Home > News content

Raytra tracking 最大 the biggest leap in 12 years NVIDIA RTX 20 series graphics card officially released

via:快科技     time:2018/8/21 7:01:55     readed:991

Rasterized rendering is always done, and the output of the image is obtained by computing triangles and polygons, but the natural flaws of this technique make it possible to simulate the real world as much as possible, without infinite approximation or even full reproduction.

Ray Tracing technology is called the “Holy Grail” of the graphic world. Simply speaking, it tracks the light of objects and the environment in real time during the graphics rendering process, accurately reflecting light reflection and refraction, global illumination, and physical shadow. Brings almost 100% real renderings, especially light and shadow effects.

The ray tracing technology is not new, it has been proposed very early, but its calculation is too large (I want to calculate the various propagations of countless rays in the scene in real time), the previous GPU can not afford real-time calculations. Until now, NVIDIA has brought a revolutionary "Turing" Turing architecture that supports ray tracing for the first time and adds AI artificial intelligence, while the API supports the latest DX12 DXR (DirectX Ray-tracing).

The Turing Core (TU102) integrates 18.6 billion transistors and has a core area of ​​754 square millimeters, which is 60% and 58% more than the Pascal Pascal architecture core. It is the second largest chip ever.

It has three different cores inside.First, the traditional SIM CUDA core array, floating point computing performance is the biggest14TFlopsThe integer calculation has a maximum performance of 14TIPS, and supports floating point, integer parallel execution, and variable rate shading.

two isRay tracing dedicated core RT Core, up to 10 billion rays per second (10 Giga Rays/sec), three isDepth calculation auxiliary core Tensor Core, FP16 floating point performance 110TFlops, INT8 integer performance 220TOPS, INT4 integer performance 440TOPS.

Because it is a new technology and lacks uniform metrics, NVIDIA has also invented many new calculation methods, such as the 78T RTX-OPS, which represents the number of ray tracing operations performed by the Turing architecture.

RTX ray tracing rendering technology was actually proposed at the GTC Technology Conference in March this year. The Tesla V100 computing card that used four Volta volts architectures can get satisfactory results, but now only one Turing core can get better. The performance of the Pascal architecture is even more incomparable.

Below are a few ray tracing switch renderings, where the effect of the Turing architecture can be done in real time:

The game's support for Turing architecture ray tracing needs to be optimized, but with a solid ecosystem and strong appeal, it has been supported by numerous masterpieces.There are more than 20 models in the starting lineup, including the upcoming Battlefield V, The Shadow of Tomb Raider, Metro Exodus, and more. The domestically produced "Reverse Water Cold" and "Jian Xia Love 3".

Ok, let's look at the product:

NVIDIA launched three new cards in one launch, all with an open dual fan design.

The top level isRTX 2080 Ti,4352 CUDA cores, core reference frequency 1350MHz, acceleration frequency public version 1545MHz, FE public version overclocking 1635MHz, with 352-bit 11GB GDDR6 memory, equivalent frequency 14GHz, bandwidth 616GB/s, whole card power consumption 260W, 8+8 pins Power, ray tracing performance 10 Giga Rays/s, 78T RTX-OPS.

Next isRTX 2080,2944 CUDA cores, core reference frequency 1515MHz, acceleration frequency public version 1710MHz, FE public version overclocking 1800MHz,Match256-bit 8GB GDDR6 memory, equivalent frequency 14GHz, bandwidth 448GB / s,The whole card consumes 225W, 8+6-pin power supply, and ray tracing performance of 8 Giga Rays/s and 60T RTX-OPS.

FinallyRTX 2070,2304 CUDA coresCore reference frequency 1410MHz, acceleration frequency public version 1410MHz, FE public version overclocking 1710MHz, with 256-bit 8GB GDDR6 memory, equivalent frequency 14GHz, bandwidth 448GB/s,The whole card consumes 175W, 8-pin power supply, and ray tracing performance is 6 Giga Rays/s and 45T RTX-OPS.

Ray tracingPerformanceMemory
RTX 2080 Ti10 GigaRays/second78T RTX-OPS11GB
RTX 20808 GigaRays/second60T RTX-OPS8GB
RTX 20706 GigaRays/second45T RTX-OPS8GB

RTX 2080 Ti 999 US dollars, National Bank 9999 yuan; RTX 2080 699 US dollars, National Bank 6499 yuan, starting from now on pre-sale, officially listed on September 20.

RTX 2070 $499The price of the Bank of China has not been announced, it is estimated to be around 4,999 yuan.

There is also a professional levelDual card combination Quadro RTX 8000 (DUAL)With 96GB GDDR6 memory, ray tracing performance of 20 Giga Rays/s, 188T RTX-OPS,$20,000.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments