When rumors surfaced several weeks ago claiming that AMD would launch Carrizo late this year, we attempted to deflate them with minimal success. Now, AMD’s John Byrne has taken to YouTube to demonstrate the new Carrizo APU and to confirm that both Carrizo and Carrizo-L won’t ship until the first half of 2015 — in line with original expectations. Managing product narratives is particularly important for a company in the midst of the kind of transitions that AMD is experiencing, and the internet’s self-referential echo chamber can become a nightmare.

More important than the shipping date is what Byrne had to say about the final product. According to him, Carrizo is the greatest single leap forward in performance per watt that AMD has ever launched, with an all-new graphics core and the aforementioned Excavator APU.

Previously, it was expected that Carrizo would deliver a very modest update to AMD’s product lineup. It’s still built on 28nm, Excavator is a point update to the Bulldozer family (and the final BD-derived “big core” CPU), and AMD may have opted to use high density libraries (HDLs) to squeeze better performance and power consumption out of the core — at the expense of scaling. Do these new announcements invalidate these expectations? I’d argue no.

Putting rumors in context

When mobile Kaveri shipped, we noted that low power envelopes were actually the best comparison point for AMD and Intel processors. The lower the TDP, the better AMD has competed, particularly in graphics.

Whether AMD adopted HDLs for the entire Carrizo APU or not, further tweaks, improvements, and better power management could still yield significant improvements to the overall design. Raw CPU performance isn’t expected to edge up much with Excavator, save in specific scenarios. We know that 256-bit AVX registers are coming in the new chip, but AMD has been mum on other improvements. Given the push behind the Steamroller core (which ended up delivering just 7-10% higher IPC in most workloads), it’s not reasonable to expect a dramatic reversal at this late date.

As for the GPU side, I still suspect we’ll see a Tonga-derived GPU core in AMD’s next-generation of APUs, though that could happen on either Carrizo, the 20nm die shrink of Puma, or both. Again, we know that AMD is working on cutting-edge memory technologies, like High Bandwidth Memory (HBM), but it’s not going to deploy radical new memory innovations on budget chips intended for lower-end systems first. Without major new developments in RAM scaling, AMD’s APUs will remain fairly memory bandwidth challenged — though they still outperform Intel’s integrated graphics.

We’d be pleased to see AMD deliver a 10-15% increase in performance-per-watt over Kaveri in both CPU and GPU performance. If the company could manage a 30% gain, it would substantially close the gap with Haswell, and could even win breathing room for thin-and-light tablets.

Ultimately, however, everything AMD is doing in its CPU and GPU business is calibrated towards its product launches in 2016. That’s when we’ll see its first custom ARM architecture, new CPU architectures, and hopefully, the nucleus of a new set of products that can move the company forward. In 2015, it’ll be Project Skybridge, AMD’s first ARM-based processors shipping in volume, and possibly a new GCN-derived GPU architecture carrying the banner, buoyed by Xbox One and PS4 sales.