Another year has come and gone. Over in the blink of an eye. If you are, like me, nearing your thirties you’ll have no doubt experienced some form of time dilation when you wake up and realise that it feels like it was just yesterday that you made your New Year’s resolutions for… what year was it even? Well, that’s what the tech industry wished it had, because 2018 was one big headache for everyone involved. Let’s review this sucker.
Spectre and Meltdown
Spectre and Meltdown are two classes of vulnerabilities that exist in almost every modern processor to date from Intel, AMD, and ARM, and they dominated the headlines when they were leaked a few days early of an official announcement from Intel along with a disclosure of the vulnerabilities they faced. Spectre was a vulnerability that allowed you to leak information from another current process in a core, yielding decryption keys and passwords in modern software. Meltdown granted you full kernel memory contents on Intel’s processors.
The combination of these attacks and their ramifications ripped through the industry like wildfire. Intel battled to contain the PR meltdown it was experiencing as new variants of these same attacks surfaced week after week, with all eyes on them as researchers turned their attention to looking for new attack vectors. Intel’s messaging was confusing and desperate, because not only did these attacks have workarounds, the workarounds they had settled on for older products came with severe performance penalties. Because of the risks, cloud vendors immediately took Intel’s fixes as they were available, and saw performance in managed services and virtual machine infrastructure fall 10-35% depending on the workload. For Intel, this was bad news. Worse still was the news in February that viruses using Spectre for attacking web browsers were already in the wild.
Meltdown and Spectre are now mitigated against in all operating systems and on all affected platforms that are able to receive updates. Intel, having known about the vulnerabilities as far back as June 2017, had already designed hardware mitigations in their new Coffee Lake refresh processors. New attacks are still discovered every month, but they now affect unpatched machines that cannot run the mitigations, and are more academic exercises today.
Intel’s continuing PR nightmare
If Spectre and Meltdown weren’t enough, Intel had a much harder time cleaning up their PR in the industry. The removal of CEO Brian Krzanich, ostensibly for having a workplace relationship with a subordinate (who may or may not have been his current wife), wasn’t very clean. Intel kept quiet despite speculation from the media that Krzanich may be guilty of insider trading, using the opportunity to put CFO Bob Swan into the driving seat, and the board instituted a number of departmental changes. Intel is still looking for a permanent CEO.
Swan was out of the frying pan and into the fire as rumours of chip shortages and another 10nm process delay arrived in the second half of 2018. Intel’s shortage was caused by the company shutting down much of their 22nm production lines on schedule, but did not increase the number of 14nm wafer production lines. There was an expectation from Intel’s foundry department that 10nm would ship on time to relieve some of the pressure, but Intel was only making one 10nm product – the much-maligned Core i3-8121U that shipped with integrated graphics disabled on the die. The result was an internal shortage of 14nm wafers, and this didn’t help as Intel was right in the middle of preparing for the launch of Coffee Lake Refresh and a bunch of Xeon processors with upwards of 22 cores. Intel publicly dressed this “demand” for their products, and this created the impression that consumers were buying up so much product at once that Intel’s foundries couldn’t keep up.
The shortage affected all of Intel’s partners, from HP and Dell to Pegatron, and everyone making motherboards. The shortage also affected all of Intel’s 14nm products including modems, processors, and chipsets. All of Intel’s partners adjusted their revenues and unit tallies lower, and many shifted their attention to AMD, who had just started selling their Ryzen 2000 series chips a few months prior. Intel finally ended speculation by pushing the launch of 10nm processors to late 2019 and invested over a billion dollars in removing 10nm capabilities from some machines and switching them back to producing 14nm wafers.
This decision has cost Intel far more than a billion dollars. Its affects will only be seen in 2019 as AMD pulls ahead with their popularity of their Ryzen processors in US and EU markets, while Intel still struggles to meet internal demand for 14nm wafers for all their products. Intel, for the record, announced that they were prioritising production of high-end silicon like their Xeon processors, limiting the production of cheaper chips such as Pentium and Core i3 processors. In response, retailers ended up hiking prices of their Core i5 and i7 chips almost 50% in some cases because they couldn’t source stock, and you’ll have no doubt noticed the lack of new budget notebooks from vendors lately. The Pentium N5000 is about all there is to find on store shelves, everything else is from the first Coffee Lake release.
Moving forward, Intel will have to do a lot to steer the ship back on course, and they’ve hired a lot of talented people to come in and fix their issues internally, but time will tell. As I write this, there still isn’t an update on the 10nm process, and Intel has bizarrely claimed that the 7nm process is on schedule.
Memory price fixing
While all this was going on, extraordinary claims and reporting on pending court proceedings suggested that memory manufacturers Samsung, SK Hynix, and Micron were being investigated by the Chinese government on suspicion of price fixing and collusion. If this sounds familiar to you, it’s because this has happened before, and attracted heavy fines for the memory makers at the time. Hagens Berman, a law firm in Seattle, even filed a class action against all three companies.
Evidence on the internet does appear to show cartel-like tactics from Samsung and Hynix. In September 2018, Samsung announced that the company was not going to increase production of flash memory in order to maintain their pricing. Not long after, Micron said nearly the same thing, warning investors about a slump in profits. SK Hynix said the same a month later. The announcements could not have come at a worse time as Intel was also struggling to manufacture enough wafers, which meant that memory sales for new systems similarly dropped. The memory market, negatively affected by Intel’s woes, sought to keep their high pricing while they continued with their expansion plans with new factories that would come online in 2019.
And the winner of the prize for worst product launch of 2018 is…
Microsoft and NVIDIA both share the trophy for worst product launch. In NVIDIA’s case, it lost favour among enthusiasts as Jen-Hsun Huang’s keynote address for the launch of GeForce RTX did not go well. Huang, unable to really put into words why anyone wanted real-time ray tracing in videogames, became meme-worthy with phrases like, “It just works!” and “The more you buy, the more you save!”
NVIDIA sunk further with the actual launch of the RTX 2070, 2080, and 2080 Ti. While the Ti was a true flagship in terms of performance, NVIDIA shifted the price-to-performance bracket one tier upwards, charging consumers $100 to $300 more for the same performance as existing products in the tier below. NVIDIA justified this through features like RTX and DLSS, and touted superior efficiency in their latest products. But for the same price, consumers could pick up last-gen products for less money, and RTX didn’t change things that much.
Those features also shipped late. RTX had to wait for the arrival of Windows 10 1809 into the consumer channel, and DLSS was only present in a single benchmark on launch, Final Fantasy XV. Of the 24 games announced to support DLSS, none of the games currently shipping actually have proper support for it.
RTX arriving late was all Microsoft’s fault. Microsoft botched up both major upgrades they bring to Windows 10 every year. The 1803 update was delayed over a month, finally shipping in early May 2018. Microsoft never revealed the official reasoning behind the delay, but some bugs causing BSODs and full system freezing might have played into it. With the silence around 1803, it was expected that Microsoft would push ahead with 1809 and put it straight into the hands of the public. Well, it did all that and then some. The company rushed to complete the update on schedule and chopped out planned features to be able to meet the deadline. It wasn’t just consumers that wanted the updates on time, it was also Microsoft’s partners. Without a shipping version of 1803 or 1809, OEMs would be unable to market Windows 10 support with key features available in these updates.
The rush to complete 1809 ended in disaster for the company. Not only had Microsoft completely ignored Insider feedback on key issues that included data loss, they shipped 1809 with pending updates for Intel graphics and sound drivers which caused BSODs, broke many HP notebook installs thanks to a keyboard driver conflict, and had not fixed lingering issues highlighted in 1803, particularly the animations for Timeline and the inconsistencies in the user interface. Microsoft pulled the 1809 update and never said a word about its status or ETA for a re-release, until the company later settled on an October 2018 launch for the update. Still more bugs popped up, including another one that deleted user files, and a separate cumulative update that broke Windows Media Player and any application that relied on it working.
If that wasn’t bad enough, Microsoft later admitted in a blog post published in early December that all users who checked for updates manually in the Settings menu were fed “betas” of updates the company was testing. The due diligence on the part of the users, who for decades had the “check for updates daily” mantra drummed into their heads, was being turned into acceptance of their being test subjects. In the blog post, Microsoft blames the quality of the information coming out of the insider program for their issues – a hoot for Insiders, who discovered this year that the company doesn’t listen to them anyway. Microsoft also admitted that they had been using the Insider program for focus testing and A/B testing for new features, exactly the opposite thing that the Insider program was designed to do. 2018 was not a good year for Microsoft.
But enough of the gloom. Let’s celebrate the positive changes.
GPU mining is one and done
It’s been months since GPU mining for altcoins has been a thing, and the market has returned to sane pricing and generally good availability of parts. The obscene loss in bitcoin’s price has been the main driver of change, as small-scale miners finally hopped off the bandwagon to minimise their losses and move to better things. With that, we no longer have shortages of components needed for mining rigs, and can now buy our computers for mostly cheap prices.
The waning interest in bitcoin also means that there was an overstock of GTX 1060 dies sitting in NVIDIA’s warehouse. The company wanted to sell these during the height of the mining craze, but couldn’t move enough stock before the crash hit the market. They’ve been trying various things to do that, such as sell the cut-down GTX 1080 variant of the GTX 1060 die with GDDR5X memory to help deplete inventories.
SSD prices are through the floor
In December 2018 I bought a HP 120GB SATA SSD for my brother’s laptop for R599 from Makro. A year ago, this same drive would have cost twice as much – and the 500GB version of the same drive is just R2,000. SSD storage has become insanely cheap in the last six months, and it’s set to drive down even lower as higher density NAND hits the market and companies move to using newer controllers that can push 3D quad-level cell storage (QLC) further down the range. Pretty soon, we’ll see the death of 120GB drives because they’re uneconomical to make. If you own a computer that has an open M.2 port, it costs you very little to now move to much faster storage for a better computing experience. There’s no reason why anyone anywhere should ever have to rely on spinning rust ever again.
AMD sees success
While Intel flailed around trying to address all the rumours and leaks and bad press it was being bombarded with, AMD sailed through 2018 with a stellar year. They launched several new products in the APU and CPU segments, and set records by shipping 32 cores on a consumer desktop socket for the first time. The company can now lay claim to the title of fastest single-socket server system thanks to the 64-core EPYC processors they now ship, and the final ramp-up of first-gen EPYC sales to customers should now be reflected in their books for the fourth quarter of 2018.
Ryzen also continues to sell well. In Europe, shortages of Intel products meant that sales of AMD processors skyrocketed, tripling in a single month as shown by German retailer Mindfactory. This success has been reportedly seen all over Europe, and here in South Africa AMD’s market share is over 10%, inching closer to 15%. AMD’s desktop strategy has been working very well, and their mobile strategy can only improve from here.
Graphics-wise, the company is still hanging in there. By lowering prices to compete with the RTX range from NVIDIA, the Radeon Vega 56 and 64 are still good value for money compared to the RTX 2070, and AMD’s just launched RX 590 clearly beats the GTX 1060 for less money. AMD is expected to finally revamp their GPU lineup in 2019 with the release of Navi, but there’s no telling when this will arrive.
Finally, RISC-V has arrived
In a final nod to good things in 2018 tech-wise, we have RISC-V just about giving the industry the heebie-jeebies because it threatens the status quo. RISC-V is another RISC architecture like ARM, but unlike ARM it is being developed as an open platform to which others can contribute and copy for free. There are some companies investigating RISC-V as an alternative to x86 should Intel or AMD flounder in the near future, but what’s more likely is that some companies will begin looking at rolling their own chips for specialised devices, because RISC-V is modular. The hardware is currently working, Linux is running on it, and someone got DOOM to run without much trouble. If it runs DOOM, it will eventually run all manner of things from a Plex server to Skyrim.
Why is it sending Intel and AMD and the rest of the industry into a tizzy? Unlike ARM, RISC-V does not attract any licence fees because it is an open platform. There’s no licences for the architecture, and you don’t need to share your improvements with other companies (although, it would be beneficial in the end to do so). A lot of companies are looking for an alternative to ARM because they want to reduce their R&D spend, and they don’t want to hitch their wagon to x86 because it is old hat. RISC-V allows them to use it wherever they please, and we’ll probably see some RISC-V announcements coming out of CES this coming week.