2024 was Nvidia’s year. Its command of mindshare and market share was unequaled among tech vendors. Here’s a recap of some of the key events of 2024 that highlight just how powerful the world’s most dominant chip player is.
Nvidia teams up with Cisco
On February 6, Nvidia announced a collaboration with Cisco to deliver AI infrastructure solutions for data centers. The news was an expansion of an existing partnership to offer integrated software and networking hardware that promises to help customers more easily spin up infrastructure to support AI applications.
The agreement deepens both companies’ strategy to expand the role of Ethernet networking for AI workloads in the enterprise. It also gives both companies access to each other’s sales and support systems.
As part of the announcement, Nvidia’s newest Tensor Core GPUs will be available in Cisco’s current M7 Unified computing System (UCS) rack and blade servers, including Cisco UCS X-Series and UCS X-Series Direct, to support AI and data-intensive workloads in the data center and at the edge.
Data center revenue spikes
Nvidia started the year with a bang, releasing financials for its fiscal year 2024 ended January 28, 2024. Its 2024 revenue came in at $60.9 billion, up 126% from the previous year. Data center revenue, in particular, jumped to $47.5 billion, a gain of 217% compared to 2023 numbers. GAAP earnings per diluted share were $11.93, which is up 586% year-over-year.
Nvidia GTC 2024
The month of March was dominated by the company’s GTC 2024 conference. The big news was the introduction of the Blackwell architecture, which Nvidia claimed would offer four times the performance of Hopper, the current architecture.
And there was much more announced at the show, including new InfiniBand networking cards and switches, significant upgrades and new features in its Enterprise AI software, and a storage partner validation service.
Nvidia also introduced its next-generation AI supercomputer, the DGX SuperPOD10. Powered by GB200 Grace Blackwell superchips, the SuperPOD10 is capable of processing trillion-parameter models for large-scale generative AI with 11.5 exaFLOPS using FP4 data types and 240 TB of fast memory.
Nvidia stock splits
For its first fiscal quarter 2025, which ended on April 28, 2024, Nvidia reported revenue of $26 billion, up 18% from the previous quarter and up 262% year-over-year. Operating income surged to $16.9 billion, a 24% increase from the previous quarter and a 690% increase from the year prior. GAAP earnings per diluted share were $5.98, a 21% increase from the previous quarter and a 629% increase from a year ago. All of these numbers significantly exceeded expectations from Wall Street analysts.
Nvidia also announced a 10-to-1 stock split, effective in June. Nvidia stock had seen a massive run up to over $1,000 per share.
Blackwell systems emerge, Rubin is teased
At the massive Computex IT hardware show, Nvidia unveiled new Blackwell-powered systems that, it said, will allow enterprises to build “AI factories” and data centers to drive the next wave of generative AI.
Nvidia also announced that its Nvidia MGX modular reference design platform, announced at the 2003 Computex show, now supports Blackwell products. It launched the new Nvidia GB200 NVL2 platform, a smaller version of the GB200 NVL72 introduced in March, that it said will speed up data processing by up to 18x, with 8x better energy efficiency compared to using x86 CPUs.
During his Computex keynote, CEO Jensen Huang disclosed the next generation microarchitecture, codenamed Rubin. Very little is known about the new platform. The Rubin AI platform will use HBM4 memory (which is not even out yet) and NVLink 6 Switch, operating at 3,600 Gbps.
Huang also introduced a new ARM-based CPU called Vera, which will be a part of a new accelerator board called Vera Rubin, just like the Grace Hopper chips combine a Grace CPU and Hopper GPU.
Strong numbers override heat rumors
For the second fiscal quarter ended July 28, revenue was $30 billion, up 15% from the previous quarter and up 122% year-over-year. GAAP earnings per diluted share was $0.67, up 12% from the previous quarter and up 168% from a year ago. Data center revenue reached $26.3 billion, a 154% increase from the previous year.
Prior to the earnings release, Nvidia was the subject of rumors and gossip about heat issues surrounding the manufacture of its new Blackwell processors. If there were any problems, they were quickly fixed and the chips remain on schedule for delivery at the end of 2024.
Anticompetitive issues arise
The US Department of Justice announced it was initiating an investigation into anticompetitive practices by Nvidia, which analysts pointed out didn’t make sense because Nvidia had no real competitors to begin with.
Oracle buys Blackwell GPUs for its cloud
Oracle announced plans to purchase 131,072 Nvidia Blackwell GPUs for its Oracle Cloud Infrastructure (OCI) Supercluster to aid large language model (LLM) training and other use cases. The company made the announcement at the CloudWorld 2024 conference. However, given the severe backlog of High-Bandwidth Memory, Oracle wasn’t likely to take delivery of the GPUs before 2026.
Elon Musk’s xAI taps Blackwell GPUs for its supercomputer
Not to be outdone by Oracle, Elon Musk and his team behind the xAI company took delivery and deployed 100,000 H200 Blackwell GPUs in an incredible 19 days. Huang had said that such a deployment should take four years complete. The cluster is in a data center outside of Memphis, Tennessee, and will be used for all things GenAI.
Heat rumors start up again, just in time for earnings
For the second time, reports of heat problems with Nvidia’s newest processors came out right before the release of earnings. In this case, the report stated that due to heat from the Blackwell chips, servers needed to be redesigned to accommodate the heat. This was dismissed by Nvidia on its earnings call – and at the same time, Nvidia reported revenue gains of 17% compared to the prior quarter and 94% compared to the prior year.
AI factory designs
Nvidia released reference designs for what it calls AI factories, or data centers dedicated to AI processing. The company had been talking about AI factories for some time, but this was finally a blueprint telling people how to build one.
Edge computing platform for AI processing
Nvidia launched a new platform called EGX Platform designed to bring real-time artificial intelligence to edge networks. The idea is to put AI computing closer to where sensors collect data before it is sent to larger data centers. This serves as a buffer to whittle down the data sent to data centers so only relevant data is sent down the wire for processing. This can be up to 90% or more of data collected is discarded.
Verizon, Nvidia team up for enterprise AI networking
Nvidia announced a partnership with Verizon to build AI services for enterprises that run workloads over Verizon’s 5G private network. Dubbed 5G Private Network with Enterprise AI, it will run a range of AI applications and workloads over Verizon’s private 5G network with Mobile Edge Compute (MEC), a colocated infrastructure that is a part of Verizon’s public wireless network.
Source:: Network World