Six years ago, AMD made a borderline outrageous claim - that they'll increase the performance/watt of their APUs by 25 times by 2020. Hence, the 25x20 meme was born.
Let's look at the context here. In 2014, AMD was coming off the launch of the spectacular failure, Bulldozer. The highest end Bulldozer was so bad, it couldn't keep up with Intel's mid-range Core i5 parts, let alone Core i7 and HEDT. Worse still, the power efficiency was abysmal, needing nearly twice the power of Intel's competitors to offer the same performance. Meanwhile, Intel was executing with perfection, and Haswell was another huge leap forward.
You can see why AMD's claims were widely ridiculed and mocked back then. Followers of my blog would know I've been covering AMD on Steem, and now Hive, since 2016. I've been wildly optimistic about this company against all odds, and have seen its stock price increase by an order of magnitude during the time. But even I couldn't imagine how they could possibly increase efficiency by 25x in 6 short years! The conventional wisdom was that Moore's law was slowing down, but even with a rather optimistic outlook showed a 2x improvement in efficiency every 2 years. That would get is to 8x. Coincidentally, that's exactly what Intel has offered. The latest Ice Lake parts offer roughly the 8x efficiency over Haswell.
So how did it get there? Of course, AMD was coming from a long way down, and it's obviously much easier to improve when you have a shitty baseline. AMD proved this in 2015 with the launch of Carrizo. It kept the same base architecture as Bulldozer, but that architecture was so bad and unrefined that it had plenty of low-hanging fruit. So, Escavator improved things significantly, with a 3.5x improvement in power efficiency. Suddenly, the 25x20 goal seemed possible, but Zen was a long way away.
As expected, the next step came with Zen 1, with Raven Ridge. This was a big leap forward, but only brought the overall tally to 10x. Over the next couple of years, we got minor iterations. The clock was ticking, time was running out.
Then, of course, came Zen 2. Now, I've written about Zen 2 extensively, and even called it the best CPU architecture in decades. You can see why. Going from AMD's 2019 APUs to the 2020 Zen 2 APUs brought an astonishing 3x improvement in performance/watt.
The end result, AMD has achieved nearly 32x performance/watt improvement in 6 short years. I wouldn't have believed you if you had said this would happen, and I'm one of the few who remained optimistic about AMD when everyone was predicting bankruptcy. As a result, AMD has outperformed Intel by 4 times. Let that sink in... They have gone from offering half the efficiency of Intel in 2014 to doubling Intel in 2020. Unbelievable stuff.
I should point out that AMD's scores also account for GPU improvements, but it's clearly evident that a bulk of the unexpected gains have come from CPU evolution.
Looking forward, Intel's 10nm is finally coming online with Tiger Lake. I do expect them to close the gap, but then, AMD is going to pull ahead once again with Zen 3. Intel has to nail 7nm and make a big leap forward to catch up to where AMD will be with Zen 4. Either way, it's an exciting time for the CPU industry - a far cry from the dull years around 2014.
Read more about it on AMD's blog: https://www.amd.com/en/technologies/25x20. AnandTech has a detailed article on the same, which I have referenced: https://www.anandtech.com/show/15881/amd-succeeds-in-its-25x20-goal-renoir-zen2-vega-crosses-the-line-in-2020