What is the upgrade of the RTX30 series compared to the RTX20?
The NVIDIA® GeForce RTX ™ 30 Series has been launched for over half a year, but many readers still do not understand what the current generation compares to the previous generation RTX 20 Series. If you use a previous generation RTX graphics card, do you need to upgrade it? Let me first explain what are the most important upgrades of the GeForce RTX 30 Series!
Nvidia Ampere Architecture SM
The current generation GeForce RTX series 30 series graphics cards use the new Ampier GPU Micro Architecture. The previous generation Turing architecture had 64 fp 32 kuda and 64 INT 32 arithmetic units on each SM module. It also had one SM partition, 1 FP and 1 INT data path, so each Turing SM and each cycle could perform 64 FP 32 64T and 32 Ft 32.
Games The average ratio of FP’s INT instructions to modern games is approximately 100: 36
As modern games have more advanced computing requirements, many workloads mix FP32 computing instructions (such as FFMA, FDD, and FMUL), but to maximize their use, such as address, addition, and floating-point comparison. Value / minimum value and other tasks such as obtaining processing results. The FP to INT instruction ratio in modern games is on average 100: 36. Therefore, the Ampere GPU adds more efficient scheduling for two types of computing instructions.
Am Better Ampere SM Design
The new ampere architecture converts the SM module into 128 FP32 CUDA arithmetic units, half of which are INT32 arithmetic capabilities, so that the 2 data paths in each SM partition can be used simultaneously for FP operations, but 1 of them can be scheduled for INPT operations for each of the 32 amp Can be done or adjusted to 64 fp 32 + 64 INT 32 work operations, i.e. FP 32 doubles the computing power, as ray tracking will have more to do after the FP 32 mathematical function is enabled, so the performance is significantly improved in the new ampere architecture of live ray tracing.
Upgrade Second Generation Ray Tracing Engine !!
Speaking of the most important part of Nvidia GeForce RTX graph graphics cards, of course RT Core is responsible for the live ray tracing function of ray tracing, which can realistically mimic and physically render the light of scenes and objects. Realistic reflections and reflections in real time, shadows and indirect lighting provide realistic lighting effects in game scenes. Although the RTX 20 Series with Turing architecture has a live light tracking engine added, due to the limited computing power of the RT Core, it can cause serious FPS drop after ray tracking is turned on. Take the previous generation RTX 2060 for example, you could say “you can not use it if you do not see it”. At most, it can only be used to post on Facebook to take a screenshot, but in the actual game, most players still choose to turn off the light chase, which is a pity.
Upgrading the current generation GeForce RTX 30 Series RT Core engine to the second generation, the BVH algorithm box greatly accelerates three-dimensional collision and triangular intersection capabilities, and the performance is twice as high as that of the first generation RT Core. In the past, Turing SMMs have improved the inability to perform other functions simultaneously in ray tracing operations. When performing ray tracing operations, synchronize the graphics or computer functions so that live ray detection can be implemented in the game. Time efficiency has been greatly improved.
Additionally, a new Dynamic Blur hardware acceleration function has been added, and the interpolate tri position is imported when performing triangular intersection calculations. Readers can understand this as a parameter calculation related to “time”, which allows them to respond to Ray’s findings. Make changes over time and finally render to the “Light Chase” image with the dynamic blur effect.
New Third Generation Tensor Course Computing Unit
Another important feature of Nvidia GeForce RTX graph graphics cards is the tensor core arithmetic unit, which is used exclusively for performing vector and matrix functions, mainly for INT 8, INT 4 accurate function operations, and high accuracy FP 16 functions. In-depth learning neural network functions, inference functions, matrix functions, etc.
The new Ampere GPU Micro Architecture has been upgraded to the third generation tensor course computing unit, adding new types of data computing modes such as sparse computing, new accuracy modes such as TF32 and biflot 16, and matrix multiplication speed. Improved. 2 times up. Although Nvidia halved the number of tensor cores per SM, the computing power of each tensor core became stronger, giving it more robust AI computing capabilities.
RTX 2060 (Turing) vs RTX 3060 (Ampere) Tensor Core Comparison
|GPU architecture||NVIDIA® Turing||NVIDIA® Ampere|
|Tensor core for each SMM||8||4|
|FP16 FMA functions for each tensor core||64||Dense: 128
|Total FP16 FMA operations per SM||512||Dense: 512
Do RTX series 20 Series users need to change their cards?
After reading the content above, you can almost see that the RTX series 30 series is a huge improvement over the RTX series 20 series. Data data indicates that the current generation game of the ETTX points 3080 can be nearly doubled. Performance increases compared to previous generation RTX80 2080 Super RTX series 20 Series Graphics Card User Want to replace the card? How do I change that?
|1080p||RTX 2070 (Super)||
RTX 3060 Ti
RTX 2080 (Super)
RTX ™ 2080 Ti
|RTX ™ 3070|
In fact, readers can refer to the list above to choose the graphics card model that best suits their game needs. This can be considered as providing a limit to the actual tracing gaming experience. If you are a user of RTX ™ 2070 (Super) and would like to try out the 2K resolution light chase gaming experience, I suggest you upgrade to RTX ™ 3070 or higher. For readers who want to experience even 4K or 8K ray tracing games, the high quality Nvidia GeForce RTX80 3080 and RTX3090 may be your only choice.