Saturday, June 22, 2024

AI accelerator

AI accelerator Neural Processing Unit NPU, Digital signal processor DSP trillion operations per second 

Apple A11 iPhone 8/8 Plus/X, Neural Engine, 600 GOPS 
Apple A12 iPhone XS/XR/iPad 8/iPad Air 3/A12X iPad Pro 3 ce/A12Z iPad Pro 4, Next-generation Neural Engine, 8 cores, 5 TOPS 
Apple A13 iPhone 11/iPad 9 ks, 8-core, 6 TOPS 
Apple A14 iPhone 12/iPad 10 hl/iPad Air 4 kh, 4-6 GB, 16-core, 11 TOPS 
  Apple M1 iPad Air 5/iPad Pro 5 ca, 8- GB, 16-core, 11 TOPS 
Apple A15 iPhone 13/13 Pro/14 kh, 4-6GB, 16-core, 15.8 TOPS 
  Apple M2 iPad Air 6 2024/iPad Pro 6, 8- GB, 16-core, 15.8 TOPS 
Apple A16 iPhone 14 Pro/15 ca, 6 GB, 16-core, 17 TOPS 
Apple A17 Pro iPhone 15 Pro, 8 GB, 16-core, 35 TOPS
 Apple M3 Mac 8 GB, 16-core, 18 TOPS 
Apple M4 iPad Pro 7, 8 GB, 38 TOPS 
Apple A18 Pro iPhone 16, 35 TOPS 

Qualcomm Snapdragon systems on chips: Snapdragon_8/8s Gen 3 (2024) 

Google Tensor with Tensor Processing Unit (Edge_TPU) used in Pixel Phones 

Microsoft Copilot+ PC Blog 

A guide to AI TOPS and NPU performance metrics 
Qualcomm Snapdragon X series Elite/Plus DSP 45 TOPS, GPU 3.8-4.6 TFLOPS 

AMD Ryzen AI 9 HX Series 300, Ryzen AI 300 series, 365/370 28W (15-54W) 50 TOPS, 375 55 TOPS, 

Intel Core Ultra 200V Series Intel Core Ultra 7 17W (8-37W): 256V 16GB, 258V 32GB 47 TOPS, 266V 16GB, 268V 32GB 48 TOPS 

Intel AMX XMX Advanced Matrix Extensions in CPU, Xe Matrix Extensions in GPU 

Microsoft Copilot+ PC 40 TOPS, 16 GB memory, 256 GB storage 

Apple Intelligence ChatGPT-4o Code Completion 16GB 

Best chatbot ChatGPT alternatives Anthropic Claude, Google Gemini for live data, Microsoft Copilot creative, Perplexity research, Inflection Pi personal, xAI Grok social, Llama open source, MetaAI fun, Poe more choices comparison 

No comments: