Industry analysis explores novel CPU-GPU hybrid system architectures for improved performance.
The high performance computing market operates in boom-bust cycles across diverse applications including technical simulation, cryptocurrency mining, gaming, video rendering, visualization, machine learning, and data analytics, making projections challenging for suppliers. However, Nvidia has positioned itself to weather these fluctuations by diversifying across multiple revenue streams that tend to balance out over time. While the GPU channel for gamers has delivered strong performance and pricing, both Nvidia and AMD pushed inventory into the channel in recent quarters and must wait for sellers to burn through the excess. Both companies have also been impacted by a downturn in specialized GPUs created for cryptocurrency mining. As a result, GPU business growth for these companies has not met expectations as 2018 got underway.
Despite these headwinds, Nvidia continues to grow faster than its compute peers, with profits expanding more rapidly than revenues—a positive indicator for any company. In the third fiscal quarter of fiscal 2019 ended in October, Nvidia reported $3.18 billion in sales, up 20.7 percent year-over-year, with net income growing 46.8 percent to reach $1.23 billion. This represents a remarkable achievement when viewed against the company's history; five years prior, breaking through $1 billion in quarterly sales was considered a major milestone. The company now regularly delivers that figure in profitability from its GPU products and services, reflecting what the company has characterized as "fractal diversification" across successive generations of GPUs.
Nvidia's R&D investments underscore its commitment to maintaining technological leadership. In the current quarter, the company spent $605 million on research and development, totaling $1.73 billion across the nine months of fiscal year 2019. For comparison, Hewlett Packard Enterprise spent only $1.22 billion in the first three quarters of its fiscal 2018, while Dell spent roughly $1.1 billion per quarter over the same period—though it is worth noting that Dell generates approximately $80 billion in annual revenue compared to Nvidia's roughly $12.8 billion, making Dell 6X larger, while HPE at approximately $32 billion is 2.5X larger than Nvidia. In the fiscal third quarter, Nvidia's R&D spending grew 30 percent, outpacing revenue growth and suggesting new GPU architectures in development. This increased investment was partially offset by a $149 million tax benefit, contributing to the higher-than-expected net income figure, though even without this benefit Nvidia would have crossed the $1 billion profit threshold.
Gaming GPU sales rose 13 percent to $1.76 billion in the thirteen weeks ended in October, but this growth was tempered by inventory buildup in the channel and a sharper-than-expected collapse in demand for cryptocurrency mining GPUs. The falloff in sales for Ethereum and other cryptocurrency mining applications hit harder than anticipated, forcing Nvidia to take a $57 million charge against mining-focused components in the quarter. The company's product evolution reflects this market dynamic: while GeForce gaming GPUs historically served as the foundation for higher-end Quadro visualization and workstation cards used in scientific and defense applications, Nvidia's highest-end offerings are now targeted at datacenter compute for HPC centers, cloud builders, and hyperscalers running simulation, modeling, and machine learning workloads. Technology developed at the top trickles downward, as exemplified by the "Turing" Tesla T4 accelerators, which derive from the "Volta" compute architecture with Tensor Core units retuned for dynamic ray tracing and machine learning inference.
Nvidia's datacenter business, encompassing DGX-1 and DGX-2 systems featuring the integrated NVSwitch interconnect that links sixteen Volta GPU accelerators into a shared memory configuration, grew 58.1 percent year-over-year to $792 million in the quarter. While growth has moderated—two years ago the datacenter segment was tripling, and last year it was doubling—a $4 billion annualized run rate with estimated gross margins of 80 percent or higher represents a substantial business. GPU acceleration for HPC and AI workloads, along with emerging database applications, remains in early stages of adoption, suggesting the current growth moderation may represent a local flattening before acceleration resumes.