Recommended Free Tools
Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
NVIDIA’s GTC 2024 keynote on March 18, 2024 introduced the Blackwell architecture and the company’s B200, GB200 and GB200 NVL72 platforms. The event was covered live by AnandTech in an article titled The NVIDIA GTC 2024 Keynote Live Blog (Starts at 1:00pm PT/20:00 UTC). NVIDIA’s announcements focused on data-center AI, networking and enterprise software—not a consumer GeForce launch.
The original AnandTech URL now redirects to the AnandTech forums, so it should be treated as an archival reference rather than a guaranteed working copy of the live blog. The official NVIDIA GTC portal and NVIDIA’s Blackwell launch announcement preserve the key announcements.
What was the NVIDIA GTC 2024 live blog?
The live blog was timestamped event coverage published by AnandTech and authored by Ryan Smith and Gavin Bonshor. It followed Jensen Huang’s keynote as announcements were made, rather than presenting a polished retrospective after the event.
- Date: March 18, 2024
- Start time: 1:00 p.m. Pacific, 4:00 p.m. Eastern, or 20:00 UTC
- Venue: SAP Center in San Jose, California
- Publisher: AnandTech
The event was significant because NVIDIA’s Hopper generation, including the H100, had become central to the generative-AI infrastructure boom. GTC had also grown beyond a conventional developer conference into a major enterprise and industry event. The move to the SAP Center reflected that scale.
#1 Best Overall
- AI Performance: 767 AI TOPS
- OC mode: 2632 MHz (OC mode)/ 2602 MHz (Default mode)
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Axial-tech fan design features a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- A 2.5-slot design maximizes compatibility and cooling efficiency for superior performance in small chassis
The original article is available by title through the AnandTech GTC 2024 archive, while a search-result mirror is hosted at inkl.com. Neither should be assumed to reproduce the original page perfectly.
The headline announcement: Blackwell
The keynote’s central announcement was Blackwell, NVIDIA’s successor to Hopper. NVIDIA said the architecture used 208 billion transistors, two large dies connected by a 10 TB/s chip-to-chip link, and a custom 4NP TSMC manufacturing process.
Those are NVIDIA-published specifications, not independent measurements. The two-die design was presented as one unified GPU, allowing NVIDIA to build a very large accelerator while maintaining high-speed communication between its dies.
NVIDIA highlighted several architectural features:
- A second-generation Transformer Engine for AI workloads.
- Support for 4-bit inference.
- Fifth-generation NVLink for communication between accelerators.
- Reliability, availability and serviceability features for data-center operation.
- Confidential-computing and security capabilities.
- A decompression engine intended to accelerate data analytics.
NVIDIA also said Blackwell could support models scaling to 10 trillion parameters. That describes a claimed platform capability; it does not mean models of that size were commonly deployed, affordable or practical at the time.
B200, GB200 and GB200 NVL72 explained
These names describe different layers of the platform and should not be treated as interchangeable.
| Product | What it is |
|---|---|
| B200 | The Blackwell-generation Tensor Core GPU. |
| GB200 | A Grace Blackwell Superchip combining two B200 GPUs with one NVIDIA Grace CPU, connected by a 900 GB/s chip-to-chip interconnect. |
| GB200 NVL72 | A liquid-cooled rack-scale system containing 36 GB200 Superchips, 72 Blackwell GPUs and 36 Grace CPUs. |
The GB200 NVL72 also includes fifth-generation NVLink and BlueField-3 DPUs for networking, storage, security and infrastructure management. NVIDIA described the rack as operating like a single large GPU, with 1.4 exaflops of AI performance and 30 TB of fast memory.
Those performance and memory figures are NVIDIA’s stated system specifications. The NVL72 is not a desktop graphics card or a single B200 GPU; it is a complete, high-density data-center system requiring specialized power, cooling, networking and facilities.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #2
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5060
- Integrated with 8GB GDDR7 128bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
How to interpret NVIDIA’s performance claims
NVIDIA claimed that the GB200 NVL72 could deliver up to 30 times the performance of the same number of H100 GPUs for certain large-language-model inference workloads. It also claimed up to 25 times lower cost and energy consumption.
These are workload-specific vendor claims, not universal benchmarks. The outcome can change with the model, precision, batch size, software stack, utilization, networking configuration and baseline being compared. “Cost” may refer to a particular cost-per-inference or system scenario rather than the purchase price of the complete infrastructure.
The keynote therefore established NVIDIA’s intended value proposition, not a guarantee that every Blackwell workload would be 30 times faster or 25 times cheaper. Training, fine-tuning, retrieval, recommendation, simulation and ordinary CUDA workloads may produce different results.
Networking was part of the product
Blackwell was presented as a full-stack AI infrastructure platform rather than merely a faster accelerator. NVIDIA announced or highlighted:
- Fifth-generation NVLink, with NVIDIA stating up to 1.8 TB/s of bidirectional throughput per GPU.
- NVIDIA Quantum-X800 InfiniBand.
- NVIDIA Spectrum-X800 Ethernet.
- Networking speeds of up to 800 Gb/s for the new platforms.
- BlueField-3 DPUs for networking, storage, security and infrastructure services.
High-speed interconnects matter because large AI models are distributed across many accelerators. At that scale, communication, synchronization and data movement can limit real-world performance just as much as the arithmetic capability of an individual GPU.
Software announcements
NVIDIA used GTC 2024 to promote the software surrounding its hardware. The lineup included:
- NVIDIA NIM inference microservices for packaging and deploying supported model-serving workloads.
- NVIDIA AI Enterprise, positioned as an end-to-end production AI platform.
- TensorRT-LLM for optimizing large-language-model inference.
- NeMo Megatron and related model-development tools.
- CUDA and NVIDIA’s broader AI software libraries.
- NVIDIA DGX Cloud for managed access to NVIDIA AI infrastructure.
The strategy was to make NVIDIA valuable at multiple layers: GPU silicon, CPU-plus-GPU systems, networking, cloud infrastructure, enterprise support and model deployment. NIM and AI Enterprise may simplify parts of a production deployment, but they do not make every model portable, every workload simple or every deployment inexpensive.
Rank #3
- Powered by the NVIDIA Blackwell architecture and DLSS 4. System Requirements: Minimum 850W PSU with 16-pin 12V-2x6 (12VHPWR) connector required. Verify before purchasing.
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability. Compatibility: 348mm (13.7") length, 3.6 slots, 4.3 lbs. Confirm case clearance and slot spacing. GPU bracket included.
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.6-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
Relevant official resources include NVIDIA NIM, NVIDIA AI Enterprise and NVIDIA DGX Cloud.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Cloud partners and availability
NVIDIA said AWS, Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure were among the first cloud providers expected to offer Blackwell-powered instances. It also named Applied Digital, CoreWeave, Crusoe, IBM Cloud, Lambda and Nebius.
The launch announcement said partner products were expected later in 2024. That wording did not establish immediate, universal availability, retail pricing or access in every cloud region. Availability can differ by provider, instance type, geography, reservation status and customer eligibility.
For a small team or independent developer, cloud access or a hosted inference service is generally more realistic than purchasing a GB200 rack. Official starting points include AWS accelerated computing, Azure virtual machines, Google Cloud GPUs, Oracle Cloud GPUs, CoreWeave and Lambda. Current pricing and Blackwell availability must be checked with each provider.
What GTC 2024 meant for different readers
AI labs and cloud providers
The announcement targeted organizations training and serving large models at scale. The combination of faster accelerators, NVLink, high-speed networking and rack-level integration was intended to reduce the cost and time of distributed AI workloads.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Enterprise IT teams
Enterprises had to evaluate more than GPU specifications. Power, liquid cooling, rack density, networking, software licensing, support, facility capacity and long-term utilization all affect the economics of deployment.
Developers
Developers gained a clearer picture of NVIDIA’s preferred stack: CUDA, TensorRT-LLM, NIM, NeMo, AI Enterprise and cloud-based DGX services. That stack can reduce integration work, but it also increases reliance on NVIDIA-specific software and hardware.
Rank #4
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5070 Ti
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
Consumers and PC buyers
The keynote was not a conventional consumer graphics-card launch. It did not establish a retail product, consumer price or immediate desktop availability for B200 or GB200 hardware. A reader shopping for a gaming GPU should not interpret the event as a GeForce announcement.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the keynote did not establish
- It did not prove that every AI workload would receive the advertised speedup.
- It did not make Blackwell hardware immediately obtainable by small teams.
- It did not eliminate the importance of software optimization, power, cooling, networking or data-center construction.
- It did not make H100 or H200 systems automatically obsolete.
- It did not provide a normal consumer GPU price.
- It did not establish that every named cloud provider had Blackwell instances available in every region.
Existing Hopper deployments may remain economically sensible because organizations have already paid for facilities, software integration, contracts and operational expertise. The value of upgrading depends on workload, utilization, energy costs, deployment timing and the availability of compatible systems.
A compact timeline of the announcement
- Before the keynote: Expectations centered on a Hopper successor, new data-center hardware, networking and software.
- March 18, 2024: Jensen Huang presented the keynote at the SAP Center in San Jose.
- Architecture reveal: NVIDIA introduced Blackwell and its two-die design, Transformer Engine and new interconnect features.
- System announcements: NVIDIA introduced B200, GB200 and the rack-scale GB200 NVL72.
- Platform expansion: Networking products, NIM, AI Enterprise, DGX Cloud and cloud partners rounded out the launch.
- Availability message: NVIDIA said partner products were expected later in 2024, without establishing universal immediate access.
Where to revisit the event
The official NVIDIA GTC page is the best starting point for keynote and conference material. NVIDIA’s Blackwell platform page and GB200 NVL72 page provide product information.
The original AnandTech live-blog URL is available here as an archival reference, but it currently redirects to the AnandTech forums. That archival uncertainty is important: a modern reconstruction can explain the event, but it should not imply that the original timestamped page remains fully accessible.
Frequently Asked Questions
When was the NVIDIA GTC 2024 keynote?
It took place on March 18, 2024, beginning at 1:00 p.m. Pacific time at the SAP Center in San Jose, California.
What was the main announcement?
NVIDIA introduced the Blackwell architecture and the B200, GB200 and GB200 NVL72 data-center platforms.
Free tools Windows power users keep installed
One-click scans. No signup required.
Was Blackwell a consumer graphics-card launch?
No. The keynote focused on enterprise and data-center AI infrastructure, not a retail GeForce product.
Is the AnandTech live blog still online?
The original URL currently redirects to the AnandTech forums. The AnandTech GTC archive and NVIDIA’s official GTC resources are better starting points for surviving material.
Were NVIDIA’s performance claims independently verified?
The headline 30-times performance and 25-times cost-and-energy figures were NVIDIA’s workload-specific vendor claims. They should not be treated as universal benchmarks.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minute
