Today the HPC world is excited about El Capitan–just as innovative with even tighter coupling between CPU and GPU–based on a single package that includes a unified cache and high-bandwidth memory. My question: will AMD take over the world by commoditizing the technology used in El Capitan? Do Nvidia and Grace Blackwell even have a chance?

Nvidia just got Professor X and Aquaman to hypnotize the world with their spiralling eyeballs (I think)! q^8

Nvidia has shown a pretty canny ability to box competitive GPUs up and marginalize them in a way they can neither earn share nor reasonable margins in PC Graphics. Now the volume opportunity is shifting to data center, a market Nvidia pretty much built single handedly and understands like few others. AMD already is in the lead? Come on TMP.

Not saying AMD won’t get some traction, they can just simply because of all the unsatisfied demand, but I’m not sure how the upstart gets elevated to incumbent-like status? An HBM device is an HBM device. AMD isn’t magically getting more from those 1024 I/O pins than anyone else can.

We all need hope, and we love to root for the underdog. But with Nvidia’s platform lead in Data Center compute — AMD basically has been able to garner low single digits of traction from the last 3 or 4 Instinct generations — I’m going to be from Missouri here with respect to MI300. Show me.

And AMD, for real, grow some balls and submit to MLPerf. Then we’ll know you have a real part. Until then it’s all just posturing.

I think there has been plenty talk about the issues with CoWoS. And I am saying it is a problem with Nvidia being able to hog capacity for HBM and CoWoS. So I am not exactly sure what your problem is. This is not SemiAnalysis — this is The Next Platform, and we are one level higher up in the stack.

As far as I know, no one else knows what the X stands for. I merely said — with humor — to re-use Xavier because it was fun. Lighten up.

What are you talking about the Jetson line is already in its latest iteration and that Jensen/Xavier branding is already utilized for some time now. Can the “technology” press be a little more accurate and with some regards to their due diligence. And Reporters used to have notebooks with such information and kept that with them there with information about product code names/product branding and especially Nvidia’s Naming for it’s GPU Micro-Architectures where there’s good documentation there for that.

I do not like the trends here with the Independent Tech Press becoming too Mind Share for hire oriented rather than Just the Facts Oriented. And really there the tech press outside of the paywall has slid into that abyss where accuracy is just getting in that way of the Marketing!

Why is the Tech Press not focused on TSMC’s part of the AI product shortages with that TSMC CoWoS packaging capacity shortage there and how that affects competition in an AI Accelerator Market that the French, and probably other, regulators are looking at! And Why is one business entity able to buy up most of the packaging/fabrication capacity there to the possible detriment of competition. And the US is definitely targeting Chip Packaging IP as part of the chips act funding but TSMC needs to be in front of a US Congressional Committee explaining just why that CoWoS capacity was not properly planned out in advance for the demand that the AI boom has placed on that limited capacity from only one IP supplier. Intel’s CEO sure made a good move there with regards to Intel maintaining it’s own In-House Chip/Chiplet Packaging IP and capacity but will Intel be willing to let others get at that packaging capacity as a service? And maybe the move towards some standardization for chiplet packaging IP needs to forced upon the industry in the name of Public Utility as the world economy is so dependent on Processors and how those are Packaged.

Nvidia should open it up for a vote I think. Some folks could vote for Marvell’s Professor X, but I’d go for Professor Popkiss, after the prescient inventor of the uniquely multi-environment, airbreathing, retrorocketed, VTOL, Clear-Vu, almost self-driving, vehicle: Supercar! Jetson is just a bit too cartoony … and C. Scarlet’s self-healing retrometabolism feels a bit too “out-there”! 8^b

You’re good. We are all part of one big run of Newton’s Method of Approximation on reality. Together, we get the right curves plotted.

Think you’re reaching pretty hard here…