For all of the hype leading up to the actual launch today, a lot of people, I think, expected some degree of disappointment, however mild.
AMD’s Zen 2 launch last summer was largely good, but it was accompanied with issues of boost clock attainment and retention, and while AMD finally won the raw IPC comparison with Intel, it was by such a small margin that their clock speed disadvantage meant the case since Ryzen’s launch remained true – Intel remained on top in gaming, although the difference was small enough that getting Ryzen made more sense then it had even previously.
Today, the case inverts.
If you want the high-level takeaway of the review coverage the Ryzen 5000 series that launched today, it is very straightforward – the de facto best gaming CPU you can buy today at this very moment (well, maybe, depending on availability) is any flavor of Ryzen 5000, literally any one of them. Even the six-core model quite often beats the more expensive Intel Core i9 10900k, and by a high enough number that it mostly escapes margin of error territory and reaches the point of comedy. Now, it isn’t always the case, and there are some titles where Intel still stays competitive and even close to or slightly above the top of the AMD lineup, but those wins are tiny and few compared to the overall victories of the AMD lineup.
Now, it is worth breaking down a little bit more than that, but only a little bit more. In general, if you’re going to buy a system today for gaming, streaming, productivity, or even a mix of tasks, generally, the AMD lineup with Ryzen 5000 is unilaterally better. There are several factors for why.
Improved IPC: Independent benchmarks now confirm that AMD’s claim of 19% IPC uplift is accurate and perhaps even lowballing depending on the workload, as the 19% number itself was an average across a multitude of benchmarks from AMD. In some cases, the performance improvement at same clocks is 20% or more, and in others, it is slightly less. However, this increase is huge and represents a significant improvement even over Zen 2, which itself was only (only) a 15% uplift over Zen 1.
Improved Clock Speeds: While clocks haven’t increased that much over the original Zen 2 launch last year or even the XT refresh parts this last summer, the speeds are generally up at each tier by 100-200 MHz. The 16 core 3950x had a maximum boost clock of 4.7 GHz, now the 5950x has 4.9 GHz listed maximum boost. This is a pretty tame increase in-line with the fact that AMD is not using a new silicon manufacturing process but rather has made small tweaks and refinement to the chip floor plan to wring out some extra edge. Speaking of…
Boost Clock Peaks Are A Pleasant, Unlabeled Surprise: The Ryzen 9 5950x lists a maximum core speed of 4.9 GHz. Many of us, myself included, expected this to be Zen 2 style boost, where 4.9 GHz was a theoretical maximum that would only ever be obtained in fantasy and that most would see it fall short of that. Instead, AMD has defied expectations and didn’t even advertise it in advance. Most reviewers with 5950x CPUs have shown a peak speed exceeding 5 GHz, with some choice silicon getting very close to boosting to 5.1 GHz, on normal stock settings and behaviors! This is an impressive result, and the polar opposite of the Zen 2 experience, where reaching the listed maximum clock was a fairy tale and it caused no end of irritation among buyers. This time around, it seems like the printed max boost specification is a safe target that every CPU can obtain out of the box, but many samples will have more headroom.
Sustained Boost Seems to Be Better: The few reviewers I’ve seen who offered clock speed over time plots have shown that while Zen’s underlying ping-pong clock speeds remain intact as a power management strategy, when a sustained load is placed onto the CPU, it tends to settle into a groove faster and stay at the higher clock speeds for longer. The difference is often seconds, but those extra seconds can be enough for a system to complete work and throttle down to save power and heat output.
No Day 1 Jank: All the reviews I’ve seen so far today have been incident-free, which is, unfortunately, not a thing that Ryzen has had previously. Memory speeds seem to work correctly, motherboards with updated BIOSes pick up immediately and run with the chips, boost behavior seems to be very consistent across all the different test configs I’ve seen so far, and no reviewer has mentioned any real issues with getting the platforms working as expected or observed any unusually odd outcomes in testing. All of that is good news!
Given all of that, it seems like the Ryzen lineup is doing really well, which is great. However, we have to discuss pricing and availability.
Pricing-wise, the $50 across-the-board price hike remains contentious, but less so with validated performance numbers to lean on. The 6-core pricing remains egregious, in my opinion, but seeing it put the i9 10900k on ice in CS:GO of all games is just the funniest thing. At that point, you know what? For pure gaming, it almost always beats the 10900k at $200 less, so suddenly that value proposition makes way more sense! The rest of the stack is basically competing with the prior gen Ryzen 3000 part equivalents, and the 5000 series always wins those comparisons by 20% or more. These gains still offset the price increase, retaining a high performance per dollar rating.
However, all of that assumes, much like the RTX 3000 series, that you can buy these parts. Currently, 3.5 hours after launch, every retailer I checked was completely sold out. I will say that my own RTX 3000 comparison was unfair in that the 5600x and 5800x retained stock for longer, but the 5900x and 5950x will be hot ticket items for some time. There are reasons to believe that this may not be a continuing trend – AMD is using the same IO die as the 3000 series, which they’ve been manufacturing for nearly 2 years now, so supply remains steady. Further, the IO die is made on Global Foundries 14nm instead of using their precious TSMC 7 nm supply of silicon wafers, further improving supply.
On the flip side, however, AMD is using TSMC 7nm for a dizzyingly large product array at this point. Zen 2 CPU chiplets (remain in supply for the low end of the market and the datacenter Epyc CPUs until the Zen 3 Epyc models launch), Renoir mobile and desktop Ryzen 4000 APUs, all 3 next-gen consoles, Radeon RX 5000 series, the upcoming Radeon RX 6000 series (with much larger dies), and to top it all off, Zen 3 CPU chiplets. Now, the chiplet design for their CPUs does mean that a single wafer of Zen 3 chiplets produces a large quantity of CPUs (roughly 650 chiplets, which translates to somewhere between 325-650 CPUs for desktop Ryzen) and AMD is now TSMCs top customer for 7nm volume, of which TSMC produces over 100,000 wafers per month. Given that volume, if we cut it in half, AMD could have as many as 50,000 wafers per month, and the production volume depends on how they shift the products around to meet demand.
That is really all we can know for now, because of course, day 1 sellout was likely to happen here anyways, as it did with Zen 2. The question remains how much they can restock. If AMD has stockpiled Zen 3 chiplets for this launch, they could be in a good situation (one month of 25,000 wafers worth of CPU chiplets would produce enough working silicon for 8 million 5900 or 5950x CPUs, and we know the design of Zen 3 taped out several months ago, meaning production has been underway for some time), but there is a tug of war over managing their supply. Microsoft needs a given number of SoCs for both Xbox Series consoles, Sony needs a good supply for PS5s, RDNA 2 launch means a lot of huge dies (the same wafer that produces up to 650ish Zen 3 CPUs can only, with rough napkin math, produce around 86 GPUs of the Radeon RX 6800/XT/6900 XT size), and maintaining supply for datacenter and mobile CPUs will remain a priority. Now, the thing with AMD is that we’ve had rumors for months about tape outs and production of parts, and I would think it very likely that they’ve produced throughout the full year of 2020 with this late year launch series in mind. We know the consoles have been finalized and likely in full production for most of the year, and AMD is only responsible for the base silicon delivery of the SoC chips in those. RDNA 2 production silicon was being used to show off a raytracing demo this last spring, which tells me that those parts have likely been in production for a handful of months now too. Zen 3 rumors were really hitting their stride over the summer, and a few months to produce their most economical product in terms of silicon area tells me that shortages are potentially short-lived.
Lastly, we know from public information that AMD nearly doubled their 7nm capacity with TSMC effective in the second half of 2020, which means they’ve had 4 months now where their capacity has been dramatically increased. If they’ve spent the calendar year working on meeting their contracts with Sony and Microsoft, they could be in a great position to mass-produce CPUs (and GPUs) now. If anything, my suspicion is that we’ll see reasonable supply restocks weekly (Zen 2 high-end SKUs had the same) and I think that the Radeon RX 6000 cards are drastically more likely to be supply constrained, due to how much larger the die is and how few of them fit on a wafer relative to the Zen chiplets. At least that is what I am telling myself since I was unable to get the 5950X of my dreams with today’s launch and am hoping it restocks at a convenient time for myself, along with me being able to secure an RX 6900 XT before any shortages there!
But that is all speculation on my part. For now, on the merits of price and performance, AMD has truly pulled off something great. Long-term, it has implications for the low-cost market (an APU with Zen 3 and Navi could be a killer low-cost gaming solution) and for Intel (latest rumors are that the top Rocket Lake part may still be able to pull out a modest win in gaming over the 5600X, but the rest of the stack remains to be seen), but for today, the best gaming CPU you can buy (theoretically, at least) is Zen 3, and given that this wasn’t even fully expected until AMD’s event last month, well, what a great surprise.