Sidenote: CES Week Rumors and GPU Hell Gets Worse

CES is the first major milestone for the year in tech, and while 2021 is going to be a virtual event, that doesn’t mean that there aren’t going to be major announcements.

So this week, expect to see a lot of new technology announced. What is exciting with CES is that for every story of an experimental tech TV that costs $100,000, in PC gaming, the event is much more down to earth, generally.

So let’s talk about the major categories of PC gaming products and what to expect!

CPUs

CES 2021 is all but confirmed to have two fairly major CPU announcements. The first and easiest to cover (mostly) is AMD, who will be bringing the Zen 3 cores of the Ryzen 5000 family to mobile with new laptop silicon. These CPUs start a new APU generation, reusing the Vega 8 graphics of the Ryzen 4000 lineup but with new Zen 3 CPU cores and marginally higher clock speeds. These are also expected (eventually) on desktop, which is only true of the Ryzen 4000 APUs if you buy a prebuilt system or find an OEM supplier who has an unboxed tray CPU on-hand. It remains unclear how PCI-Express lane allocation will work (while based on Zen 2, mobile Ryzen 4000 had no support for PCI-E Gen 4, had only 8 lanes for a discrete graphics card, and because of the lack of PCIE Gen 4, was stuck with slower SSDs (I mean, very minor issue, but one all the same). Whether or not you can buy a laptop with these new CPUs remains to be seen!

While a Ryzen 5000 Threadripper lineup has been rumored, I can’t imagine AMD will launch that lineup with Epyc Milan Zen 3 parts still not shipping in volume. Unless all the Zen 3 chiplets are leaky, high-power silicon, in which case it could be announced for a later launch, I suppose. The only real rumors I’ve heard about Threadripper is a launch of a small number of WRX80 workstation motherboards for Threadripper Pro 3000 CPUs, which have currently only been available in Lenovo workstations, and a rumor that the Threadripper 5000 series will have a 16-core part as well. This might sound weird at first, considering you can get a 16-core AM4 Ryzen, but for segmentation, it makes sense. Threadripper has much more IO capacity (a desktop Ryzen CPU has 20 lanes direct to the CPU on-offer, enough for most users to get 1 GPU and 1 NVME SSD, but Threadripper offers 64, meaning they make de facto better systems for non-gaming multi-GPU usage, high throughput storage with RAID-arrayed NVME SSDs, or things like add-in cards for 10 gigabit LAN, multiple capture devices, and other professional devices. I would be a bit surprised to see this announced with a launch date, but I could see them touching on it.

Intel, meanwhile, should be firming up details for their 11th-gen desktop launch with the Rocket Lake codename. These CPUs, which I’ve discussed previously, lock in at 8-cores maximum on Intel’s aging 14nm lithography. However, rumors point to the top-end chip being a beast. It will be Intel’s first new core design on desktop since 2015, and the early rumors show it possibly retaking the single-threaded performance crown, on a mix of increased IPC and maintaining 5GHz+ single core boost frequencies. If this is indeed true, I fully expect a demo or live benchmark during their video presentation. Rocket Lake also finally drags Intel to PCI-E Gen 4, meaning faster GPU and SSD interfaces, and with a new core design, it should in theory be free of the crippling security vulnerabilities that Intel has faced over the last 3 years and had to patch and repatch. The rumored launch for these is in March, and alongside them will be a new Z590 motherboard chipset (Z490 is said to still work, as the signaling hardware needed for PCIE Gen 4 was built into most motherboards and the socket remains LGA1200), so expect to see around a dozen new motherboards from everyone.

GPUs

On the Nvidia front, they seem poised to announce at least one GPU, and maybe, maybe a second one. In the Turing era of RTX cards, RTX stopped at the 2060, with all the comfortably midrange and mainstream cards being the bizarrely named GTX 1600 series. Currently with Ampere, the lineup stops at the recently-launched RTX 3060 Ti, a pretty good card at its $400 MSRP that meets and exceeds the performance of the GTX 1080 Ti. The curveball is this – the RTX 3060 is looking fairly likely to come in with 12 GB of VRAM, more than the higher-tier 3060 Ti and the 3070, oddly. This amount of VRAM is meant to counter AMD’s offerings almost fully, as the resolutions the RTX 3060 would best serve will be unlikely to use it during the service life of most of these cards. At this point, the RTX 3060 is almost a certainty – leaks of an Asus GPU box show all the pertinent details, backed up by a Lenovo Legion desktop part listing including this option.

The less-likely launch, in my eyes, is the RTX 3050 Ti. This card was also seen on the Lenovo leak, and shows a 6 GB VRAM card built on the same GPU die as the 3060, GA106. This would mark the first time Nvidia’s RTX technologies make their way into an xx50 GPU, and this card would likely come in between $200-$300, making it an attractive value play. RTX in the low-end cards isn’t good for actual ray-tracing – it has far too much of a performance hit even on the higher end cards – but it would mean a cheap card can run DLSS, which would allow access to higher resolutions via intelligent upscaling. However, there is also a rumored RTX 3050 non-Ti at 4 GB, so my suspicion is that Nvidia will baby step down the lineup, with a 3060 at $350, 3050 Ti at $300, and the 3050 at $250. I fully expect this because Nvidia remains able to keep prices high with insufficient supply (and we’ll talk more about this in a minute, because…yikes!)

Lastly, while the rumor mill has been continually full of RTX 3080 Ti/3070 Ti word, with higher-spec GPUs and doubled VRAM, it seems like Nvidia will remain happy to ride the wave of selling existing versions of these GPUs prior to launching these cards, which seems to be a February possibility.

For AMD, the only likely news is a Radeon RX 6700 lineup, in all likelihood with both an XT and non-XT variant. Both of these cards are likely to be on the Navi 22 silicon, with around half the CUs on-die at 40 max for the XT version and likely 36 CUs for the non-XT. Both should have 12 GB of VRAM on a reduced-width memory bus, but what we don’t know is if Infinity Cache is included or not. My guess would be that it uses a nearly fully-cut down half-die of Navi 21, and would thus have 64 MB of Infinity Cache. On the higher-tier parts, the cache hit rate for 1440p and 1080p resolutions is much higher than the slightly over 50% at 4k, which leads to the performance leads the Radeon RX 6800 XT can pull out over even the RTX 3090 in some benchmarks at these lower resolutions, while at 4k, the Nvidia cards often win and by a good margin. With the likely CU counts, the RX 6700 series will be a strong current-gen console match, as 36 CUs is the number in the PS5 and the likely 40 CUs of an RX 6700 XT would be right about in the middle of the PS5 and Xbox Series X. Cost-wise, I expect these to be marked up slightly over the RX 5700 equivalents, with a likely $450 MSRP on the XT and a $400 for non-XT. They could always do the incredibly stupid “jebaited” shit they did with the 5700 lineup and drop prices in response to an Nvidia announcement, but I would hope not, because the 5000 series Radeons were full of this kind of response – between late-changing prices and a so-late-it-wasn’t-in-first-batch-cards change in memory clocks for the RX 5600 XT.

Other Gaming Tech

Expect to see some high-speed DDR4 memory kits discussed, a thing that a lot of manufacturers have started pushing. Nearly every memory brand now offers some lineup of highly-binned, hand-tested modules with DDR4 clock speeds in excess of 5000 MHz, and the midrange of most lineups is moving up rapidly in speed as we get closer to the first DDR5 platforms in the next year or so.

A standard rollout at CES is announcements of new CPU coolers, fans, case designs, and water cooling components. In years past, CES generally brought the initial announcements with no clear release dates, and then usually at Computex in June, release dates would be locked in for the fall and winter. This year, as Computex is likely going to end up being a virtual event too…well, it is much harder to predict. CES may very well see some actual launches for these types of components, and it is especially likely given that a lot of products that would have likely launched in 2020 in a better world were delayed. It could be more interesting, as CES generally is a proof-of-concept phase for many of these products, but seeing more finished and ready products would be fascinating.

For gaming overall, both PC and console, expect to see a larger variety of HDMI 2.1 displays. First-gen HDMI 2.1 TVs have had some teething problems with the current-gen consoles, and some displays use lower bandwidth than full spec, which leads to quality reductions, most notably with HDR, where a lower bandwidth requires chroma subsampling to bring enough of the quality while being able to manage the reduced speed connection. Other HDMI 2.1 devices should be seen too, like new audio receivers, capture cards, splitters, hubs, and probably some snake-oil cables!

Lastly, outside of what I might personally call gaming tech, Samsung is likely to launch a new Galaxy smartphone series, and as mobile is the largest gaming platform in the world, that is an advancement worth watching.

The GPU Hellscape Gets Worse and Worse

At the end of 2020, I wrote that expectations were high that we might see some improvements to the GPU availability landscape, but not likely until 2H 2021. I also wrote triumphantly about getting my own RTX 3080 finally after months of monitoring the availability situation and trying to jump in a number of times, which is very lucky in retrospect.

What is worse than expected now? Well, on top of continued low availability in general, a handful of issues are compounding to make things even worse. Firstly, supply constraints continue, with core components like substrates, interposers, and capacitors remaining in short supply, alongside reduced supply of PCBs and a chain of supply hits to GDDR6 memory, which is the backbone memory technology of all new GPUs except the RTX 3080 and 3090, all 3 new consoles, and a myriad of older GPUs that remain in production. While the port and shipping situation no longer has the holiday rush, those remain longer-term problems and we’ll be waiting for decisions made late in 2020 to have an impact for several more weeks still yet.

However, a new contender has arrived, and absolutely wrecked the GPU market in the US!

The tariffs on Chinese-manufactured goods, enacted by President (and attempted coup-encourager) Donald Trump, previously contained an exemption on tariffs for several categories of components, graphics cards included. As of the new year, the GPU exemption has expired, which has brought the a 25% tariff into play on newly imported graphics cards. For the sake of consistency (and some profiteering on the cards already in the US, of course!) Asus has been the first brand to step forward and increase pricing, with egregious examples such as their Strix RTX 3090, previously $1,799, now selling for $1,979, and their RTX 3070 KO going from $549 to $639.

Other GPU brands have yet to really step forward and confirm what this will mean for them – several brands have been working on moving production lines to Taiwan or other southeastern Asian countries to ensure this very event would be mitigated, and many manufacturers are holding action until they see how things develop. With a new incoming US administration, we could see these tariffs removed or reduced, although that isn’t guaranteed. For some higher-end brands, they may have the margin to eat the tariff on some product lines, but this is a big breaking point for some. The pre-launch word on the RTX 3060 Ti, as reported first by Gamersnexus, indicated that there was a lot of pressure from Nvidia for board partners to hit aggressive sales prices that eat into their margins, force them to cut costs through using cheaper coolers and lower-tier component selections for power delivery and PCB, or both.

For a 3090, some partners may very well have margin built in to allow them to maintain pricing, or to use the margin here to offset losses elsewhere – after all, everything is selling out right now! However, many companies cannot or, more likely, will not do so – seeing Asus move to charge the difference immediately, including on cards likely already cleared through customs prior to the change, is wholly unsurprising – and I suspect many other brands will follow suit. However, moving quickly to change could be a problem if the tariffs get reduced or removed – you then either have to switch back rapidly, or hold the line and justify the price increase continuing on.

Then, to top things off, Ethereum cryptocurrency mining has grown, and both Nvidia’s Ampere cards and AMD’s RDNA 1 cards have been flying off the shelves into crypto rigs, creating a situation akin to that of the prior cryptocurrency runs on GPUs in 2017 and 2018, but worse because of reduced availability. And, to make matters even even even worse, due to iffy deals with crypto farms during the rush of 2017 and 2018, both AMD and Nvidia are likely to not push production much higher than what they suspect gamers will buy, as AMD was burned by overproduction of Vega cards, while Nvidia faced a shareholder lawsuit that was dismissed in 2020, because of their claims that the gaming segment was not being driven by cryptocurrency mining. AMD Radeon RX 5000 series cards that remain available have started increasing in price due to mining, and a part of the Ampere shortages is due to, once again, backroom dealing from miners with suppliers to get their hands on handfuls of cards.

Now, CES announcements of new GPUs are still worth being excited about, because mid-range cards tend to be easier to get during mining crazes because they cut enough hardware out to reduce mining efficiency, and the much-smaller silicon dies mean they are vastly easier to manufacture, yield, and produce a finished card for. They also require a smaller quantity of components on the card – a smaller power delivery, fewer power connectors, smaller coolers with fewer fans, and less filtering capacitors. If these cards are already produced and imported, it could be fine – and even then, a 25% tariff would be less impactful on a lower priced card. However, currently people in the US have or are in the process of receiving a second round of economic stimulus, and that may drive additional GPU purchases (I expect that with the broad audience for console gaming versus enthusiast PC gaming, this is actually more impactful to console supply).

Lastly, with the console point being a good segue, AMD CPUs and GPUs remain in short supply (Frank Azor’s tweet ages worse and worse by the second!), and we may see a new round of issues here. AMD has a continually limited supply of wafers for 7nm silicon production at TSMC, and they are adding more products to that stack with a new GPU, new mobile CPU lineup, and continuing to work to build out enough parts for the Epyc Milan launch this quarter, on top of existing Zen 3 desktop CPUs, Zen 2 chiplets for datacenter and desktop, Zen 2 mobile, Navi 21 GPU dies, and 3 different console SOCs. AMD has even had this boil up publicly, as Microsoft publicly stated on an Xbox podcast that they’ve been pushing AMD to get a larger allocation of SoCs. It may just be an issue of increased demand, but generally, given how such business relationships work, this becoming public seems to indicate some amount of trouble for AMD and may push a higher amount of production Microsoft’s way. Microsoft has two fairly large SoCs in production with AMD, which complicates things even further – if Sony wants more PS5 chips, there is one, and the production scaling is pretty predictable. For Microsoft, though, if the split is for more Series X consoles, that is the largest of the 3 SoCs, which would eat into AMD’s wafer allocation even more.

So in short – CES will have some cool stuff, and also, everything is fucked up if you’re trying to build a PC right now!

One thought on “Sidenote: CES Week Rumors and GPU Hell Gets Worse

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.