Qualcomm had more hints than releases at Computex but they were pretty big hints. Nothing should be brand new to SemiAccurate readers, just updates and details.
The big bang came at the end of Qualcomm CEO Christiano Amon’s keynote speech. The speech itself was mainly a pat on the back to partners who are waving the Snapdragon PC flag, fair enough. The devices shown off for the first time last year were supplemented by another batch of designs, basically the yearly device refreshes. But what was that big bang we mentioned?
Finally it is ‘officially’ announced
In a shock to no one, Christiano admitted that Qualcomm has datacenter ambitions for real. This isn’t just AI AISCs but the full CPU side of things as well. The datacenter plans at Qualcomm have had a rocky road in the past, more due to external events and finance than product merit. This is the long way of saying that although there was much pomp, ceremony, and even devices shown, nothing ever came out. SemiAccurate saw, used, and actually like the Centriq line but alas that too sunk under the waves.
This time around we think the story will have a different ending, a much happier one. Qualcomm was one of the partners that Nvidia announced for their ‘opening’ of the NVLink protocol, but more on that later. Couple this with the Nuvia derived server cores, basically the top of the three core tiers, then add an AI100 SoC and you have one side of the market covered. NVLink to Nvidia GPUs gets you to the other, and off Qualcomm goes. Details have been promised for ‘soon’ so expect details in the not so distant future but there is a dark shadow looming over this too.
Back to happier themes, and no we don’t mean the Qualcomm 40th anniversary, we mean Docker. As you might guess when the COO of Docker makes an appearance at a Qualcomm keynote, Docker now supports Qualcomm/ARM CPUs natively. This is a really good thing for developers and would only be surpassed if Qualcomm supported a real OS. Don’t hold your breath, the doorstop some claim to be a laptop by my ankle is testament to that. (Authors note: After over a year bumping my ankle, it was moved to behind my chair against the wall to keep the cats from opening the knee wall door so it isn’t entirely useless)
Finally, accessible AI100 hardware
Last up we have the Dell Pro Max Plus laptop/mobile workstation. Why is this at a Qualcomm event? Well if you take the discrete GPU out and put in a Qualcomm AI100 module, you have, err, a laptop with an AI100 card. Other than being an easy entry point for AI100 hardware access, it gives devs a known starting point to start coding. What is known about it? Look at the specs on the card to the side.
Yup, it is indeed the first discrete NPU in a laptop, the Intel offerings were not called NPUs but to be fair, they were really in a different and much lower class of products. 450+ TOPS is nothing to be sneezed at, nor is the 80W it pulls but that is still a solid perf/Watt ratio. The snarky bit is the listed CPU, or lack thereof, for the Pro Max itself. A quick perusal of the Dell site shows there are two models, the 14 and 16, both with Intel CPUs, no AMD to be found there. Or Qualcomm, but this is a good thing. Why? All the tools devs use and are familiar with will work so devs will dev rather than fight compatibility with ARM headaches. If you want to develop for Qualcomm AI hardware, this is your tool.
In the end the Qualcomm Computex messaging was more about the future than the immediate releases. Snapdragon X Elite Gen 2 was promised for a September unveiling and a Q1/26 launch unless someone *HINT* does it a tad earlier. Servers were ‘soon’, and more hints promised a pretty hectic 2H/25 for the company. Even if Qualcomm’s release schedule didn’t coincide with Computex, they did a pretty solid job of keeping our attention.S|A
