DF Weekly: AMD's Zen 5 - why isn't there a solid consensus in reviews?

In this week's DF Direct, the team discusses a range of topics but it's the subject of the Ryzen 9000/Zen 5 CPU reviews that made the biggest impression on me. Actually, it was a question raised by one of our supporters, asking about discrepancies in results between various outlets. We saw some reasonable, if modest, improvements comparing the new Ryzen 7 9700X against the last-gen 7700X, while others only saw fractional improvements: two to three percent is barely outside the margin of error in CPU-bound scenarios. So why isn't there the same kind of consensus we more typically see in GPU reviews?

All of this relates to how tricky it is to get the measure of CPU performance in gaming - but the simple answer is that in a world where in-built game benchmarks only rarely stress the CPU, results will vary drastically based on the actual material being benched. Even the same game tested with the same components on identical settings can produce very, very different results - and this speaks to the diversity of tasks the processor is asked to tackle.

Going back in time here, I remember benchmarking The Witcher 3 using the Core i3 4130 up against AMD's price-equivalent six-core FX-6300. In cutscenes and the open world, the i3 was faster. However, in the dense city of Novigrad, the FX-6300 ran a lot faster. So, which processor is actually faster? It all depends on what you test, but I'd say that whichever scene is more demanding overall - which area tanks frame-rate hardest - should be the focus of testing - and that's the Novigrad run. However, both sets of data are valid as they both represent processor performance and both benchmarks together may illustrate the strengths and weaknesses of different CPUs. The question is, how deep do you want to go?

DF Direct Weekly #175: PSVR 2 PC Issues, Zen 5 Controversy, Half-Life 3?! Jedi Survivor Last-Gen!

• 0:00:00 Introduction
• 0:00:54 News 01: PS VR2 tested on PC - without an adapter!
• 0:12:49 News 02: Zen 5 launches to mixed reviews
• 0:29:57 News 03: Space Marine 2 previewed
• 0:40:50 News 04: Half-Life 3 confirmed???!!!!
• 0:50:24 News 05: Jedi: Survivor last-gen screenshots revealed
• 1:06:32 News 06: Classic Doom games get revised port
• 1:11:07 News 07: Final Fantasy 16 PC release may be imminent
• 1:18:51 News 08: Ready at Dawn shut down
• 1:25:56 Supporter Q1: Why isn't every new game using DirectStorage?
• 1:30:56 Supporter Q2: How could a "Stop Killing Games" directive work for online-only titles?
• 1:39:13 Supporter Q3: Should we be less judgmental about generative AI use?
• 1:48:18 Supporter Q4: Is the Ryzen 5800X3D one of the greatest ever CPUs?
• 1:55:35 Supporter Q5: Why does the PS4 refuse to die?

As I said, CPU benchmarking is tricky - and that's just one element of processor reviews that can cause confusion. Another controversial topic is how you get the metrics to begin with. It's pretty much established now that to test, say, a GPU, you pair it with the fastest CPU available. So logically, to test a CPU, you hook it up with the fastest GPU around. However, what resolution should you choose? Some say that higher resolutions like 1440p upwards better represent the actual gaming experience. Others believe that the lower the resolution, the less intrusive the GPU is, giving you 'pure' CPU performance.

I'm inclined to side with the latter argument, but in putting together the latest DF benchmarking suite, we provide 1080p, 1440p and 2160p benchmarks - and in many of our new tests, we're finding that with DLSS active, you're CPU limited on all of them, even with the latest and greatest. And with the arrival of upscaling technologies pushing up frame-rates, the notion that you're rarely CPU limited is certainly looking somewhat anachronistic - especially in a world where the likes of Star Wars Jedi: Survivor, Dragon's Dogma 2, Baldur's Gate 3 and Microsoft Flight SImulator can spend much of their play time at CPU limits.

Hopefully this explains why we're seeing different performance deltas between various CPUs from different outlets - and it's a reminder that no matter how much effort any given reviewer puts into their CPUs, the sheer volume of potentially CPU-bound scenarios in any given game could result in many, many different tests. Therefore, in line with just about any hardware review really, it's strongly recommended that any purchasing decisions are derived from looking at however many trusted sources you may have.

Ultimately though, I'm not sure there is that much of a disparity between the reviews I've seen. The key point is that for gaming at least, Ryzen 7 9700X does not beat the existing gaming champion, the Ryzen 7 7800X3D. In fact, looking at Will Judd's review on Eurogamer last week, the numbers reveal that owners of existing AM4-based platforms are still very well serviced by the Ryzen 7 5800X3D, while those on AM5 can go for the top-of-the-line 7800X3D but still get good gaming performance from the lowly Ryzen 5 7600X. Those two chips have something like a 30 percent performance differential between them but Ryzen 9000 processors mostly sit in the middle between those extremes.

As to why the new Zen 5 architecture isn't providing the kind of stratospheric gains users were hoping for, I'd suggest there are many potential explanations. The new chips have been lauded for their power efficiency, which may suggest that AMD isn't really targeting gaming or even general purpose computing. Far more attractive are the datacentre and laptop markets, where power efficiency is hugely important.

Secondly, in the transition from AM4 to AM5, the Ryzen line moved from DDR4 to DDR5 memory. We've seen on Intel processors (which support both) that the same CPU running at the same clock speeds can produce very different levels of performance depending on whether they're running DDR4 or DDR5. AMD has enjoyed that advantage with Ryzen 7000 vs Ryzen 5000, but obviously Ryzen 7000 and 9000 are on the same platform. Indeed, AMD recommends the same memory for optimal performance on both architectures.

With new Intel processors on the horizon, there's clearly an opportunity for the firm here in taking on the Zen 5 line with its upcoming Arrow Lake-S CPUs - supposedly a radical departure from the Raptor Lake chips with all the issues that have presented in recent times. We're certainly looking forward to testing them - and the upcoming 12/16-core Ryzen 9 9950X and 9900X products.

Those reviews should be coming soon and we'll be expanding our range of gaming benchmarks to support Starfield and Counter Strike 2. This is all part of a truly exciting project we're working on at Digital Foundry behind the scenes, automating benchmarking and even automating gameplay on some titles in order to deliver those custom scenes that actually stress-test any given CPU. A good example of this is the Cyberpunk 2077 benchmark above. Run the YouTube video and check it out. And yes, this is an example of the Ryzen 7 9700X beaten by the 7700X... and indeed the 9600X.

In terms of our major benchmarking project, there's a supporter-only video on the DF Supporter Program right now that goes into depth on this - which brings us back to this week's episode of DF Direct Weekly: remember that supporters help shape the show by suggesting topics, commenting on news stories and suggesting questions for our Q+A section. And they get the show days ahead of everyone - join us!