Imagine you have 1000 trucks to move your goods, but only a 2 lane road to your company. If your goods aren’t moving fast enough will you buy more trucks or invest in a wider road? As obvious as this example seems, when it’s about IT, people buy the trucks.
When you process big data, run AI/ML workloads, reporting or databases/data warehouses, you want to keep their data in memory as much as possible, that’s faster. But once it’s in memory, you also need to get it to the CPU fast, or the CPU will spend most of it’s time being idle, waiting for the memory contents to arrive. That’s where memory bandwidth comes in. Memory bandwidth is about how fast you can move or process your data, not how much data you can store.
Memory bandwidth is almost always overlooked when sizing systems or analyzing performance issues. Most performance tooling has no clue about memory bandwidth usage, and will not alert you when this bandwidth becomes scarce. They will tell you to buy more trucks, not widen the road.
Did you know that AMD Epyc CPU’s have 8 memory channels where Intel scalable CPU’s only have 6? Choosing an AMD platform will give you 33% more memory bandwidth at the same memory speed. Choosing an AMD platform in the cloud will give you extra speed, often at no extra cost.
Get help with your data intensive workloads. Book a free 30 minute call with us and we’ll help you widen your road.