While Sony’s partnership with AMD for Project Amethyst has been an ongoing once since it was announced earlier this year, the two companies have now revealed more details about the partnership in a new video. Featuring PS5 and PS5 Pro lead architect Mark Cerny and SVP and GM of AMD’s Computing and Graphics Group, Jack Huynh, the video goes into detail about the breakthroughs that the two companies have achieved.
In the video, Huynh goes into the role machine learning players in modern game development when it comes to offering developers cleaner pipelines and more efficient ways to render visuals while still having the technological headroom needed to create the massive worlds that players want.
“The challenge comes in how we implement these systems,” said Cerny. “The neural networks found in technologies like FSR and PSSR are incredibly demanding on the GPU. They’re both computationally intensive and require speedy access to large amounts of memory. The nature of the GPU fights us here.”
The technical discussion between Cerny and Huynh goes into detail about how the design of modern GPUs can often end up creating bottlenecks, since the smaller chips that GPUs are often made with also means that problems that compute units have to tackle have to similarly be broken up into smaller “bite sized pieces”.
To tackle this issue, Huynh revealed that Sony’s and AMD’s partnership gave rise to a new technology referred to as Neural Arrays. The general idea behind the technology is to have compute units teaming up to tackle large problems together rather than each compute unit handling its own individual smaller problem.
“We’re not linking the entire GPU into one mega unit,” explained Huynh. “That would be a cable management nightmare. But we are connecting [compute units] within each shader engine in a smart, efficient way. And that changes the game for neural rendering. Bigger [machine learning] models, less overhead, more efficiency, and way more scalability as workloads grow.”
The efficiencies offered by Neural Arrays as a concept have been described by Cerny as being a game changer for developers, especially in the development of next-generation image upscaling and denoising technologies like FSR and PSSR. Huynh also noted that these efficiencies will also lead to brand new uses for ML that engineers have just started to imagine thanks to the recent breakthroughs between the two companies.
Ray tracing has also been one of the subjects of research for the partnership. However, Cerny noted that the current iterations of ray tracing have been hitting limits of what can be achieved with modern hardware. To deal with this, AMD and Sony have spent two years rethinking the path tracing pipeline, from hardware all the way to software.
“Earlier this year at Computex, we introduced Neural Radiance Caching, a key part of FSR Redstone,” said Huynh. “Now we’re building on that with Radiance Cores, a new dedicated hardware block designed for unified light transport. It handles ray tracing and path tracing in real time, pushing lighting performance to a whole new level. Together, these form a brand new rendering approach for AMD.”
Radiance Cores will essentially take over all of the technical responsibilities of ray traced lighting that compute units typically have to deal with along with managing their shader software. This, in turn, frees up the compute units to tackle other things, while Radiance Cores can focus on path tracing, ray tracing, and ray traversal, all of which tend to be quite compute heavy.
The final thing revealed in the video revolves around the constraints faced by modern GPUs when it comes to memory bandwidth. Dubbed Universal Compression, the feature can evaluate every piece of data headed to memory, and compresses it when possible. This means that memory bandwidth usage can be reduced dramatically, since only the most important data is sent through the memory bus.
“That means the GPU can deliver more detail, higher frame rates, and greater efficiency,” said Huynh. Cerny noted that this new technology will allow GPUs to even exceed the paper specs of its memory bandwidth thanks to the high level of efficiency offered by the compression technique.
“There’s a multitude of benefits from this, including lower power consumption, higher fidelity assets, and perhaps most importantly, the synergies that Universal Compression has with Neural Arrays and Radiance Cores, as we work to deliver the best possible experiences to gamers,” said Cerny.
These technologies are still quite new, however, and, at least for the time being, they only really exist in simulation form. However, the results from this partnership have seemingly been promising, with Cerny noting that we might get to see them in future console generations as well. Huynh noted that these technologies will also make their way on to other gaming platforms as well.
Sony and AMD Partnership Led to Development of Neural Arrays, Radiance Cores and Universal Compression
Source: Buzz Trends Updates
0 comments:
Post a Comment