BIOS updates here, patches there, and now we have a much more performant 9000 series… but the black eye of premature release may stick with us for a little while yet. Unfortunate thaat it came to this.
The initial Zen 5 reviews were wrought with issues, including our own. This wasn’t the fault of reviewers; it was a lack of communication from AMD on how to properly test its new CPUs. AMD tested on a Windows version with elevated privileges, and in particular, access to better branch prediction. A branch predictor is exactly what it sounds like. It’s a circuit that predicts which of two paths the signal will go inside the CPU. This works with software — i.e. Windows — and the Windows versions available at launch didn’t come with improved branch prediction.
Get your weekly teardown of the tech behind PC gaming
Reviewers retested, and there were performance improvements. But, there were also performance improvements for previous generations of Ryzen CPUs, making the uplift essentially null. Yes, you would get more performance out of a Ryzen 9000 chip, but the dynamic between them and previous-gen Ryzen 7000 CPUs really hadn’t changed.
Jacob Roach / Digital Trends
So, AMD tackled the lineup from two directions. For the Ryzen 9 9950X and Ryzen 9 9900X, AMD focused on improving CCD latency. These two CPUs use 16 and 12 cores, respectively, and they’re split into two Core Complex Dies, or CCDs. A CCD can only hold eight cores with AMD’s current architecture, so these CPUs need two. With two separate dies, there’s some amount of latency as the two CCDs communicate with each other, which can have a performance penalty if you’re activating cores on both dies at the same time. AMD figured it could decrease the latency between the CCDs, thereby improving performance.
The Ryzen 5 9600X and Ryzen 7 9700X are in a different situation. Since they both have eight cores or less, they only use a single CCD. To improve performance on these chips, AMD introduced a 105-watt power limit — both previously had a limit of 65W — that was fully covered under warranty with AEGSA PI 1.2.0.2. You could essentially download more performance just by updating your BIOS, and in a way that AMD was still covering under warranty.
That’s where we are now. If you have a single CCD chip, you have more power at your disposal, and if you have a dual CCD chip, you have improved latency, and all four CPUs have access to improved branch prediction inside the most recent Windows versions. And true to its word, AMD delivered quite the performance improvement.
This is a positive move, and I don’t want to distract from that. Still, it’s important to recognize that AMD marketed and sold a product that wasn’t ready. These aren’t the marginal performance increases we see over the course of a year and a dozen BIOS updates — this is foundational performance that was totally absent from the CPUs AMD sold to customers at launch. It’s great that the extra performance is here now, but we should’ve had it back in August when the CPUs released.
The folks working on these parts at AMD — and at Intel and Nvidia, for that matter — are unironically some of the smartest engineers in the world. They knew how the Ryzen 7 9700X performed at 105W; they knew about branch prediction differences in Windows; and they knew about CCD latency on the two Ryzen 9 chips, and I would bet they knew all of that before the CPUs released. I don’t think AMD intentionally restricted the chips, or that it made any changes to them out of malice. To me, this reads like a company with deadlines that released a product that wasn’t ready.
Thankfully, it looks like Ryzen 9000 is ready now. There’s still plenty of discussion around the value of last-gen AMD components, and the competition from Intel as we stare down the release of Arrow Lake CPUs. But for now, after two months on the market, Ryzen 9000 chips have finally been brought up to par.
