Xiaomi Mi Band 3
The hottest and most interesting thing of late is definitely the release of the new Turing architecture RTX 20 series graphics card by NVIDIA. Compared to the previous generation of Pascal architecture graphics, Turing, in addition to the overall performance boom, but also to make the quality of the game more realistic light tracking and DLSS technology. I personally think that the latter two new technologies are the biggest motivation for players to replace Turing graphics cards. However, RTX 2080Ti is very out of stock now, and it is not easy for me to find agent friends to grab a piece. It is really a robbery. Many trenches who are not short of money are actually willing to increase their prices to buy cards.
First, what new technology and performance improvements does the new Turing architecture bring?
Let’s take a look at what new technologies and performance improvements Turing’s new architecture has brought us. NVIDIA’s recent graphics card architectures are named after scientists. This time Turing refers to (Alan Mathison Turing), the father of human computer science and the father of artificial intelligence, which seems to suggest that the new architecture of this generation is one that highlights artificial intelligence. The Turing core, built using TSMC 12nm FFN technology, integrates 18.9 billion transistors and has a core area of 754 square millimeters. Compared to the previous generation of Pascal architecture, Turing has a larger core area and more advanced manufacturing process. In addition to the continued increase in the size of the CUDA core and the upgrading of Shading shading rendering, the new technology is mainly the addition of the new RT CORE, which supports up to 10 billion ray calculations per second, and the introduction of AI artificial intelligence accelerated TENSOR CORE for real-time game graphics.
1. RT CORE and real-time ray tracing technology.
NVIDIA’s 20-series graphics card prefix from GTX to RTX, to a large extent has something to do with RT CORE, but also shows how proud Lao Huang is of this new technology. Ray tracing is not a new concept, ray tracing technology is actually light source light produced by reflection, refraction and other light changes and the resulting impact on the shadow through the calculation of accurate reflection into the picture. Make the effect of light and shadow closer to reality. However, the previous technologies need to consume a lot of computing resources, it is easy to CPU and GPU down, time-consuming and laborious, so these technologies have not been widely promoted and popularized, ordinary gamers are not experienced in the game. It’s still mainly used in filmmaking. The new RT CORE in Turing’s core is designed specifically for ray-tracing technology. The combination of RT CORE and ray tracing makes it easy for Turing’s new graphics card to carry the huge amount of computing needed to create a light-and-shadow tracking effect.
The above is the official NVIDIA contrast picture, notice the second picture on the body of the fireworks reflection, than in the first picture without the use of shadow tracking technology effect is more reasonable and true. Later in the game, we can even hide behind the wall and stick out a mirror or gun barrel to observe the activities of the enemy.
2, TENSOR CORE and DLSS.
Another important new feature of Turing’s architecture is the “deep learning, super sampling, anti-aliasing” (DLSS) technology based on AI artificial intelligence. A variety of anti-aliasing technology with gamers through the year after year, this technology so that players are both love and hate, love is to open the anti-aliasing can make the picture more realistic, delicate; Hate is the higher the quality of the anti-aliasing card resources occupied by the larger, means that the card performance is not enough, then there will be a variety of Catton. What kind of thing is DSLL? To put it in a nutshell, NVIDIA works on a supercomputer by collecting multiple reference images that are 64 times the size of a supersample. Then select some pictures in a certain way as the “standard answer” of perfect rendering, and then through the DLSS artificial intelligence depth learning, compare the standard resolution pictures with these pictures, and generate an optimal picture. Then compared with 64 times of super sampling, the differences are fed back to the neural network after the difference is obtained, and the training cycle is carried out. After many iterations in this way, AI networks can learn how to render standard images to near 64 times the size of the super-sampled original. These learning results will be regularly updated through software (mainly GFE software) to use Turing GPU card, through the TENSOR CORE can be real-time comparison, so that the lower resolution of the picture “brain fill” for the high resolution picture; To improve the quality of the picture. This is the process of most of the operations are in the NVIDIA supercomputer to complete, DLSS basically does not need to consume local GPU resources, so can greatly liberate the labor of GPU, let it concentrate on rendering the game.
There are very few games that support both RTX ray-tracing and DLSS antialiasing, and games like Tomb Raider: the Shadow and final Fantasy 15 have to wait for an official upgrade patch to support them.
Analysis of the hardware parameters of Turing display Card.
The NVIDIA released a total of three new Turing graphics card models, are high-end products, including: RTX 2080Ti, RTX 2080, RTX 2070, respectively corresponding to the TU102, TU104, TUF106 and other three different core. However, the RTX 2080 Ti does not use a complete TU102 core, but a streamlined TU102-300A,CUDA of 4352, 272 texture units, 88 ROP units, 68 RT cores, and 544 Tensor cores. The basic frequency of the GPU core is 1350MHz, the boost accelerating frequency is 1635MHz, the non-official 1545MHz (more and more non-official ones are on the market, the frequencies are also different), the GDDR6 grain is adopted, the capacity is 11GB, the memory bit width is 384bit, and the equivalent frequency is 14GHz. The bandwidth is 616GB / s, and the power consumption of the whole card TDP is 260W.
RTX 2080 uses a simplified version of the TU104-400 core, CUDA for 2944, 184 texture units, 64 ROP units, RT core 46, Tensor core 368. The basic frequency of GPU core is 1515MHz, boost accelerating frequency is 1800MHz, non-standard is 1710MHz, GDDR6 memory grain is also used, capacity is 8GB, memory bit width is 256bit, equivalent frequency is 14GHz, bandwidth is 448GB / s, the power consumption of the whole card is 225W. The core frequency is 1515MHz, the boost accelerating frequency is 1800MHz, the non-standard is 1710MHz, the capacity is 8GB, the bit width is 256bit, the equivalent frequency is 14GHz / s. RTX 2080 hardware specifications in comparison with the RTX 2080Ti reduction is quite large.
RTX 2070 uses the complete TU106 core, 2304 CUDA core, 144 texture unit, 64 ROP unit, 288 Tensor core, 36 RT core. The GDDR6 memory of 8GB is 256-bit wide, and the equivalent frequency is still 14GHz. The basic frequency of GPU core is 1410MHz and boost acceleration frequency is 1710MHz and non-standard 1620MHz. The power consumption of the whole card is 185W.
III. Opening the box.
The specific model of the graphics card is Sothel RTX2080 Ti 11GD6 X GAMING OC, packaging box and so on, nothing to dry up, direct drying graphics card. The overall color matching of the graphics card is still based on the traditional black and gray of Sotel, with three 8mm fans on the huge radiator. Graphics card thickness 5.3 cm, need to occupy two and a half PCI slot, length 31. 1 cm, is a relatively long graphics card, the chassis has certain requirements, so when installing the box must pay attention. In the aspect of video output interface, three DisplayPort 1.4, one HDMI 2.0 and one TYPE-C interface are provided. The back of the graphics card is covered with metal heat dissipation armor, which is hollowed out at the back of the GPU core, which is more conducive to heat dissipation.
Sotel RTX 2080Ti X GAMING OC’s PCB is much the same as the public version, with some optimizations. The power supply part uses the 16-phase power supply plan of 13 + 3. GPU uses a TU102-300, core area much larger than the previous generation of GTX 1080Ti. Display memory using Meguiar GDDR6 particles, the default frequency of 14 GHz, a single 1 GB, a total of 11 components of the total capacity of 11GB. This capacity is a bit odd, so we can see that there is a memory grain bit is left blank, memory bit width 352bit, memory bandwidth up to 616GB/s. The external power supply is 8pin+8pin. The SLI bridge interface uses NVLink 2.0, which is much larger than the previous one, and requires the purchase of a dedicated NVLink bridge, which currently only supports dual card connectivity.
The radiator is a pure copper endothermic base, which is closely connected with the storage particles through a heat sink, plus 5 8mm pure copper nickel-plated heat pipes and dense large heat dissipation fins, coupled with a heat dissipation system of three 8cm fans. In the RGB rampage today, do not support the RGB lights are embarrassed to market, this radiator is also with RGB lights, can be adjusted through the software FireStorm Sotai various lamp settings.
Performance measurement and overclocking.
1. Theoretical performance test at default frequency.
The nominal frequency of Sotel RTX 2080Ti X GAMING OC is slightly lower than that of the common version. The basic frequency of GPU is 1350MHz and the boost frequency is 1605MHz. The memory frequency is 14GHz. The maximum Boost frequency in operation is 1965MHz (higher than the public version). The TDP power of BIOS is 260W, the floating upper limit is 105%, and the power is 272W. Adjustable power up to 115%, up to 300W. Full load run of the game for two hours, maximum GPU core temperature is no more than 62 ℃. The maximum speed of the fan is within the range of 2000RPM, and the noise is low.
2. Overclocking test.
Overclocking test is the use of Sotheart matching FireStorm software. The GPU core frequency setting for the Turing graphics card is a little different from what it was before, and the frequency pulled up in the software is actually 1/2 of the actual boost frequency. In other words, you increase the frequency of the 10MHz, the actual overclocking effect is to enhance the 20MHz.
The overclocking potential of GDDR6 memory particles is good, and the overclocking potential of silent 14GHz is easy to overclocked to 16GHz. RTX 2080Ti’s GPU core overclocking potential is also good, the actual operation of the maximum Boost frequency from 1965MHz to 2085MHz is very easy, close to 2.1GHz. After overclocking, the performance is about 8% higher than the default frequency.
3. Actual game performance test.
Because there is no official version of the game yet to support RTX ray tracing and DLSS, the actual game performance test is mainly aimed at the mainstream popular DX12 and DX11 games. From the actual game comparison, in the DX12 game, 1080p resolution, the RTX 2080Ti than the previous generation of flagship GTX 1080Ti has about 23-29% performance improvement, up to 4K resolution, the increase is 23-35%. In the DX11 game, 1080p resolution, the RTX 2080Ti than the previous generation of flagship GTX 1080Ti only about 9-20% performance improvement, performance improvement is not as obvious as the DX12 game, and at 4K resolution, the increase to 25-37%. The biggest significance of Turing architecture is to focus on improving the performance of DX12 games and 4K resolution, without wasting too much resources in DX11 games and 1080p resolution.
4. RTX ray tracing and the first experience of DLSS.
Currently available for RTX ray tracing and DLSS only “Star Wars DEMO”. But this DEMO is just a beginner’s experience. There is no switch in DEMO to turn on / off RTX raytracing and DLSS effects. By default, both RTX and DLSS are turned on. So can not be directly from the picture quality effect and frame number gap to compare the same card before and after the difference between the opening. This just a few minutes of DEMO, is mainly from the ground and the character armor to the environment of the various reflections to highlight the characteristics of RTX ray tracing technology. In particular, the armor’s reflection on the outside world is as realistic as a real movie clip. As the character moves, the angle of the external object reflected from the armor and the brightness of the light can change accordingly, which really deserves a lot of praise. At 4K resolution, the number of frames in the RTX 2080Ti can reach more than 40 fps, while the GTX 1080Ti can only play one frame like a slide show. The performance gap between the two when the RTX is turned on is definitely more than 30% of the conventional performance gap. But by a few orders of magnitude.
Experience in use:
One of Turing’s biggest improvements over Pascal is the addition of the RT Core (ray-tracing core, which is designed to track light and shadow. After turning on the RTX, the image can be made to look like the real thing. Although DLSS antialiasing has no obvious improvement compared with TXAA in image quality, it can greatly reduce the resource consumption of graphics card under the condition of almost the same image quality. So as to improve the number of game frames, improve the game flow. The hassle is that the games that claim to support both RTX and DLSS have not yet been released or need to wait for the release of an upgrade patch, so Turing’s huge advantage in this area has yet to be fully felt. We can only wait for the release of the corresponding game quickly.
The advantages and disadvantages of Sotel RTX2080Ti 11GD6 X GAMING OC:
1, advantage: better than the heat release, higher than the actual operation frequency of the board, lower than the price of the board, a good overclocking potential.
2. Disadvantages: the effect of RGB on the radiator is relatively monotonous. Willing to spend 10, 000 yuan to buy graphics card users, graphics card in all aspects of the requirements must be relatively high, this card RGB light efficiency is relatively simple.
III. Purchasing proposals.
RTX 2080Ti gives me the biggest feeling is an uppercase Shuangzi. Everything is much better than GTX 1080Ti, but the price is also more sour, compared to the price of GTX 1080Ti has almost doubled. What can you buy for nearly twice the amount of money you spend? You can buy 30% of the conventional performance improvements and RTX ray tracing and DLSS technology not available on the previous generation of graphics cards. If 30% performance improvement and DLSS are not attractive enough, I personally think that the RTX ray tracing project is really a qualitative change in the quality of the graphics card.
Given that the AMD does not currently have a competitive product on the market (which is unlikely to happen any time soon, judging from AMD’s product blueprint), RTX 2080Ti prices should remain strong for a long time to come. Especially in the current general shortage of goods in the market situation is not to expect Lao Huang to RTX 2080Ti to do price promotion. So do not pursue new, the pursuit of cost-effective users, you can first consider the price more affordable GTX 1080Ti/1080. If you want to enjoy the experience of the first super cool game users (not lack of money), then there is no way but to pay for it. Overall, the RTX 20 series is indeed a cross-generational improvement over the GTX 10 series, and it makes sense that it is so expensive to sell. Whether it’s worth it or not depends on what you want from the game and how thick your wallet is.