/
Navigation
Chronicles
Browse all articles
Explore
Semantic exploration
Research
Entity momentum
Nexus
Correlations & relationships
Story Arc
Topic evolution
Drift Map
Semantic trajectory animation
Posts
Analysis & commentary
Pulse API
Tech news intelligence API
Browse
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
Concept Search
Semantic similarity search
High Impact Stories
Top coverage by position
Sentiment Analysis
Positive/negative coverage
Anomaly Detection
Unusual coverage patterns
Analysis
Rivalry Report
Compare two entities head-to-head
Semantic Pivots
Narrative discontinuities
Crisis Response
Event recovery patterns
Connected
Search: /
Command: ⌘K
Embeddings: large
TEXXR

Chronicles

The story behind the story

days · browse · Enter similar · o open

Nvidia unveils its A100 AI chip with 54B transistors and 5 petaflops of performance, about 20 times more than the previous-generation Volta

Dean Takahashi / VentureBeat :

VentureBeat Dean Takahashi

Context & Ripple Effects

This launch extends a cadence Nvidia set with the Tesla P100 in 2016, when it first paired a huge transistor count with high-bandwidth memory for deep learning. The A100 jumps to 54B transistors and roughly 5 petaflops — about 20x the previous-generation Volta — making it the biggest single-generation performance leap in the company's datacenter line.

The chip immediately drew head-to-head scrutiny against Intel and GraphCore on performance, economics, and software (ZDNet's Ampere comparison), and it went on to become the ~$10K GPU at the center of the generative AI boom, with an 80GB memory variant following within months for supercomputer deployments.

First-order effects

  • Buyers of AI training compute get a claimed ~20x performance step over Volta, resetting the price-performance baseline every cloud and supercomputer procurement now measures against.
  • Intel and GraphCore face an immediate benchmark gap: their competing parts are judged directly against Ampere on throughput, cost per inference, and software maturity.

Second-order effects

  • Nvidia converts the lead into a product ladder rather than a one-off part, shipping an A100 80GB variant for supercomputers and keeping customers inside its roadmap instead of waiting out competitors.
  • Rivals are pushed to compete on economics and software ecosystems rather than raw specs alone, since matching a 54B-transistor part on paper does not close the deployment gap.

Third-order effects

  • If the pattern holds, each architecture generation becomes the de facto standard for the next AI wave — the A100 did exactly that as the workhorse of generative AI, entrenching Nvidia's dominant machine-learning GPU share through software lock-in as much as silicon.
  • Datacenter AI compute consolidates around a single vendor's cadence, turning GPU supply into strategic infrastructure that governments and hyperscalers plan around.

The trend: Datacenter GPUs are evolving from graphics parts into the standardized utility layer of AI infrastructure, with each Nvidia architecture generation defining what the industry builds on next.

Discussion

  • @jwangark James Wang on x
    1. Moore's Law lives! Nvidia's new 54 billion transistor GPU continues the march of Moore's Law, tho costs are no longer staying constant. 2. In 2020, the world's #1 chip is from a ‘gaming’ company and #2 chip is from a ‘web retailer’. How the leaderboard has changed.. https://tw…
  • @etherealmind @etherealmind on x
    Couple of things about this keynote 1. Its from a kitchen, with the obvious “baking something” cheap gag. 2. there are no customers involved. 3. its less than 15 minutes long. This might be the perfect keynote for virtual events. cc @jtroyer https://twitter.com/...
  • @libertyrpf Liberty on x
    “A data center powered by five DGX A100 systems for AI training & inference running on just 28 kilowatts costing $1m can do the work of a data center with 50 DGX-1 systems for AI training and 600 CPU systems consuming 630 kilowatts & costing over $11m” https://blogs.nvidia.com/..…
  • @nvidia @nvidia on x
    Announcing the new NVIDIA Ampere Architecture, designed to be the heart of the elastic data center of the future: https://nvidianews.nvidia.com/ ... #GTC20
  • @nvidiaembedded NVIDIA Embedded on x
    Announcing the new @NVIDIA Jetson Xavier NX Developer Kit with 10X the power of Jetson TX2 and cloud-native support. Its unique combination of size, performance and power opens the door for bringing new #AI products to market fast! Buy now. https://www.nvidia.com/... #JetsonNX ht…
  • @benbajarin Ben Bajarin on x
    In data center accellerators, NVIDIA has ~65% share in Azure, AWS, and GCP. When it comes to deep learning training NVIDIA is the standard. https://twitter.com/...
  • @gavinsbaker Gavin Baker on x
    1) Is Ampere being 20x better than Volta good or bad for all of the AI silicon startups that are “10x” better than Volta on a single algorithm on one framework? 😀 Especially given that *none* of them except Habana have posted the 10x results on MLPerf. https://venturebeat.com/...