I am not talking about the FirePro S9000, which was — how shall I say it? — not very good and was not a real Instinct card. The first CUDA-enabled GPUs were also not really datacenter-class products, either. I see your point, but I suspect you can see mine.

This is wrong. Vega was the last GCN architecture, not the first. The first was Tahiti, which released at the end of 2011. The top-end consumer GPU using it was the 7970. The server GPU based on Tahiti, released in mid-2012, was the FirePro S9000.

After the Vega-based Instinct cards, AMD evolved GCN into CDNA, which is the architecture used in all the Instinct cards from the MI100 onwards.

Can’t wait to hear that buccaneer’s cannon fire!

I’ll add that, in this, both Motley Fool and Tom’s Hardware get points off for publishing an inaccurate automated transcription of AMD’s 10/31/23 earnings call that so mis-stated the timeline of MI300A’s shipments to El Capitan! Humans (hopefully knowledgeable ones) still got to double-check those error-prone automated gizmos (or maybe just stick to reading the more accurate and trustworthy TNP).

I’m sure AMD is keeping their powder dry to make a big splash at SC23.

Right. I keep thinking of the old Semi-Custom and Embedded division pre-Xilinx.

As for GAAP versus non-GAAP. I think they are equally disingenuous. You can’t call an underlying business more profitable if the overlying business is a huge cost, or has to take a writeoff, or whatever shenanigans (or benefits) happen from time to time. The way I see it, only real net income counts. Because that is the money you can spend,

It would probably also be better to use free cash flow or non-GAAP profits in the first chart, which for AMD better highlights their true profitability. Using GAAP numbers is skewed by funny accounting gimmicks (see that big spike in Q4 2020, AMD claimed old tax losses. And the near zero profits recently is AMD writing off fake losses and intangibles when they bought Xilinx with shares, not cash).

Agreed. But it is a 20 percent tweak in features on Arcturus to get to Aldebaran, not a complete from scratch redo.

No. $50 million incremental to the $400 million in Q1 2024. Still badly written by me, and my model says $400 Q1, $450 Q2, $550 Q3, $600 Q4.

But overall, I love your work here, it’s really interesting and a good dive into what’s happening in the base layers of systems and hardware and how that will affect IT pros. And glimpses of where the world might be going. Personally I think the AI stuff is way over hyped, like Bitcoin and all that stuff.

P.s. Akismet seems to think this comment is a duplicate of my original one. Sigh….