Обложка канала

Spark in me - Internet, data science, math, deep learning, philosophy

2440 @snakers4

Канал про интересные мне темы - интернет - статистика - наука о данных Без рекламы и буллшита.

Spark in me - Internet, data science, math, deep learning, philosophy

4 года назад
Открыть в
Ada Lovelace GPUs Shows How Desperate Nvidia Is - AMD RDNA 3 Cost Comparison semianalysis.substack.com/p/ada-l…esperate TLDR - good for AI, bad for gaming (price-wise): Due to high wafer costs, GPU die costs are up massively, but the die is only a portion of a GPU’s total bill of materials (BOM). The BOM of a GPU also includes memory, packaging, VRMs, cooling, and various other board-level costs. When moving from the previous generation 3090/3090ti (GA102) to the new 4090 (AD102), these board-level costs remain the same. As such, the MSRP increase from $1499 to $1599 is enough for Nvidia to maintain margins and deliver substantial gains in performance per dollar. The MSRP cannot be compared directly as the 3090ti GPU sells for $999, or even less, meaning performance per dollar in traditional rasterization rendering is flat. .... In short, AMD saves a lot on die costs by forgoing AI and ray tracing fixed function accelerators and moving to smaller dies with advanced packaging. The advanced packaging cost is up significantly with AMD’s RDNA 3 N31 and N32 GPUs, but the small fan-out RDL packages are still very cheap relative to wafer and yield costs. Ultimately, AMD’s increased packaging costs are dwarfed by the savings they get from disaggregating memory controllers/infinity cache, utilizing cheaper N6 instead of N5, and higher yields. Memory BOM utilizes the full memory bus width using single-sided 16Gb G6 or 16Gb G6x memory. Nvidia likely has a worse cost structure in traditional rasterization gaming performance for the first time in nearly a decade. Nvidia is desperate to maintain margins, as shown by AD104’s 4080 12GB pricing and branding. They still have far too many GPUs in the channel. If Nvidia wants to maintain its market position, marketing and game partnership teams will need to emphasize areas where their GPUs perform better, such as ray tracing and AI-based rendering. We expect AMD to gain a decent market share in laptops with the N33 GPU and superior mobile APUs. On the desktop, the market share shift will depend on how many wafers AMD allocates to gaming GPUs versus Genoa and Bergamo Zen 4 server CPUs. We expect AMD to rise to 30% to 35% market share on discrete desktop GPUs. AMD could raise its margins aggressively from historical levels to well above 50%. While Nvidia will still retain an advantage in ray tracing and AI-based rendering techniques, many gamers care more about the games they play today than where the industry is headed.
Ada Lovelace GPUs Shows How Desperate Nvidia Is - AMD RDNA 3 Cost Comparison

Nvidia's first cost disadvantage versus AMD in a decade

SemiAnalysis