Why it issues: AMD is within the distinctive place of being the first competitor to Nvidia on GPUs – for each PCs and AI acceleration – and to Intel for CPUs – each for servers and PCs – so their efforts are getting extra consideration than they ever have. And with that, the official opening keynote for this yr’s Computex was delivered by AMD CEO Dr. Lisa Su.
Given AMD’s huge product portfolio, it is not shocking that Dr. Su’s keynote coated a broad vary of matters – even a couple of unrelated to GenAI (!). The 2 key bulletins had been the brand new Zen 5 CPU structure (coated in full element right here) and a brand new NPU structure referred to as XDNA2.
What’s notably attention-grabbing about XDNA2 is its help for the Block Floating Level (Block FP16) knowledge sort, which provides the velocity of 8-bit integer efficiency and the accuracy of 16-bit floating level. In line with AMD, it is a new trade customary – which means current fashions can leverage it – and AMD’s implementation is the primary to be executed in {hardware}.
The Computex present has an extended historical past of being a essential launch level for conventional PC elements, and AMD began issues off with their next-generation desktop PC elements – the Ryzen 9000 collection – which do not have a built-in NPU. What they do have, nonetheless, is the sort of efficiency that players, content material creators, and different PC system builders are always searching for for conventional PC purposes. Let’s not neglect how necessary that also is, even within the period of “AI PCs.”
After all, AMD additionally had new choices within the AI PC house – a brand new collection of mobile-focused elements for the brand new Copilot+ PCs that Microsoft and different companions introduced a couple of weeks in the past.
AMD is branding them because the Ryzen AI 300 collection to mirror the truth that they’re the third era of laptop computer chips constructed by AMD with an built-in NPU. A little bit-known reality is that AMD’s Ryzen 7040 – introduced in January 2023 – was the primary chip with built-in AI acceleration, adopted by the Ryzen 8040 on the finish of final yr.
We knew this new chip was coming, however what was considerably sudden is how a lot new know-how AMD has built-in into the AI 300 (codenamed Strix Level). It options new Zen 5 CPU cores, an upgraded RDNA 3.5 GPU structure, and a brand new NPU constructed on XDNA2 that gives a formidable 50 TOPs of efficiency.
What’s additionally shocking is how shortly laptops with Ryzen AI 300s are coming to market.
What’s additionally shocking is how shortly laptops with Ryzen AI 300s are coming to market. Programs are anticipated as quickly as subsequent month, simply weeks after the primary Qualcomm Snapdragon X Copilot+ PCs are set to ship.
One huge problem, nonetheless, is that the x86 CPU and AMD-specific NPU variations of the Copilot+ software program will not be prepared when these AMD-powered PCs are launched. Apparently, Microsoft did not anticipate x86 distributors like AMD and Intel to be executed so quickly and prioritized their work for the Arm Qualcomm units. Because of this, these can be Copilot+ “Prepared” techniques, which means they’re going to want a software program improve – doubtless in early fall – to turn out to be full-blown next-gen AI PCs.
Nonetheless, this vastly sped-up timeframe – which Intel can be anticipated to announce for his or her new chips this week – has been spectacular to observe. Early within the growth of AI PCs, the frequent thought was that Qualcomm would have a 12 to 18-month lead over each AMD and Intel in creating an element that met Microsoft’s 40+ NPU TOPS efficiency specs.
The robust aggressive risk from Qualcomm, nonetheless, impressed the 2 PC semiconductor stalwarts to maneuver their schedules ahead, and it appears to be like like they’ve succeeded.
On the datacenter aspect, AMD previewed their newest Fifth-gen Epyc CPUs (codename Turin) and their Intuition MI-300 collection GPU accelerators. As with the PC chips, AMD’s newest server CPUs are constructed round Zen 5 with aggressive efficiency enhancements for sure AI workloads which can be as much as 5x quicker than Intel equivalents, in keeping with AMD.
… as Nvidia did final evening, AMD additionally unveiled an annual cadence for enhancements for his or her GPU accelerator line and provided particulars as much as 2026.
For GPU accelerators, AMD introduced the Intuition MI325, which provides twice the HBM3E reminiscence of any card available on the market. Extra importantly, as Nvidia did final evening, AMD additionally unveiled an annual cadence for enhancements for his or her GPU accelerator line and provided particulars as much as 2026.
Subsequent yr’s MI350, which can be primarily based on a brand new CDNA4 GPU compute structure, will leverage each the elevated reminiscence capability and this new structure to ship a formidable 35x enchancment versus present playing cards. For perspective, AMD believes it is going to give them a efficiency lead over Nvidia’s newest era merchandise.
AMD is likely one of the few corporations that is been capable of achieve any traction towards AMD for large-scale acceleration, so any enhancements to this line of merchandise is certain to be well-received by anybody on the lookout for an Nvidia different – each massive cloud computing suppliers and enterprise knowledge facilities.
Taken as an entire, the AMD story continues to advance and impress. It is superb to see how far the corporate has come within the final 10 years, and it is clear they proceed to be a driving drive within the computing and semiconductor world.
Bob O’Donnell is the president and chief analyst of TECHnalysis Analysis, LLC, a market analysis agency that gives strategic consulting and market analysis providers to the know-how trade {and professional} monetary group. You may comply with Bob on Twitter @bobodtech
