- AMD’s head of client CPUs says it is investigating dedicated NPU accelerators
- These would be equivalent with a discreet GPU, but for AI tasks
- Such boards would reduce demand on higher end GPUs as they were no longer purchased for AI work as they are in some cases
AMD looks at a future where it may not just produce free graphics cards for stationary PCs, but similar boards that would be equivalent with an AI accelerator – a discreet NPU, in other words.
CRN reports (via WCCFECH) that Amds Rahul Tikoo, head of his client CPU business, said Team Red is “talking to customers” about “utility cases” and “potential options” for such a dedicated NPU accelerator card.
CRN points out that there are already movements along these lines on the foot, such as an incoming Dell Pro Max Plus -Bearing computer set to boast a few Qualcomm AI 100 PC infernic card. These are two discreet NPU boards with 16 AI kernels and 32 GB of memory, for 32 AI kernels and 64 GB RAM in total.
To put it into perspective, current integrated (on -chip) NPUs, such as those in Intels Lunar Lake CPUs, or AMDS Ryzen AI chips, offer about 50 tops – ideal for copilot+ PCs – while looking at up to 400 tops with the mentioned Qualcomm AI 100.
Tikoo observed: “It’s a very new set of use cases, so we look at that space carefully, but we have solutions if you want to get into that space – we will be able to.”
AMD Exec would not be drawn to give a tip on a timeframe where AMD may be planning to realize such discreet NPU ambitions, but said “it’s not hard to imagine that we can get there pretty quickly” considering the ‘breadth’ of Team Red’s Technologies.
Analysis: Potentially removing the pressure from the high-end GPU demand
So does that mean it won’t be too long before you might look at buying your desktop PC and Muller a discreet NPU with a GPU? Well, not really, this is still not consumer area as such – as mentioned, it’s more about AI power users – but it will have an important impact on everyday PCs, at least for enthusiasts.
These standalone NPU cards are only needed by individuals working on more heavy weight AI tasks with their PC. They will offer the benefits of running large AI models or complex workloads locally rather than on the cloud, with far more responsive performance (evading the delay factor that is inevitably brought into the mixture as you pipe work online into the cloud).
There are obvious benefits of privacy by keeping the work on device instead of switching to clouds, and these discreet NPUs will be designed to be more effective than GPUs that take on this kind of workload — so there will be power savings to have.
And this is where we get to the core of the case for consumers, at least enthusiast PC players who look at buying more expensive graphics cards. As we have seen before, individuals buy with AI, Top -end GPUs – such as RTX 5090 or 5080 – for their rigs. When dedicated NPUs come out of AMD (and others), they will offer a better choice than a higher end GPU – which will remove pressure from the graphics card market.
So especially when a new series of GPUs come out and there is an inevitable haste to buy, there will be less overall demand for models with a higher end -which is good news for delivery and pricing, for players who want a graphics card for, yes, play PC games and not turn down to AI workloads.
Roll on the development of these standalone NPUs, so – it has to be a good thing for players in the end. Another thought of the much further away future is that these NPUs may eventually be needed for AI routines in games when complex AI-run NPCs are brought to. We have already taken some steps down this path, Sky-wisdom, though it is a good thing or not, is a matter of meaning.
