Differences

This shows you the differences between two versions of the page.

--- en:multiasm:cs:chapter_3_5 [2025/12/02 12:52] – ktokarz
+++ en:multiasm:cs:chapter_3_5 [2025/12/02 13:28] (current) – [SIMT] ktokarz
@@ Line 25: / Line 25: @@
 ===== SIMT =====
-Single Instruction Multiple Threads. Originally defined as the subset of SIMD. In modern construction, Nvidia uses this execution model in their G80 architecture
+Single Instruction Multiple Threads. Originally defined as the subset of SIMD. The difference between SIMD and SIMT is that in pure SIMD, a single instruction operates on all elements of the vector in the same way. In SIMT, selected threads can be activated or deactivated. Instructions and data are processed only on the active threads, while the data remains unchanged on inactive threads.
-((https://www.nvidia.com/content/PDF/fermi_white_papers/NVIDIA_Fermi_Compute_Architecture_Whitepaper.pdf)) where multiple independent threads execute concurrently using a single instruction.
+We can find SIMT in modern constructions. Nvidia uses this execution model in their G80 architecture
+((https://www.nvidia.com/content/PDF/fermi_white_papers/NVIDIA_Fermi_Compute_Architecture_Whitepaper.pdf)), where multiple independent threads execute concurrently using a single instruction.
+In x64 architecture, the new vector instructions have masked variants, in which an operation can be turned on or off for selected vector elements. This is done by an additional operand called a mask or predicates.

en/multiasm/cs/chapter_3_5.1764679968.txt.gz · Last modified: 2025/12/02 12:52 by ktokarz