At GTC, Nvidia not only introduced the first three Geforce RTX 4000 cards, but also DLSS 3. One of the most important questions to be raised about the new expansion phase of Deep Learning Super Sampling is whether or not DLSS 3 is exclusive to the RTX -40 series? The short answer is yes, it’s exclusive. The long answer is a bit more complicated.
Brian Catanzaro, Vice President of Applied Deep Learning Research at Nvidia, confirmed that DLSS 3 in its current form only works on the Geforce RTX 4000. According to him, it is theoretically possible to equip older generations with DLSS 3. In theory – whether Nvidia will bring DLSS 3 To RTX 30 (Ampere), RTX 20 (Turing) or even older cards is somewhat questionable. According to Catanzaro, DLSS 3 does not offer any major advantages for owners of older Geforce graphics cards.
Does DLLS 3 work on Geforce RTX 3000 (Turing) and RTX 2000 (Ampere)? (2)
Source: Nvidia
The background is the tensor cores of Ada Lovelace, the codename for the latest generation, and the optical flow accelerator. Both are already on Turing cards, but with Lovelace they are greatly improved, according to Nvidia. At the moment, the main conditions are the productivity values and higher performance factors that the units are now supposed to achieve. This probably means a performance penalty for DLSS 3 on older cards. How dramatic or not, only time will tell. Until then, Nvidia’s marketing data remains unchecked in the room. For example, the optical flow accelerator is said to have 2 to 2.5 times the performance of its predecessor. According to Nvidia, DLSS framework creation is only supported in Lovelace.
| DLSS sub feature 3 | GPU hardware support |
|---|---|
| Generate a DLSS framework | GeForce RTX 40 Series GPU |
| DLSS Ultra Resolution (also known as DLSS 2) | GeForce RTX 20/30/40 . GPU Series |
| NVIDIA Reflex | GeForce 900 series and the latest GPU |
As an example: Intel’s XeSS does hardware acceleration on Arc graphics cards, but it also works without that hardware acceleration. Independent tests have yet to show the size of the performance penalty. However, it does show that new technologies can certainly be introduced comprehensively, even if older or non-adapted devices achieve less results.
However, with DLSS 3, Nvidia also includes Reflex in the overall picture and it would be counterproductive if older hardware performed worse, while on the other hand, latency was announced as an important point. Perhaps Nvidia is still at a disadvantage as it was with G-Sync and is offering DLSS 3 as a “light version” over Ampere and Turing. You will find out from us when the time is right.
Source: via Twitterreddit
