“Micron is delivering a trifecta with this HBM3E milestone: time-to-market management, best-in-class {industry} efficiency, and a differentiated energy effectivity profile,” mentioned Sumit Sadana, government vice chairman and chief enterprise officer at Micron Technology. “AI workloads are closely reliant on reminiscence bandwidth and capability, and Micron may be very well-positioned to assist the numerous AI progress forward via our industry-leading HBM3E and HBM4 roadmap, in addition to our full portfolio of DRAM and NAND options for AI functions.”
The efficiency that Sadana alludes to is blistering pin pace of 9.2 gigabits per second (GB/s), leading to reminiscence bandwidth of greater than 1.2 terabytes per second (TB/s). It’s additionally energy environment friendly—Micron claims its HBM3E reminiscence consumes 30% much less energy than its rivals.
This does not imply an entire lot for customers, not less than in a roundabout way, AMD flirted with HBM on its gaming GPUs (bear in mind Vega?), however the price to efficiency ratio simply is not aggressive with GDDR. While uncooked bandwidth is not unimportant, amount takes precedence within the shopper realm. We additionally do not need to think about how way more gaming graphics card would price of AMD, Intel, and NVIDIA determined to make a wholesale swap to HBM as an alternative of GDDR.