIA uses GDDR6 memory with the GeForce RTX 20 series. NVIDIA reaches a memory bandwidth of 616 GB / s thanks to the still 352-bit memory interface. The price of € 1.259 is, of course, a house number and reaches a level that we only know from the titanium variants so far.

The & # 39; mid-range & # 39; of the GeForce RTX 2080 is the GeForce RTX 2080. It comes with 2,944 shader units that can operate up to 1,800 MHz. The graphics memory has a capacity of 8 GB and has a memory bandwidth of 448 GB / sec. The price of the GeForce RTX 2080 Founders Edition is 849 Euro.

Access to the GeForce RTX-20 series starts with the GeForce RTX 2070. 2,304 shader units are used here. These work in the basic clock with 1.410 MHz and are boosted by 1.710 MHz. The expansion of the graphics memory is 8 GB identical to the larger model. This also applies to the width of the memory interface and thus the memory bandwidth. For a price of 639 euros, you can no longer be a beginners card.

With connections, the DisplayPort 1.4 cards offer three times, once HDMI 2.0a and again VirtualLink.

Prices and availability

The Founder editions of the GeForce RTX 2080 Ti, RTX 2080 and RTX 2070 are available directly through NVIDIA. From today a pre-order is possible, the delivery starts on 20 September. The GeForce RTX 2070 will be available later. Prices are € 1.259 for the GeForce RTX 2080 Ti, € 849 for the GeForce RTX 2080 and € 639 for the GeForce RTX 2070.

NVIDIA has not lost anything from the partner cards.

The hardware is followed by the software

We had a few days ago to align NVIDIA on the ray tracing. At the latest with the announcement of the RTId interface of NVIDIA and compatibility with Microsoft's DXR API for Ray Tracing, it is clear where the journey will go with the Turing generation of GeForce cards.

With the Quadro RTX series of workstation cards and the accompanying architecture modifications, NVIDIA has also determined the course in hardware. The GeForce 20 series brings this technology to the end customer. But now the question can be asked, how it is ordered for compatible games.

NVIDIA presented a number of titles at the event. These include Shadow of the Tomb Raider, Metro Exodus, Assetto Corsa and Battlefield V.


Source link

Leave a Reply