This is part of Digital Foundry’s current coverage of the Xbox Series X. You can find the rest here:
- The full specifications of the Xbox Series X! How powerful is the new generation console?
- How big is the Xbox Series X compared to the Xbox One X?
- Xbox Series X: Why Faster SSD Storage Changes Everything
- Xbox Series X: the fight against input lag and heartbreak
- Xbox Series X: How Old Games Work Better on Next Generation Console
The ray tracing difference
RDNA 2 fully supports the latest DXR Tier 1.1 standard and, like the Turing-RT core, speeds up the creation of the so-called BVH structures, which are necessary for the exact mapping of beam paths and intersections, which are tested against geometry. In short: just as light “reflects” in the real world, hardware acceleration for ray tracing maps the crossing and intersection of light at speeds of up to 380 billion intersections per second.
“Without hardware acceleration, this work could have been done in shaders, but alone would have used more than 13 TFLOPs,” says Andrew Goossen. “With the X series, that work is moved to dedicated hardware and the shader can continue to run at full power in parallel. In other words, the X series can effectively achieve the equivalent of well over 25 TFLOPs in ray tracing. “
However, it’s important to put this in the right context. Although the workloads can run at the same time, calculating the BVH structure is only one part of the ray tracing method. Standard GPU shaders must also do their part so that things like lighting calculations continue to be performed on standard shaders. The DXR API adds new stages to the GPU pipeline to perform this task efficiently. Yes, RT is generally associated with a decrease in performance which also applies to the console implementation. With the benefits of a solid console design, however, we can expect developers to optimize more aggressively and be innovative as well. The good news is that Microsoft is providing low-level access to RT acceleration hardware.
« [Series X] Goossen goes even further than the PC standard and offers developers more performance and flexibility, ”reveals Goossen. In keeping with console tradition, we also support programming at the lowest level, including support for offline BVH design and optimization. With these building blocks, we expect ray tracing to be an area of amazing graphics and a big innovation for developers over the life of the console. ”
During our visit to the Redmond campus, Microsoft used a very first tech demo of the Minecraft DXR Xbox Series X based on the Minecraft RTX code we saw at gamescom last year to demonstrate the Console’s RT capabilities – they looked very similar, even though they were running on a completely different GPU. Which is somewhat ironic: the base Nvidia code has been tweaked and works on AMD X-series ray tracing hardware. What’s impressive is that there is full path tracing. Other than the skybox and the moon in the demo we saw, there are no raster elements. The entire presentation supports ray tracing. This shows that the Xbox Series X can deliver the most ambitious and impressive implementation of ray tracing, in real time, despite the limitations of delivering RT in a console with limited power and silicon budgets.
Minecraft DXR is an ambitious statement – full ray tracing if you will – but we have to expect the technology to be used in very different ways. “We’re very excited about DXR and the hardware raytracing support,” says Mike Raynor, CTO of The Coalition and Gears 5. “We have some compute raytracing features in Gears 5, we have shadows raytracing and that [neue] Global screen space illumination is a form of image-based GI with ray tracing. Therefore, we are interested in how ray tracing equipment can be used to use such techniques and then move on to using DXR cores.
” I think, […] that in the future we’ll see hybrid rendering between traditional rendering techniques, and then use DXR for that – whether it’s shadows, GI, or adding reflections – and these are things that can really expand a scene and [wir können] use the whole chip to get the best visual quality at the end. ”
Efficiency in design
For me, one of the most important information about X-series silicon was not just performance, but design efficiency. With all the new graphics features and 12 teraflops of consistent computing power, we had to imagine a monstrously large and unacceptable processor design – in fact, a very expensive console. However, SoC size means 360mm2ethat the chip is much smaller than what was originally expected. The 15.3 billion transistors means that the density of transistors has more than doubled compared to the 16 nmff design of the Xbox One-X processor, and yet the performance is once again significantly better at just twice that.
In order to achieve the objectives set in terms of performance, performance and chips, innovative thinking on the part of Microsoft was needed. Graphics performance isn’t just about teraflops. Computing power requires memory bandwidth, which is a challenge for the console. Microsoft’s solution for the main memory subsystem here means an idiosyncratic 320-bit interface, with ten 14 Gbps GDDR6 modules on the motherboard – six 2 GB chips and four 1 GB chips. it is divided for the developer is fascinating.
“Memory performance is asymmetric. It’s not something I could do with a PC, ”says Andrew Goossen. “Ten gigabytes of physical memory operates at 560 GB / s. We call this the optimal GPU memory. Six gigabytes work with it. 336 GB / s. This is what we call standard memory. Optimal and Standard GPUs provide the same performance for CPU, audio, and file I / O. The only hardware component that sees a difference is the GPU. ”
Games are allocated a total of 13.5 GB in how memory is allocated, which includes the full 10 GB of optimal GPU memory and 3.5 GB of standard memory. That leaves 2.5GB of GDDR6 memory from the slower pool for the operating system and front-end interface. From Microsoft’s perspective, it is still a unified storage system, although performance may differ. “Based on conversations with developers, games quickly fill their standard memory quota with CPU, audio, stack data, executable data, and script data, and developers prefer to do this when they can. fits. gives more bandwidth, ”Goossen said.
It sounds like a complex solution, especially if Microsoft itself has a more traditional solution with the One X’s large memory interface. But the desire to work with much faster GDDR6 memory posed some challenges. “When we spoke to the systems team, there was a lot of talk about the complexity of signal integrity and other things,” Goossen said. “We had a 384-bit interface on the Xbox One X, but now at these incredible speeds, 14 Gbps with GDDR6, we wanted to go as far as possible and we think 320 is a good compromise to get the best performance.” and at the same time building a system that actually works and that we can deliver. ”
Essential performance is well covered, but it’s not just about raw computing power – the feature set is essential, too. In 2016, a year before work on the Xbox One X was completed, the Xbox Chips team was already working on the X series. This is where the work on the next-gen feature architecture began, we’ll see now. in 2020 – just as a reminder of the time it takes to develop new technologies. Even then ray tracing was already on the agenda and the need for a revolutionary approach to memory was clear. This brings us to the second key point in the X-Series hardware design: a fundamental shift away from mechanical hard drives, completely towards solid-state storage.
You will learn more about this tomorrow.
Source : Twitter Feeds