Tensor cores significantly accelerate matrix operations in neural networks, offering substantial performance improvements over traditional CUDA cores, especially in MLP inference tasks.
About:
The metadata does not provide specific information about the author. The site is named 'Interplay of Light', which may also be the author's pseudonym or the primary focus of the website.
Website:
Incoming Links:
Subscribe to RSS:
The post discusses the implementation of spatial hashing to optimize raytraced ambient occlusion (RTAO) in 3D graphics. It highlights the limitations of traditional 3D texture methods, such as high memory usage, and introduces a s...
This blog post explores the performance impact of vertex shader exports in GPU rendering, specifically focusing on how increasing the number of float4 vertex exports affects drawcall costs. The author conducts an experiment render...
The blog post delves into the architecture of GPUs, focusing on the role of SIMD units in rendering tasks. It discusses the importance of optimizing VALU throughput and identifying bottlenecks caused by fixed function units. Techn...
Exploring the use of multilayer perceptrons in rendering, this post shares experiments on encoding radiance and depth, revealing challenges and potential in graphics programming.
The text discusses the interplay of light and the use of asynchronous computing to improve GPU utilization and performance. It explains how to pair tasks with the graphics queue and the compute queue, and the use of fences to coor...
Mesh shaders are an evolution of the geometry pipeline that removes fixed function units and replaces them with a simpler, programmable pipeline. They are responsible for reading and processing geometry data and feeding it to the ...
0The hidden cost of shader instructions
2025-01-19 •
The post discusses the 'hidden' cost of shader instructions, focusing on GCN/RDNA architectures. It covers the causes of the 'hidden' cost, such as no hardware support for the instruction, hardware implementation of the instructio...
0An introduction to workgraphs part 2: Performance
2024-09-09 •
The blog post explores the performance of workgraphs, focusing on a hybrid shadowing system and the implementation of FidelityFX's SSSR technique. It discusses the allocation of ring buffers for each workgraph node, the performanc...
0A quick introduction to workgraphs
2024-06-29 •
The text is a high-level tutorial on workgraphs, a new feature added to DirectX12 with hardware support from NVidia and AMD. It explains how to use workgraphs to enable a GPU to produce and consume work without involving the CPU i...
The text discusses the influence of the presentation 'Low-level thinking in high-level shading languages' and its followup on shader programming. It emphasizes the importance of understanding the compiler output and manual optimiz...
0A gentler introduction to ReSTIR
2023-12-17 •
The text is a qualitative discussion of ReSTIR, a technique for lighting a scene with a large number of shadowed lights. It explains the theory behind ReSTIR, the Weighted Reservoir Sampling technique, and the challenges of implem...
0Raytraced Order Independent Transparency part 2
2023-07-30 •
The blog post discusses the use of raytracing to achieve order independent transparency (OIT) and compares it to other OIT methods. It explores the use of anyhit shader in DXR to process all intersections through the acceleration ...
0Raytraced Order Independent Transparency
2023-07-15 •
The text discusses the implementation of raytraced transparency and compares it to other transparency techniques. It explains the process of raytracing, the cost and memory requirements, and the advantages and disadvantages of ray...
0Experimenting with fp16, part 2
2023-01-08 •
The author discusses the performance difference between enabling fp16 and fp32 for a particular shader. The shader uses variance and mean radiance estimation in the pixel’s neighbourhood to clip history. The author also discusses ...
0Experimenting with fp16 in shaders
2022-12-30 •
The text discusses the use of 16 bit floating point numbers and operations in shaders, focusing on the advantages and potential drawbacks of using fp16 representation. The author shares notes from their investigation into fp16 at ...
0Stream compaction using wave intrinsics
2022-12-25 •
The text discusses the importance of stream compaction in achieving good performance on the GPU, particularly when dealing with work generated on the GPU. It explains the use of wave intrinsics to efficiently compact waves, remove...