AMD's CPU division is booming as CEO Dr. Lisa Su says sales 'far exceeded my expectations'
AMD CEO Dr. Lisa Su says the company's CPU sales have "far exceeded my expectations in terms of demand." That's the good news. The more ambivalent addendum, at least for we mere PC gamers, is that you can probably guess why AMD CPU sales are going gangbusters. Rejoice, because it is of course AI.
"We’re seeing a significant CPU demand, frankly, as a result of the inference demand picking up," Su told investors at the recent Morgan Stanley Conference. Inference, of course, involves the running of AI models and delivering AI services, as opposed to training or building AI models in the first place.
Inference has slightly different requirements from a number of angles. The compute load is of a different character—and broadly more CPU-intensive than training—and so are the software requirements and platforms. Nvidia's GPUs are particularly dominant in training and not just in terms of the hardware and raw performance. Arguably, Nvidia's CUDA software framework is just as (if not more) important in explaining why it is so dominant in the AI training market.
Indeed, AMD is increasingly tended to pitch its Instinct AI GPUs at inference rather than training. As a for instance, the new mega-deal between AMD and Meta involving semi-custom hardware is thought to be very much an inference-optimised design.
Anywho, it is likely in reference to inference workloads that Su also explained that AMD's customers now think, "the demand for CPU compute sitting along AI was perhaps something that was under-forecasted."
As for what all this means for ye olde consumer PCs, well, there are mixed signals. On the one hand, Su spoke of supply constraints due to massively increased demand for CPUs, which isn't good. On the other, she also "we are very, very well positioned from a supply standpoint to meet a large percentage of that demand."
What's more, AMD's latest server Zen 6-based CPU, codenamed Venice, is being built on TSMC's most advanced N2 node. That's not a rumour, that's according to AMD. What we don't know is what node AMD will use for its Zen 6 consumer CPUs.
AMD desktop and laptop CPU chiplets are currently built on TSMC N4 silicon. So, using N3 rather than the latest N2 would still provide a full node jump but also mean that consumer CPUs aren't competing with those new Venice server CPU chiplets for the most advanced TSMC N2 silicon.
The bottom line, as ever, is that we just don't know for now. All this AI demand certainly helps when it comes to giving AMD lots of resources to spend on engineering new chips, including chips for gaming PCs. But it also makes everything more expensive.