[ad_1]
As part of the Game Developers Conference (GDC), Nvidia announced a new driver supporting DXR beyond Turing graphics cards. With the driver, Raytracing via Microsoft DXR is now also supported on graphics cards from the Pascal generation.
Nvidia unlocks DXR in the driver for Pascal
The new GeForce driver that supports DXR for the GTX 1060 graphics cards with 6GB GTX 1080 Ti, as well as the Titan X and Titan XP workstation models, will be released in April. An exact date is not yet known. DXR media will also be available for the new Turing GTX 1660 (test) and GTX 1660 Ti (test) models, which also do not have special processing units (RT cores) for ray tracing, which are only available on the GeForce. RTX can be found.
The new driver thus allows the use of real-time ray tracing in games via DXR as part of DirectX 12 without proprietary RT cores and thus on traditional computer cores. However, if DXR effects are enabled, performance will be significantly lower than Turing with specialized RT cores – so it remains to be seen what the lens is. AMD could also use DXR on its Radeon graphics cards, if the driver allows it. AMD has not yet released any official ray tracing driver.
DXR on more and more APIs and engines
Nvidia expects an increase in the use of ray tracing in games. In addition to DXR as part of Microsoft DirectX 12, the Vulkan API will also receive a DXR extension during the year. With Unreal Engine and Unity Engine, two large-caliber video game engines support ray tracing. CryEkine engines from Crytek or Frostbite from DICE (EA) are also involved.
Metro Exodus (test) of 4A Games is a recent example of extensive ray tracing effects. "To come upAccording to Nvidia, Shadow of the Tomb Raider should receive a fix for Raytracing via DXR, as expected.
DLSS remains exclusively on Turing RTX
As of now, Raytracing will no longer be available exclusively on GeForce RTX at Nvidia. But with DLSS (Deep Learning Super Sampling), the result is different. AI-based alternative anti-aliasing will continue to be provided only with the RTX Family's dedicated tensor cores, Nvidia said during the Q & A after the GDC's presentation to media representatives.
At least theoretically, it would be possible to create a counterpart to the proprietary DLSS via the Microsoft Windows Machine Learning (WinML) API or DirectML. DirectML takes advantage of the universal computing units of GPUs, with which even AMD's Radeon graphics cards could offer such a function, as had already been suggested by an AMD employee in an interview.
Nvidia builds a gateway to RT cores
An interesting anecdote was provided by Nvidia's Gv presentation on the decision to use dedicated ray tracing nuclei in Turing architecture. As shown in the figure below of Nvidia, it would be enough for Pascal to become a monster chip with traditional processing units to achieve the same performance as a GeForce RTX 2080 in ray tracing mode.
For about 60 FPS in Metro Exodus with a resolution of 1440p and DXR enabled, a Pascal chip with 35 billion transistors, a performance of 44 TFLOPS FP32 with a chip area of 1324 mm² and a TDP of 650 watts were needed. Due to technical and economic constraints, Nvidia considered the creation of RT cores and specialized tensors as a better way. However, with 445 mm² (TU106), 545 mm² (TU104) and 754 mm² (TU102), RTX GPUs remain very bulky semiconductor chips and are therefore expensive to manufacture. This is also noticeable at the relatively high prices of the GeForce RTX.
Source link
