laptops for designing and training AI apps

see also:

• artificial intelligence
• computers for running artificial intelligence

• the main criteria is a fast GPU with lots of RAM (at least 8Gb) and lots of NVidia CUDA / tensor cores
• in comparison,
  • for video editing, you need 6Gb for 4K videos, 8Gb for 6K, and 12Gb for 8K videos ¹⁾
  • for 3D modelling, you need 6Gb for small models, 8Gb for medium models, 12Gb for large models, and 16Gb for extra large models
  • for AI image generation, you need at least 8Gb
  • for training LoRAs for AI image generation, you need 18-24Gb
  • for running 5Gb small LLMs such as PrivateGPT, you need at least 8Gb
  • for running larger LLMs, you may need at least 24Gb
  • for training LLMs, you will need ~8x48Gb
• CUDA capability 9 is a higher level than 8.9 and adds:
  • Tensor Memory Accelerator (TMA) unit
  • DPX Instructions for Accelerated Dynamic Programming
  • Distributed Shared Memory
  • Thread Block Cluster
  • Atomic addition operating on float2 and float4 floating point vectors in global memory
  • 32 max resident blocks per SM
  • 64 max resident warps per SM
  • 2048 max resident threads per SM
  • 228kb max shared memory per SM
  • 227kb max shared memory per thread block
  • seems the only Nvidia GPU in 2023 with level 9 is the Nvidia H100 Data Center GPU
• CUDA capability 8.9 has²⁾:
  • Streaming Multiprocessor (SM) consists of:
    • 128 FP32 cores for single-precision arithmetic operations
    • 2 FP64 cores for double-precision arithmetic operations
    • 64 INT32 cores for integer math
    • 16 special function units for single-precision floating-point transcendental functions
    • 4 warp schedulers
  • 24 max resident blocks per SM
  • 48 max resident warps per SM
  • 1536 max resident threads per SM
  • 100kb max shared memory per SM
  • 99kb shared memory per thread block

• NVIDIA® RTX™ 2000 Ada 8Gb is similar to GeForce RTX 4060 but ? around half the speed:³⁾
  • 128-bit bus; 3072 cores, CUDA 8.9, DirectX 12 Ultimate (12.2), OpenGL 4.6, Open CL 3.0, Vulkan 1,3,

• https://www.intel.com.au/content/www/au/en/products/docs/processors/core/13th-gen-core-mobile-brief.html
• all have 8 p-cores except i7-13650HX which has 6
• all have a base power of 55W and max of 157W
• all i9s and i7-13850HX support DDR5 memory up to 5600MT/s (other i7s support max 4800MT/s)
• support max 128Gb RAM
• all have Intel UHD graphics:
  • HDMI max 4096 x 2160 @ 60Hz
  • DP max: 7680 x 4320 @ 60Hz
  • eDP max: 5120 x 3200 @ 120Hz
  • max. 4 displays
  • DirectX 12.1, OpenGL 4.6; OpenCL 3.0
  • 845GFLOPS
  • max memory 64Gb
• 00HX and 50HX models are Intel vPro® Eligible for enterprises for remote management

• all have 6 p-cores
• all have 8 e-cores except i7-13620H which has 4
• all have 20 Total Processor Threads except i7-13620H which has 16
• all have 24MB Intel® Smart Cache (L3)
• all have a base power of 45W and max of 115W
• support max 64Gb RAM up to max 5200MT/s
• all have Iris XE graphics:
  • has 96 execution units vs 32 for Intel UHD graphics
  • 2230GFLOPS
  • max memory 32Gb
• 00H models are Intel vPro® Eligible for enterprises for remote management and extra hardware security measures

• most now have at east one Thunderbolt 4 port which supports display and power delivery (PD) - these use USB-C inrerfaces
• most also have legacy USB 3.2 Gen 2 Type-A ports which are extremely useful for all our older USB devices

• many now come with SD Express readers designed for SD Express cards which would allow speeds up to 985Mb/s
  • unfortunately when used with older SD card types, even UHS-II or UHS-III, the compatibility issues drop max speed back to UHS-I of 104Mb/s!
• a UHS-II SD card reader will read a UHS-II card at 156Mb/s and an SD Express card at up to only 104Mb/s

• most now have Wi-Fi 6E(802.11ax) (Dual band) 2*2

• most now have at least Bluetooth v5
• some have Bluetooth 5.3