--- title: 'How GPUs Accelerate Data & Analytics with AI' source: 'https://youtube.com/watch?v=zqowMgMgSjA' video_id: 'zqowMgMgSjA' date: 2026-09-15 duration_sec: 322 channel: 'IBM Technology' --- # How GPUs Accelerate Data & Analytics with AI > Source: [How GPUs Accelerate Data & Analytics with AI](https://youtube.com/watch?v=zqowMgMgSjA) ## Summary This video explores why GPU acceleration is becoming essential for modern analytical workloads. It explains how GPUs complement CPUs to improve performance and cost efficiency, and why this shift is driven by the parallel nature of analytics. ### Key Points - **The cost problem of CPU-based analytics** [00:14] — Traditional CPU-based analytics are becoming expensive to scale as teams process billions of records and support AI workloads. - **GPU acceleration as a new approach** [01:13] — GPU acceleration improves performance and infrastructure efficiency without changing how users write SQL or analyze data. - **Heterogeneous computing: CPUs and GPUs together** [01:43] — GPUs don't replace CPUs; they complement them. A host CPU coordinates query execution while GPUs accelerate parallel portions of the workload. - **Why analytical workloads are parallel** [03:27] — Analytical workloads are highly parallel — many SQL operations repeat the same calculations across millions of rows, making them ideal for GPU acceleration. - **Performance improves cost economics** [04:10] — Faster queries reduce compute cost and infrastructure run time, allowing more users to share resources and lowering cost per query. - **The future of analytics compute** [05:06] — The future of analytics will rely on CPUs and GPUs working together to handle growing workloads, not CPUs alone. ## Transcript So why is GPU changing analytical workloads? AI has relied on GPU for years to accelerate complex workloads. Now that same technology is beginning to change analytics. As teams analyze more data and support more AI workloads, traditional CPU-based analytics are becoming more and more expensive to scale. GPU acceleration introduces a new approach, using both CPUs and GPUs to work better together. In this video, we'll explore why the cost and performance of analytical workloads are changing, how GPUs fit into modern analytics, and why this hints at a broader shift in the future of data processing. Modern analytical workloads are fundamentally changing. Teams are processing billions of records, supporting thousands of users, and preparing data for AI applications, all on the same platform. But everyone working with large sets of data experienced that as the size of your data grows, the query execution speed will slow down. Therefore, the industry is beginning to rethink how analytical workloads are executed. GPU acceleration offers a new approach by improving both performance and infrastructure efficiency without changing how users write SQL or analyze data GPU accelerated analytics doesn't replace the CPU and depending on the size and shape of the data CPU might still be faster as you can see right here in this graph. Or using an approach called heterogeneous computing CPUs and GPUs can work together. There would be a host working on CPU, which processes the incoming query and returns the outcome result. Depending on the nature of the workload, it might use CPU or GPU to execute the query. The CPU continues coordinating query execution and managing your overall workflow, while GPU accelerate the highly parallel portions of analytical workloads that benefit from this parallel computation. The key question is why are GPUs suddenly becoming relevant for analytics? The answer comes down to parallelism. Many analytic workloads perform the same operation across enormous amounts of data. And that turns out to be exactly the kind of problem GPUs were designed to solve Let look Let look at three reasons why The first reason is that GPU and CPU have different strengths by design CPUs are incredibly powerful and are relying on a small number of cores. They execute complex instructions, manage applications, and coordinate workloads extremely well. GPUs are different. They contain thousands of smaller processing units that are great at handling things in parallel. Rather than replacing CPUs, GPUs complement them, as you can see in this overview right here. Overall, GPU will give you a better compute density and therefore better performance per unit of compute cost, depending on the nature of your workloads of course. The second reason is that analytical workloads are highly parallel. Many SQL operations repeat the same calculations across millions or billions of rows. You might be filtering, grouping or aggregating. Assume you are selecting data, you're selecting it from a number of tables and you might be having some classes such as where to filter. In between, you might have additional things like ons or joints, and at the end you might be grouping or ordering your data as well Depending on your SQL engine these operations can be executed at the same time instead of one by one And this would make them well suited for GPU acceleration The third reason is better performance changes the economics. The real benefit isn't just making your query execute faster, it's also making sure your cost will stay low. Because the longer your query takes to execute, the higher the cost will be. It's changing compute efficiency and execution characteristics when queries complete more quickly. Because your infrastructure runs for less time, more users can share the same compute resources and your compute cost per query execution decreases. Because your insights are arriving sooner, performance and cost improve together. This is an architectural shift in compute execution models. GPU acceleration isn't about replacing existing analytic platforms, it's about evolving how analytics engines execute work. Workloads will continue to grow, but compute capacity and cost constraints won't grow at the same pace. That's why the future of analytics won't rely on CPUs alone. Instead, we will have CPUs and GPUs working together to allow teams to deliver faster.