Every significant change to TensorCore.si is listed here, newest first, so you can see what was added or corrected and when. Individual articles also show the date they were last updated. Spotted something out of date? Email hi@tensorcore.si.
6 November 2026
- New guides: How AI works, from beginner to master and What is superintelligence?, plus a new Superintelligence section.
- New on the homepage: How AI actually runs on a computer. Seven diagrams explaining how AI works, the processor (CPU), RAM, VRAM, how they work together, the AI factory and AI agents, plus a memory speed ladder from SSD to HBM4.
5 November 2026
Data updates and corrections
- GPU specs database expanded from 12 to 28 chips. Added NVIDIA Rubin, GB200, GB300, H100 NVL, L40S, L4, RTX PRO 6000 Blackwell, RTX 5090, RTX 4090 and DGX Spark; AMD Instinct MI250X and MI455X; Intel Gaudi 3; AWS Trainium2 and Trainium3; and Google TPU v5e. New filters for vendor and for data-center vs desktop cards. Open the database.
- Rubin memory bandwidth corrected to 19.2 TB/s and NVLink 6 to 3 TB/s per GPU, following NVIDIA's revised specifications in September 2026. The earlier figures were 22 TB/s and 3.6 TB/s.
- VRAM calculator: 31 model presets, up from 20, including Llama 4, Gemma 3, Qwen3-Coder, Kimi K2 and Mistral Large 2, plus 9 more GPUs.
- AI Model Explorer: 31 models, up from 17, including GLM-4.5, MiniMax-M1, Phi-4 and Command A.
- Glossary: 87 terms, up from 37.
New pages and features
- LLM benchmarks by GPU: measured tokens per second for 14 GPUs, with our analysis of how close each comes to its memory bandwidth.
- An embeddable version of the VRAM calculator that other sites can add with a credit link.
- Editor profile, editorial policy and this updates page.
- Topic hubs with a recommended reading order on each category page.
- Dropdown menus for the Challenge, Games and AI Models sections.
New articles
- NVIDIA Rubin explained, AMD MI355X vs NVIDIA B200, B200 vs B300 and AMD vs NVIDIA for AI
- RTX 5090 for AI, NPU vs GPU vs TPU, TOPS vs TFLOPS and GPU vs CPU for AI
- What is CUDA?, What is DLSS? and Tensor Cores vs RT Cores
- How much it costs to train an LLM, prefill vs decode, FlashAttention explained and Groq LPU vs GPU
- Best GPU for local LLMs, KV cache explained, LLM quantization, TPU vs GPU, DGX Spark explained, H100 vs A100, H100 vs H200, B200 vs H100, GB200 NVL72 explained and NVFP4 explained
New tools
4 November 2026
- TensorCore.si launched with the interactive GPU teardown on the homepage, the Tensor Core challenge and certificate, five learning games, the AI Model Explorer, the glossary, the GPU specs database, the number format converter and Tensor Core facts.
- First guides: What is a Tensor Core?, Tensor Cores vs CUDA Cores, Tensor Core generations, precision formats, how a GPU works, Hopper vs Blackwell, what is HBM, NVLink and NVSwitch and how to read a GPU spec sheet.