LogoLogo
Continuum WebsiteContinuum ApplicationsContinuum KnowledgeAxolotl Platform
Continuum Knowledge
Continuum Knowledge
  • Continuum
  • Data
    • Datasets
      • Pre Training Data
      • Types of Fine Tuning
      • Self Instruct Paper
      • Self-Alignment with Instruction Backtranslation
      • Systematic Evaluation of Instruction-Tuned Large Language Models on Open Datasets
      • Instruction Tuning
      • Instruction Fine Tuning - Alpagasus
      • Less is More For Alignment
      • Enhanced Supervised Fine Tuning
      • Visualising Data using t-SNE
      • UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
      • Training and Evaluation Datasets
      • What is perplexity?
  • MODELS
    • Foundation Models
      • The leaderboard
      • Foundation Models
      • LLama 2 - Analysis
      • Analysis of Llama 3
      • Llama 3.1 series
      • Google Gemini 1.5
      • Platypus: Quick, Cheap, and Powerful Refinement of LLMs
      • Mixtral of Experts
      • Mixture-of-Agents (MoA)
      • Phi 1.5
        • Refining the Art of AI Training: A Deep Dive into Phi 1.5's Innovative Approach
      • Phi 2.0
      • Phi-3 Technical Report
  • Training
    • The Fine Tuning Process
      • Why fine tune?
        • Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations?
        • Explanations in Fine Tuning
      • Tokenization
        • Tokenization Is More Than Compression
        • Tokenization - SentencePiece
        • Tokenization explore
        • Tokenizer Choice For LLM Training: Negligible or Crucial?
        • Getting the most out of your tokenizer for pre-training and domain adaptation
        • TokenMonster
      • Parameter Efficient Fine Tuning
        • P-Tuning
          • The Power of Scale for Parameter-Efficient Prompt Tuning
        • Prefix-Tuning: Optimizing Continuous Prompts for Generation
        • Harnessing the Power of PEFT: A Smarter Approach to Fine-tuning Pre-trained Models
        • What is Low-Rank Adaptation (LoRA) - explained by the inventor
        • Low Rank Adaptation (Lora)
        • Practical Tips for Fine-tuning LMs Using LoRA (Low-Rank Adaptation)
        • QLORA: Efficient Finetuning of Quantized LLMs
        • Bits and Bytes
        • The Magic behind Qlora
        • Practical Guide to LoRA: Tips and Tricks for Effective Model Adaptation
        • The quantization constant
        • QLORA: Efficient Finetuning of Quantized Language Models
        • QLORA and Fine-Tuning of Quantized Language Models (LMs)
        • ReLoRA: High-Rank Training Through Low-Rank Updates
        • SLoRA: Federated Parameter Efficient Fine-Tuning of Language Models
        • GaLora: Memory-Efficient LLM Training by Gradient Low-Rank Projection
      • Hyperparameters
        • Batch Size
        • Padding Tokens
        • Mixed precision training
        • FP8 Formats for Deep Learning
        • Floating Point Numbers
        • Batch Size and Model loss
        • Batch Normalisation
        • Rethinking Learning Rate Tuning in the Era of Language Models
        • Sample Packing
        • Gradient accumulation
        • A process for choosing the learning rate
        • Learning Rate Scheduler
        • Checkpoints
        • A Survey on Efficient Training of Transformers
        • Sequence Length Warmup
        • Understanding Training vs. Evaluation Data Splits
        • Cross-entropy loss
        • Weight Decay
        • Optimiser
        • Caching
      • Training Processes
        • Extending the context window
        • PyTorch Fully Sharded Data Parallel (FSDP)
        • Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation
        • YaRN: Efficient Context Window Extension of Large Language Models
        • Sliding Window Attention
        • LongRoPE
        • Reinforcement Learning
        • An introduction to reinforcement learning
        • Reinforcement Learning from Human Feedback (RLHF)
        • Direct Preference Optimization: Your Language Model is Secretly a Reward Model
  • INFERENCE
    • Why is inference important?
      • Grouped Query Attention
      • Key Value Cache
      • Flash Attention
      • Flash Attention 2
      • StreamingLLM
      • Paged Attention and vLLM
      • TensorRT-LLM
      • Torchscript
      • NVIDIA L40S GPU
      • Triton Inference Server - Introduction
      • Triton Inference Server
      • FiDO: Fusion-in-Decoder optimised for stronger performance and faster inference
      • Is PUE a useful measure of data centre performance?
      • SLORA
  • KNOWLEDGE
    • Vector Databases
      • A Comprehensive Survey on Vector Databases
      • Vector database management systems: Fundamental concepts, use-cases, and current challenges
      • Using the Output Embedding to Improve Language Models
      • Decoding Sentence-BERT
      • ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT
      • SimCSE: Simple Contrastive Learning of Sentence Embeddings
      • Questions Are All You Need to Train a Dense Passage Retriever
      • Improving Text Embeddings with Large Language Models
      • Massive Text Embedding Benchmark
      • RocketQAv2: A Joint Training Method for Dense Passage Retrieval and Passage Re-ranking
      • LLM2Vec: Large Language Models Are Secretly Powerful Text Encoders
      • Embedding and Fine-Tuning in Neural Language Models
      • Embedding Model Construction
      • Demystifying Embedding Spaces using Large Language Models
      • Fine-Tuning Llama for Multi-Stage Text Retrieval
      • Large Language Model Based Text Augmentation Enhanced Personality Detection Model
      • One Embedder, Any Task: Instruction-Finetuned Text Embeddings
      • Vector Databases are not the only solution
      • Knowledge Graphs
        • Harnessing Knowledge Graphs to Elevate AI: A Technical Exploration
        • Unifying Large Language Models and Knowledge Graphs: A Roadmap
      • Approximate Nearest Neighbor (ANN)
      • High Dimensional Data
      • Principal Component Analysis (PCA)
      • Vector Similarity Search - HNSW
      • FAISS (Facebook AI Similarity Search)
      • Unsupervised Dense Retrievers
    • Retrieval Augmented Generation
      • Retrieval-Augmented Generation for Large Language Models: A Survey
      • Fine-Tuning or Retrieval?
      • Revolutionising Information Retrieval: The Power of RAG in Language Models
      • A Survey on Retrieval-Augmented Text Generation
      • REALM: Retrieval-Augmented Language Model Pre-Training
      • Retrieve Anything To Augment Large Language Models
      • Generate Rather Than Retrieve: Large Language Models Are Strong Context Generators
      • Active Retrieval Augmented Generation
      • DSPy: LM Assertions: Enhancing Language Model Pipelines with Computational Constraints
      • DSPy: Compiling Declarative Language Model Calls
      • DSPy: In-Context Learning for Extreme Multi-Label Classification
      • Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs
      • HYDE: Revolutionising Search with Hypothetical Document Embeddings
      • Enhancing Recommender Systems with Large Language Model Reasoning Graphs
      • Retrieval Augmented Generation (RAG) versus fine tuning
      • RAFT: Adapting Language Model to Domain Specific RAG
      • Summarisation Methods and RAG
      • Lessons Learned on LLM RAG Solutions
      • Stanford: Retrieval Augmented Language Models
      • Overview of RAG Approaches with Vector Databases
      • Mastering Chunking in Retrieval-Augmented Generation (RAG) Systems
    • Semantic Routing
    • Resource Description Framework (RDF)
  • AGENTS
    • What is agency?
      • Rephrase and Respond: Let Large Language Models Ask Better Questions for Themselves
      • Types of Agents
      • The risk of AI agency
      • Understanding Personality in Large Language Models: A New Frontier in AI Psychology
      • AI Agents - Reasoning, Planning, and Tool Calling
      • Personality and Brand
      • Agent Interaction via APIs
      • Bridging Minds and Machines: The Legacy of Newell, Shaw, and Simon
      • A Survey on Language Model based Autonomous Agents
      • Large Language Models as Agents
      • AI Reasoning: A Deep Dive into Chain-of-Thought Prompting
      • Enhancing AI Reasoning with Self-Taught Reasoner (STaR)
      • Exploring the Frontier of AI: The "Tree of Thoughts" Framework
      • Toolformer: Revolutionising Language Models with API Integration - An Analysis
      • TaskMatrix.AI: Bridging Foundational AI Models with Specialised Systems for Enhanced Task Completion
      • Unleashing the Power of LLMs in API Integration: The Rise of Gorilla
      • Andrew Ng's presentation on AI agents
      • Making AI accessible with Andrej Karpathy and Stephanie Zhan
  • Regulation and Ethics
    • Regulation and Ethics
      • Privacy
      • Detecting AI Generated content
      • Navigating the IP Maze in AI: The Convergence of Blockchain, Web 3.0, and LLMs
      • Adverse Reactions to generative AI
      • Navigating the Ethical Minefield: The Challenge of Security in Large Language Models
      • Navigating the Uncharted Waters: The Risks of Autonomous AI in Military Decision-Making
  • DISRUPTION
    • Data Architecture
      • What is a data pipeline?
      • What is Reverse ETL?
      • Unstructured Data and Generatve AI
      • Resource Description Framework (RDF)
      • Integrating generative AI with the Semantic Web
    • Search
      • BM25 - Search Engine Ranking Function
      • BERT as a reranking engine
      • BERT and Google
      • Generative Engine Optimisation (GEO)
      • Billion-scale similarity search with GPUs
      • FOLLOWIR: Evaluating and Teaching Information Retrieval Models to Follow Instructions
      • Neural Collaborative Filtering
      • Federated Neural Collaborative Filtering
      • Latent Space versus Embedding Space
      • Improving Text Embeddings with Large Language Models
    • Recommendation Engines
      • On Interpretation and Measurement of Soft Attributes for Recommendation
      • A Survey on Large Language Models for Recommendation
      • Model driven recommendation systems
      • Recommender AI Agent: Integrating Large Language Models for Interactive Recommendations
      • Foundation Models for Recommender Systems
      • Exploring the Impact of Large Language Models on Recommender Systems: An Extensive Review
      • AI driven recommendations - harming autonomy?
    • Logging
      • A Taxonomy of Anomalies in Log Data
      • Deeplog
      • LogBERT: Log Anomaly Detection via BERT
      • Experience Report: Deep Learning-based System Log Analysis for Anomaly Detection
      • Log-based Anomaly Detection with Deep Learning: How Far Are We?
      • Deep Learning for Anomaly Detection in Log Data: A Survey
      • LogGPT
      • Adaptive Semantic Gate Networks (ASGNet) for log-based anomaly diagnosis
  • Infrastructure
    • The modern data centre
      • Enhancing Data Centre Efficiency: Strategies to Improve PUE
      • TCO of NVIDIA GPUs and falling barriers to entry
      • Maximising GPU Utilisation with Kubernetes and NVIDIA GPU Operator
      • Data Centres
      • Liquid Cooling
    • Servers and Chips
      • The NVIDIA H100 GPU
      • NVIDIA H100 NVL
      • Lambda Hyperplane 8-H100
      • NVIDIA DGX Servers
      • NVIDIA DGX-2
      • NVIDIA DGX H-100 System
      • NVLink Switch
      • Tensor Cores
      • NVIDIA Grace Hopper Superchip
      • NVIDIA Grace CPU Superchip
      • NVIDIA GB200 NVL72
      • Hopper versus Blackwell
      • HGX: High-Performance GPU Platforms
      • ARM Chips
      • ARM versus x86
      • RISC versus CISC
      • Introduction to RISC-V
    • Networking and Connectivity
      • Infiniband versus Ethernet
      • NVIDIA Quantum InfiniBand
      • PCIe (Peripheral Component Interconnect Express)
      • NVIDIA ConnectX InfiniBand adapters
      • NVMe (Non-Volatile Memory Express)
      • NVMe over Fabrics (NVMe-oF)
      • NVIDIA Spectrum-X
      • NVIDIA GPUDirect
      • Evaluating Modern GPU Interconnect
      • Scalable Hierarchical Aggregation and Reduction Protocol (SHARP)
      • Next-generation networking in AI environments
      • NVIDIA Collective Communications Library (NCCL)
    • Data and Memory
      • NVIDIA BlueField Data Processing Units (DPUs)
      • Remote Direct Memory Access (RDMA)
      • High Bandwidth Memory (HBM3)
      • Flash Memory
      • Model Requirements
      • Calculating GPU memory for serving LLMs
      • Transformer training costs
      • GPU Performance Optimisation
    • Libraries and Complements
      • NVIDIA Base Command
      • NVIDIA AI Enterprise
      • CUDA - NVIDIA GTC 2024 presentation
      • RAPIDs
      • RAFT
    • Vast Data Platform
      • Vast Datastore
      • Vast Database
      • Vast Data Engine
      • DASE (Disaggregated and Shared Everything)
      • Dremio and VAST Data
    • Storage
      • WEKA: A High-Performance Storage Solution for AI Workloads
      • Introduction to NVIDIA GPUDirect Storage (GDS)
        • GDS cuFile API
      • NVIDIA Magnum IO GPUDirect Storage (GDS)
      • Vectors in Memory
Powered by GitBook
LogoLogo

Continuum - Accelerated Artificial Intelligence

  • Continuum Website
  • Axolotl Platform

Copyright Continuum Labs - 2023

On this page
  • Advantages of Flash Memory
  • Limitations of Flash Memory
  • Cost and Competitive Advantages
  • Applications
  • Structure and Working Principle
  • Flash cells come in different types based on the number of bits they can store
  • Flash memory supports three basic operations
  • Flash Translation Layer (FTL)
  • Current Trends in Flash Memory
  • Future of Storage
  • Storage-Class Memory (SCM)
  • Conclusion
  • Flash Memory Challenges
  • Related work section

Was this helpful?

  1. Infrastructure
  2. Data and Memory

Flash Memory

Flash memory is a type of non-volatile memory that has gained widespread popularity due to its unique characteristics and advantages over traditional storage technologies.

Flash memory has revolutionised the storage industry, enabling faster, more compact, and energy-efficient storage solutions compared to traditional hard disk drives (HDDs).

Flash memory is based on NAND flash technology, which stores data in an array of memory cells.

Each cell consists of a floating gate transistor that can store electrical charges, representing binary data. NAND flash offers high density and fast read access, making it suitable for storage applications.

Advantages of Flash Memory

  • Non-volatile: Flash memory retains data even when power is turned off.

  • High speed: Flash memory offers fast read and write speeds, significantly faster than traditional hard disk drives (HDDs).

  • Durability: With no moving parts, flash memory is more resilient to physical shocks and vibrations compared to HDDs.

  • Low power consumption: Flash memory consumes less power than HDDs, making it suitable for portable devices.

  • Compact size: Flash memory chips are small and lightweight, enabling the development of compact storage devices.

Limitations of Flash Memory

  • Limited write cycles: Flash memory cells have a finite number of write cycles before they wear out, typically ranging from 10,000 to 100,000 cycles.

  • Higher cost per gigabyte: Flash memory is generally more expensive than HDDs on a per-gigabyte basis.

  • Slower write speeds compared to RAM: Although faster than HDDs, flash memory has slower write speeds compared to volatile memory like RAM.

Cost and Competitive Advantages

  • The cost of flash memory has been steadily decreasing over the years, making it more affordable for a wide range of applications.

  • Factors influencing cost include capacity, type, performance, and market trends.

  • Flash memory offers competitive advantages such as speed, durability, energy efficiency, compact size, silent operation, and long-term performance.

Applications

  • Flash memory is widely used in various devices and applications, including:

    • Solid-state drives (SSDs) for computers and servers

    • USB flash drives and memory cards for portable storage

    • Mobile devices like smartphones, tablets, and digital cameras

    • Embedded systems and IoT devices

    • Automotive and industrial applications

Flash memory has revolutionised the storage industry, providing faster, more durable, and energy-efficient storage solutions. Its advantages have made it a preferred choice for a wide range of applications, from consumer electronics to enterprise storage systems.

Structure and Working Principle

  • Flash memory is made up of an array of memory cells, each consisting of a floating gate transistor.

  • The floating gate is electrically isolated by an oxide layer, allowing it to store electrons.

  • When electrons are stored in the floating gate, it changes the threshold voltage of the transistor, representing a binary "0" or "1".

  • The two main types of flash memory are NAND and NOR, with NAND being more commonly used in storage devices.

NAND Flash Memory

  • NAND flash memory is organised in a series of blocks, each containing multiple pages.

  • It offers higher storage density and faster write and erase speeds compared to NOR flash.

  • NAND flash is used in solid-state drives (SSDs), USB flash drives, memory cards, and other storage devices.

  • It is well-suited for sequential read and write operations, making it ideal for storing large files like photos, videos, and documents.

NOR Flash Memory

  • NOR flash memory allows random access to individual memory locations, similar to RAM.

  • It offers faster read speeds compared to NAND flash but has slower write and erase speeds.

  • NOR flash is used in applications that require fast random access, such as storing program code in embedded systems.

Flash cells come in different types based on the number of bits they can store

  • Single-Level Cell (SLC): Stores one bit per cell, offering the highest performance and endurance.

  • Multi-Level Cell (MLC): Stores two bits per cell, providing a balance between capacity and performance.

  • Triple-Level Cell (TLC): Stores three bits per cell, increasing storage density but with lower performance and endurance compared to SLC and MLC.

  • Quad-Level Cell (QLC): Stores four bits per cell, further increasing storage density but with reduced performance and endurance.

Flash memory supports three basic operations

  • Read: Retrieving data from the flash cells.

  • Write (Program): Writing data to the flash cells.

  • Erase: Clearing the contents of a block of flash cells.

Flash memory follows a write-once, erase-many principle.

Data can only be written to empty cells, and erasing is performed at the block level, which is much larger than the page size used for reads and writes.

Flash Translation Layer (FTL)

The Flash Translation Layer is a critical component of flash-based storage systems.

It manages the mapping between logical addresses used by the host system and the physical addresses of the flash memory. The FTL handles wear leveling, garbage collection, and error correction to optimize the performance and longevity of the flash memory.

Current Trends in Flash Memory

NAND 3D

NAND technology has become prevalent in recent years. It stacks multiple layers of flash cells vertically, enabling higher storage densities and lower costs per gigabyte.

3D NAND has allowed for the development of high-capacity SSDs and has reduced the price gap between SSDs and HDDs.

NVMe and PCIe Interfaces

NVMe (Non-Volatile Memory Express) and PCIe (Peripheral Component Interconnect Express) interfaces have gained popularity for flash-based storage.

NVMe is a protocol designed specifically for SSDs, leveraging the high-speed PCIe bus to deliver low-latency and high-throughput performance. NVMe SSDs offer significantly faster data transfer rates compared to traditional SATA SSDs.

Datacentre and Enterprise Adoption

Flash memory has seen widespread adoption in datacentres and enterprise environments.

The high performance, low latency, and energy efficiency of flash storage have made it a preferred choice for demanding workloads such as databases, virtualisation, and big data analytics.

All-flash storage arrays and NVMe-based storage systems are becoming increasingly common in enterprise storage infrastructures.

Future of Storage

Continued Density Increases

The future of flash memory is expected to bring further increases in storage density.

Advancements in 3D NAND technology, such as increasing the number of layers and implementing novel cell architectures like split-gate and charge trap flash, will enable even higher storage capacities in smaller form factors.

Improved Performance and Endurance

Ongoing research and development efforts aim to enhance the performance and endurance of flash memory.

Techniques such as advanced error correction algorithms, intelligent wear leveling, and optimised garbage collection mechanisms will contribute to faster and more reliable flash-based storage solutions.

Emerging Technologies Several emerging technologies show promise for the future of storage:

  • MRAM (Magnetoresistive Random Access Memory): MRAM offers non-volatility, high speed, and unlimited endurance, making it a potential candidate for storage-class memory.

  • ReRAM (Resistive Random Access Memory): ReRAM exhibits fast switching speeds, low power consumption, and high scalability, making it a promising alternative to flash memory.

  • DNA Storage: DNA-based storage is an experimental technology that aims to store vast amounts of data in synthetic DNA molecules, offering extreme density and long-term stability.

Storage-Class Memory (SCM)

Storage-class memory, also known as persistent memory, is an emerging category that bridges the gap between volatile memory (DRAM) and non-volatile storage (SSD).

SCM technologies like Intel Optane offer near-DRAM performance while providing non-volatility and larger capacities.

SCM has the potential to revolutionise data storage and processing by enabling new architectures and applications.

Conclusion

Flash memory has transformed the storage landscape, offering high performance, energy efficiency, and increasing storage densities.

As flash technology continues to evolve, with advancements in 3D NAND, NVMe interfaces, and datacentre adoption, it is poised to remain a dominant force in the storage industry.

The future of storage looks promising, with ongoing improvements in density, performance, and endurance, as well as the emergence of new technologies like MRAM, ReRAM, and DNA storage.

The advent of storage-class memory further blurs the line between memory and storage, opening up new possibilities for data-intensive applications.

As the demand for fast, reliable, and high-capacity storage continues to grow, flash memory and its successors will play a crucial role in shaping the future of computing and data storage.

Flash Memory Challenges

The paper highlights two unique characteristics of flash memory that introduce challenges in its management:

Write-once with bulk erasure

Once a page is written with data, it cannot be overwritten unless the entire block it belongs to is erased.

When data on a page needs to be updated, the new data is written to a different free page, and the old page is marked as invalid.

This approach is called out-place updating. As a result, the physical location of data on flash memory can change over time, requiring a logical-to-physical address mapping to keep track of where each piece of data is stored.

Wear-leveling

Each block on flash memory has a limited number of erase cycles it can endure before becoming unreliable (typically around 1 million erase cycles).

To prevent certain blocks from wearing out faster than others, wear-leveling techniques are used to evenly distribute erase operations across all blocks, extending the overall lifetime of the flash memory.

These characteristics lead to two major issues in flash memory management:

Address Translation

Since the physical location of data can change due to out-place updates, an efficient method is needed to map logical addresses (used by the host system) to physical addresses (actual locations on flash memory).

Space Management

As data is updated and invalidated, the number of free pages on flash memory decreases.

Garbage collection is needed to reclaim the space occupied by invalid pages, which involves copying valid data to other free pages and erasing the blocks containing invalid pages. Additionally, wear-leveling policies are used to evenly distribute erase operations across blocks to maximize the flash memory's lifetime.

RAM Space Requirements vs. On-line Performance

The paper then discusses a common technique used in flash memory storage systems: using two RAM-resident tables for address translation and space management. The address translation table maps logical addresses to physical addresses, while the space management table keeps track of the status of each page (free, valid, or invalid).

Related work section

Block-device emulations: Some designs propose emulating flash memory as a block device so that existing file systems designed for disk storage can access flash memory without modifications.

Native flash-memory file systems: Other designs propose flash-memory-friendly file systems that do not impose disk-aware data structures on flash memory management.

NOR vs. NAND flash memory: Storage system designs for NOR and NAND flash memory differ significantly because NOR flash memory supports bit-wise operations, while NAND flash memory operations are page-oriented.

Challenges in flash memory management: Researchers have been investigating how to address the new challenges introduced by the characteristics of flash memory when integrating it into existing storage systems.

PreviousHigh Bandwidth Memory (HBM3)NextModel Requirements

Last updated 11 months ago

Was this helpful?

Page cover image