2019 was a hell of an exciting year in PC gaming tech.
The year saw a lot of new technology, exciting horse races between the dominant players in CPUs and GPUs, and a lot of foreshadowing of the battles to come in the next decade.
While working on a more gaming-centric post, I thought it would be fun to look back at the technological side of things and talk about the tech powering our games.
2017 kicked off a war which hasn’t been seen in years, with AMD coming back from the sustained disaster of the “construction” cores that made up the FX lineup and launching Ryzen, the first generation Zen architecture. 2017 and 2018 saw AMD building and iterating on a solid base – doubling mainstream CPU core counts and positioning themselves as a solid budget alternative and no longer the choice of paupers.
2019, however, saw them really capitalize. Intel, to put it mildly, has been complacent in mainstream desktop hardware, iterating the same core design since late 2015 and having a lineup that is still almost entirely built on this relatively old tech today! AMD, however, came out with their Zen 2 architecture in 2019, and that changed the game. Intel left their lunch unattended and AMD snuck up on them, fired a warning shot, and then ate Intel’s lunch. Zen 2 has a lot of really cool features – for the layperson, the increased clock speeds, better power efficiency, and higher instructions per clock make a huge difference and take the experience in games and apps far above the original Zen and Zen+ designs. For the techies, there is a lot to love about Zen 2 – the chiplet design that separates the IO and memory controller into their own silicon separate from the CPU cores while using different manufacturing processes allowed AMD to take lower successful yields from TSMC’s new 7nm process and capitalize by building a complete product stack around a single core product that is easy to manufacture, small, and therefore yields terrifically while allowing for advanced binning – power efficient parts go to the Epyc server CPUs, high-clock speed chiplets go to desktop mainstream parts, and the things in-between have multiple places up and down the stack they can be plugged into.
It wasn’t all sunshine and roses, which I can say as someone who bought the 12-core Ryzen 9 3900x at launch. BIOS revisions were slow to come out and performance failed to hit the mark of advertised boost speeds up front. The chiplet design, while it maintains the thermal envelope of the prior-generation hardware, can be somewhat trickier to cool since the thermal density of the chiplet design is much higher, and the CPU cores are off-center inside the package. Motherboard compatability was iffy for prior-gen boards due to necessary BIOS updates, meaning that the best value to consumers (a third gen CPU like the 6-core 3600 or 8 core 3700x with a second gen motherboard with the X470 or B450 chipsets) was difficult to pull off without an existing second-gen CPU, a friend with one, a BIOS-flashback equipped board, or sending off to AMD for a bootkit and waiting for that to come in the mail.
However, as of the end of 2019, AMD has mostly stuck the landing – third gen Ryzen is stable, powerful, and has products from 4 cores (with Zen+ only currently, sadly) up to 16 cores in the mainstream desktop lineup, and a new iteration of the popular high-end Threadripper CPUs offering 24 and 32 core variants, with a 64 core monster coming next year. Newer second-gen motherboards are shipping with working third-gen ready BIOSes out of the box, and further BIOS revisions have improved performance and ensured more frequent appearances of the maximum rated boost clock.
Intel hasn’t been dead this whole time, and still has a high market penetration due to years of AMD ineptitude and Intel’s dominance over OEM partner deals, but has felt the sting. Their major releases this year on desktop consisted of a refresh of their top gaming CPU, the 9900k, now with a 9900ks variant offering a 5 GHz all-core speed that makes it a better pure gaming CPU. Intel also launched Cascade Lake-X, their new high end lineup, which isn’t actually new but is also a refresh, however the prices per SKU have been cut in half over the prior generation versions to compete with Threadripper. Which would have worked, if it weren’t for the fact that even the 16-core Ryzen mainstream part wins in most comparisons, Threadripper widens the gap with even larger wins, and Intel’s shady marketing tactics (moving their review embargo on these parts to 6 hours before the third-gen Threadripper launch to avoid direct comparisons) made them less popular, even as many reviewers simply held their content until the later Threadripper embargo time (or, like Linus Tech Tips, showed ominous hidden bars of performance much larger than the new Intel parts).
Intel’s woes through the last 5 years, but especially in the rise of AMD, lay with their fabrication process. Intel, unlike modern AMD, fully manufactures their own chips, including one of the largest silicon foundry businesses in the world. As a part of this, Intel creates and manages their own manufacturing process. Intel has, since 2015, been with a 14 nanometer process, which has worked very well, but their next move, to 10 nanometer, has been heavily delayed and only laptop hardware has started shipping this year in 10nm. The process has tradeoffs thanks to the amount of time Intel spent on 14nm, so while the new Ice Lake 10nm mobile CPUs have a solid increase in IPC over the 14nm parts, they cannot scale to nearly the clock speed that 14nm did, so while they gain up to 18% in IPC performance, they lose almost the same amount of clock speed. This has led to new rumors that Intel is simply going to skip 10nm on desktop altogether, going straight to a next-generation 7nm process instead. This leaves AMD with more opportunities going forward, as that means the likely future on desktop is Intel continuing to scale the same architectures or struggling to backport their 10nm designs to 14nm while AMD has the next two generations of Zen architecture in the labs with leapfrogging design teams.
2019 was a solid, if unremarkable, year for GPU technology. The year started with a mainstream-viable Nvidia RTX card, as CES saw them announce and launch the GeForce RTX 2060. Since then, however, Nvidia has largely rested, launching better-binned GPUs under the “Super” label and launching a low-cost variant of the RTX GPUs without raytracing hardware under the GTX 16xx lineup.
AMD, meanwhile, started the year with the Radeon VII, a low-production card designed to use salvage silicon from workstation graphics cards to keep the Radeon name in the conversation for gamers. It was a decent enough card in theory, as the 7nm Vega GPU powering it was still pretty good and it having 16 GB of HBM2 memory with an effective 1TB per second of memory bandwidth was a first in the consumer market. However, Vega itself was a troubled architecture and the card was never intended to stay out for long. The summer launch of the third-gen Ryzen CPUs from AMD was coupled with the launch of their Navi GPUs in the Radeon RX5700 and 5700XT lineup. These cards, through what AMD calls their RDNA architecture, made some fundamental improvements to the GCN design that AMD has used for close to a decade, which has resulted in some good improvements in horsepower in the new cards. The launch was plagued with AMD-designed reference cards that had poorly-designed blower coolers on them, which get terribly loud under any reasonable workload and still don’t keep the parts that cool, but as late summer and early fall came, partner designs wtih far better coolers came out that made the GPU more worthwhile.
In performance, AMD meets Nvidia in the upper-midrange segment, with the RX 5700XT matching the original GeForce RTX 2070 and getting close to the RTX 2070 Super. It has, however, marked another year where AMD’s GPU unit has not kept pace with the high end of the market, leaving Nvidia’s cards from the RTX 2080 upwards to take marketshare uncontested. Rumors for 2020 are that AMD will launch the “big” Navi core, which should give them a viable upmarket competitor, but until then, AMD remains a reasonable choice at lower price points and is doing a better job than they were a year ago.
The future is interesting here, though, as next year around this time we’ll likely also be discussing Intel’s Xe GPU lineup, which is very likely to change the game in the GPU space in a huge way. Especially once you begin to consider the partnerships Intel has with system integrators and the large footprint they have in the datacenter, the idea of a competitive Intel in GPUs is a fascinating prospect.
In 2018, I finally went from a standard SATA solid state drive to an NVMe drive, a small stick of gum-looking circuit board attached directly to my motherboard. The speed is insane – so insane that storage isn’t really the bottleneck in most things I do with it anymore. It holds Windows, my drivers and core applications, and my main games. It hasn’t made a night-and-day difference mainly because I was already using a fast-enough SATA SSD for those things before, but it nets me marginally more time – saving 1-2 seconds more per loading screen in WoW and saving up to 5 seconds per screen in Final Fantasy XIV has been nice!
2019 has seen price cuts that make SSD technology, whether SATA or NVMe, far more attainable for most people building a new system. Improvements to software caching setups that use a faster SSD to buffer a slower spinning hard drive have made them more interesting, too, although such technology still means first boots or playing a game after a long absence from said game takes a while to cache. 1 TB of SSD space has come down to a very affordable level, and if you’ve never had one before, the single biggest improvement you can make to a new machine is to get an SSD to use for booting your OS and your most played game. Being able to easily afford a terabyte of this storage also means that the compromises of the early-era SSD are gone – my first SATA SSD was a whopping 60GB and it took a lot of cautious effort to ensure I did not max it out with Windows and one game. Now, I have Windows and about 8 games installed to an SSD, alongside drivers, apps like Chrome, Adobe CS, and Sketchup, and I even had room in my budget for a separate SATA SSD to use as backup storage for speed-senstive files like video used for rendering.
2019 saw Samsung push their QLC flash for SSDs, which has meant cheaper, higher capacity SSDs from them at the cost of some read performance and much lower sustained write speeds – which, for the average person, doesn’t matter much anyways. External SSDs for backup are becoming more common, and just in the last few weeks, Western Digital launched a “gaming” external SSD that uses USB 3.2 Gen 2×2 technology to connect an external SSD over a USB-C port that runs fast enough to match most internal SSDs, allowing you to port games and files around very quickly.
Storage is an underrated technology in my opinion, especially for gamers – the true open worlds and no-loading screens future I want depends on fast, low-latency storage with strong sequential and random performance becoming the norm. Yet today, other than initial loading times or load screens between destinations, gaming today is built to work around storage rather than with it. My MMO experience is limited by server response rather than storage speed, but if rumors of the next-gen consoles getting NVMe drives instead of slow laptop hard drives is true, then I would expect games to get hungrier for storage bandwidth very, very soon.
What Does The Future Hold?
2020 should be an exciting year in tech. Zen 3 from AMD should offer an incremental upgrade, although rumors point to a similar IPC increase to that of Zen 2 over the original Zen design, which would make them worthwhile to consider for a lot of people, even those owning Zen 2 CPUs. Mobile should be more competitive as 2020 will have Zen 2 based laptop processors from AMD to compete with Ice Lake – a battle which Ice Lake can potentially win, but it would at least put AMD into the upper level of laptop performance and would allow for higher clock speeds in their mobile offerings. Graphics will hopefully be interesting – Nvidia has a deal in place with Samsung to manufacture their next-generation GPU’s on Samsung’s 7nm silicon process, which hopefully will see an easier ability to integrate RTX features over a smaller physical footprint to allow more room for traditional rasterization hardware to push GPU performance ever higher, while AMD’s big Navi or even Navi 2 are rumored to launch, which may help push AMD into the high-end market again (or might not, depending on thermal performance of the larger GPUs given what we’ve seen of the stock design of current Navi in the 5700 lineup).
I’m most excited to see the new consoles start to come out, in actuality. I haven’t touched on them here because 2019 saw only the barest mentions of them existing without real specs, hardware previews, or pricing. Rumors point to some fascinating advancements that should help nudge PC gaming forward. The biggest news to me is that both Microsoft and Sony have apparently made deals with Samsung to put NVMe SSDs into the new consoles. If this is true, then fast storage being an assumed default in consoles means that new games should begin taking advantage of the speeds on offer there. Both systems are also going to be moving forward to Zen 2 based CPU designs with Navi-based GPU designs from AMD, but I say “based” because there is scuttlebutt that the CPUs may not use simultaneous multi-threading (which is a core Zen feature, but not one that many games benefit from outside of margin of error framerate increases or odd decreases for poorly multi-threaded titles) and that the GPUs feature hardware raytracing (which is NOT a feature that currently exists in Navi, outside of the DXR DirectX implementation, which uses the programmable shader hardware in every GPU since 2008 to crunch the numbers on raytracing a scene very, very slowly). The possibility is that these are somewhat custom parts, as has been the trend since last generation – AMD’s business salvation was in their “semi-custom” hardware, taking their base designs from mainstream PC hardware but making small tweaks for the consoles. For example, both the Xbox One and PS4 are built on variants of the AMD “Jaguar” cores – which were available in some of their APU products, but weren’t their core lineup. These parts were tweaked to lower clock speeds for better power efficiency, and then partnered with a GCN-based GPU that doesn’t quite neatly fit to a single design on the desktop (and changed over time anyways, as the PS4 Pro and Xbox One X use upgraded versions of both components, although both manufacturers have different designs of the same base hardware in use).
PC gaming has comprised largely of unique offerings that are PC-first and PC only, and games that are designed for consoles and then ported to PC (pretty much every AAA title that you see on both platforms). If the base consoles are no longer using hardware with origins in 2012/2013 that have only received minor tweaks and upgrades in refresh designs, it follows that PC gaming will quickly begin to see increases in visual fidelity, storage requirements, and implementation of technologies like raytracing more regularly. That is exciting to me because I like the idea of seeing games push more boundaries (and I’m probably eyeing an upgrade to whatever 7nm GPU Nvidia releases next year, since I’m currently in a spot where only the GeForce RTX 2080 Ti or Titan RTX are upgrades in all cases for performance, but want something new to push harder and to justify doing a system rebuild to finally move to a custom watercooling loop).
In terms of games, well, I’ll write that up closer to the end of the year, because 2020 has at least one major release I’ve already technically purchased (Shadowlands) but also content updates for other games I play!