The most anticipated next generation of hardware from Nvidia has finally hit the market. The Nvidia GeForce RTX 2080 is available to buy right now while the mighty RTX 2080Ti has been delayed a week and will be available on 27Th September with the RTX 2070 to be released in October. Almost all preordered RTX 2080 has been delivered, and the general availability of the Graphics card can be seen in the market.
The two-year long reign of the Pascal GPUs seemed like a century when compared to the 2-year cycle of both Maxwell iterations because there was a lot happening in the market. The market was going very down due to the cryptocurrency mining, then we saw specialized Graphics cards for mining, Nvidia released Volta architecture, and most importantly Intel announced its return in the GPU market would be in 2020.
After a lot of rumors and speculations, Nvidia announced their next generation of Graphics cards on the stage of Gamescom. Most of the time on stage was taken by how the new features of the Graphics cards would take the world by storm. Because the real time ray tracing is commonly known as the “Holy grail” of graphics and implementing that in the consumer product is quite a task that Nvidia was boasting on stage. The other feature called DLSS has its particular advantages (Greater than raytracing for me), and it had its own share of time on stage. The critique that people made on Nvidia’s presentation were that it only showed one side view of the story, what they should have done was to talk about the good old rasterization along with these features. People should have gone home with a notion of what has changed in two years (which they did pretty well) AND (pardon the caps) how these new shiny pieces of hardware devices stand compared to their old Pascal Graphics cards.
Instead, Nvidia chooses a delusional approach; they introduced a new unit of reference called RTX OPS in place of TFLOPS which made things hard for the consumer in general. Additionally, during their presentation, they downplayed the specifications of their Pascal devices (which are still n the market if you want). As a result, people awed the new features and missed the old comparisons. Even though Nvidia’s presentation was not on par, the new RTX series got the much-awaited hype. The pre-order limit of these Graphics cards filled in a matter of days and now we finally have the new Founder’s edition Nvidia GeForce RTX 2080 Graphics card in the house. So, without further wait let’s get into the full review and see if Nvidia rushed the new features.
The Nvidia RTX 2080 is based on the Turing architecture which Nvidia introduced during the reveal of new Quadro RTX Graphics cards at SIGGRAPH. Turing architecture in a very crude way a hybrid of the Pascal and Volta architecture, although it is vastly superior to both. It has the enhanced rasterization capabilities of Pascal, the AI capabilities of Volta and the raytracing capabilities which are new to the table. The TU 104 GPU inside the RTX 2080 gets its power of playing with pixels from a huge silicon die that is 545mm2 in size, for comparison the die size of GP 102 in GTX 1080Ti (we’ll be comparing the RTX 2080 with the GTX 1080Ti a lot after that so bear with us) was 471mm2. While all of the 471mm areas in the GTX 1080Ti was devoted for rasterization, the story is different in the RTX 2080. It has room for the traditional rasterization units which are commonly known as shader units (CUDA cores) along with Tensor cores (for AI) and RT cores (for raytracing). So, we can say that 1080Ti has more room for the CUDE cores hence the CUDA core count in the RTX 2080 is slightly less than the GTX 1080Ti.
The chip is built with TSMC’s 12nm FinFET process, and by far it’s the third largest chip made under the FinFET process. The TU104 has eight Turing streaming processors, for each Graphic cluster and there are six of these, though two of these are inactive. Which makes 46 SMs in total, 88 Rendering processing units 64 of them are active along with eight memory controllers. These improved memory controllers can now handle eight ROPs in one second making rendering more efficient and smoother. Lastly, the Turing GPU offers concurrent integer execution paths for the SMs which means that the GPU can execute the floating point number FP32 and integers AL32 side by side which will also enhance the rendering capabilities.
One of the reasons why the Founder’s edition cards are sold separately at a higher price (usually $100 more) is due to their aesthetics and materials. If you look at these devices, they look strikingly different than their predecessors. Nvidia has completely revamped the design aesthetics of the Graphics card from “intense blower” to “casual wind fans.” The outside shell of the Graphics card is made up of Aluminum with Plastic at the center. There are two wind blade fans instead of a single blower fans, and these fans are slightly raised from the surface which gives the card a strange “circular blocky” look. Though they still look very pretty and stealthy. Lastly, there is a plastic backplate too which will prevent the GPU from wear and tear.
Coming to one of the most important aspects of the review the Specifications of the GeForce RTX 2080. Every SM is made up of 64 CUDA cores, and there are 46 SMs on the die so, the total number of CUDA cores become 2,944 which is less than the number of CUDA cores (3,584) in the GTX 1080Ti. These are clocked at a base speed of 1515MHz with boost clock speed of 1710MHz and factory overclock speed of 1800Mhz. For AI there are 384 Tensor cores in total, of which 368 are active that can produce 322 TOPs of rendering power. Lastly, for accelerating the bounding hierarchy of the physical properties of the rays of light, there are 48 RT cores in total with 46 active cores capable of tracing 8GigaRays/s.
The memory RTX 2080 will be using is GDDR6 module made by Samsung with 8GB size. The memory is running at a speed of 14Gbps with a 256-bit interface for a total of 448GB/s memory bandwidth that is very close to the memory bandwidth of the GTX 1080Ti (484GB/s). Considering the size of the memory is less so, the leap from GDDR5X to GDDR6 is huge. Lastly, the single precision score of the RTX 2080 is 10.1 TFLOPS that is slightly less than 11 TFLOPS of the GTX 1080Ti.
Now let’s talk about the new features that the Graphics card will introduce.
Starting with the most important achievement in the field of rendering real time raytracing, the “Holy grail” of Graphics. Rendering light is one of the most taxing tasks of the GPU, so the developers try to use fake lights in their games so that the Graphics cards do not have to bother about the reflections and refractions of light. The use of fake light creates a dulls scene that is far from real, and the sole goal of the field of Graphics is to portray reality. In pursuit of the reality, many scientists tried and failed hence raytracing got the name “Holy Grail” of Graphics. Now, Nvidia opted for a clever route, instead of tracing all rays (9infinity rays emit from an area light) coming from the light source only trace those rays which will eventually hit the screen or the viewer’s eyes.
So, the function of these RT cores is to trace as many rays of light as it can handle from the screen space back to the light source while closely simulating all properties of light. It is called RTX. When it is enabled, the whole scene looks close to the reality and creates much more options for the developer since he would not have to deal with fake lights.
Anti-aliasing is also one of the most taxing tasks that the GPU has to perform during gaming. If your GPU is not strong, it is usually recommended to turn it off. Now with the use of AI Nvidia has introduced a new technique of Anti-aliasing called DLSS (Deep learning supersampling). The purpose of this technique is to increase the resolution of the image without much load on the shader units. Nvidia’s plan to achieve it is by the use of their Supercomputer called Saturn V. The process is very simple Nvidia will train the neural networks on the Saturn V computer by providing at millions of pictures and letting it think which pixel will add where to increase the resolution of the image.
They will regularly train the neural networks, and these will be provided to the consumers through regular updates. So, in theory, it should not have the usual effects of anti-aliasing. It is not sure however how much of a load the DLSS will be on the CUDA cores.
Coming to the performance of the new Graphics card. Starting with the benchmarks, it got a solid score of 6,418 on 3DMark’s TimeSpy Extreme test, which is very close to the 6,839 the score of GTX 1080Ti. While the story is reversed in Unigine Superposition where the RTX 2080 won with the slightest margin. Both these benchmarks show that these Graphics cards are at least comparable to each other. Now let’s see how well the RTX 2080 compares against the GTX 1080Ti in games.
Shadow of the Tomb Raider is the most recent game to date, and the RTX 2080 was able to get 28FPS at the highest possible settings at 4k resolutions while the GTX 1080Ti fall right behind at 25FPS. Both these Graphics cards struggle to tame the game at the highest settings. However, not much of a difference here, the Shadow of Tomb Raider is one of those games that would benefit with Raytracing and DLSS. But unfortunately, the current build of the game does not support either, the developers have said that they will add these functionalities at a later date. Next, we have the most demanding game that is often called the benchmark of the Graphics cards Ashes of Singularity. The RTX 2080 showed solid performance in this game too. However, the GTX 1080Ti got the slightest lead here. At 4k Extreme Quality settings, the RTX 2080 got a respectable framerate of 74.3 while the GTX 1080Ti was able to score 75.7 Frame per second.
The latest title in the portfolio of Ubisoft Farcry 5 has a deep and woody open world which can also test any device to its fullest. Here the RTX 2080 got its lead back, though the margin was pretty small again. At 4k Ultra settings it could only get three more frames per second while neither of the Graphics cards could hit the 60FPS mark. Lastly, we have the Battle Royale game that takes the juice out of every rig PUBG. Only a few Graphics cards can run the game at 4k, and the RTX 2080 is one of these cards, and it can produce enough framerates to help you in a competitive scenario, essentially Battle Royale.
Before the final words, let’s ponder on the promises Nvidia made during their presentation. During the presentation, none of them felt like far from reach, but now they seem like a little too grandiloquent. The strengths of the new RTX Graphics cards are Raytracing, and DLSS and only one game Infiltrator Demo (full game has not been released yet) support DLSS while no currently available game supports Raytracing. These features are essentially useless currently, which makes the release of the Raytracing and DLSS rushed by Nvidia.
However, the RTX2080 does not slack on rasterization, in most of our games it was very close to the GTX 1080Ti. However, the price difference between both these Graphics card is huge. Which is the only reason why it would be safe to recommend GTX 1080Ti for now? The game developing market mostly revolves around console development which is also one of the reasons why raytracing and DLSS may delay. All in all, the RTX series seems kinda rushed if you were looking for raytracing and DLSS, it is recommended to wait a while and see how things unfold and how the graphics card performs in a raytraced game.