In the past year, fans of computer hardware had many reasons to upgrade the hardware component of their PC. Among the reasons conducive to an upgrade, one can single out the fall in prices for RAM and solid state drives, as well as the release of new central processors and video cards. Now, in the last days of December, it’s time to brush up on all those events and releases that made 2019 not boring. The author takes responsibility for choosing the most interesting topics, and you, dear readers, can always express your opinion in the comments under this article.
By tradition, let’s start with the processor market, where a tense struggle has unfolded for the hearts and wallets of PC enthusiasts. The second half of the year definitely passed under the sign of AMD. The “red” chipmaker brought to market 7nm mainstream Ryzen 3000 processors with up to 16 cores, monstrous EPYC server chips and a new Socket sTRX4 HEDT platform along with the first 3rd generation Ryzen Threadripper. Moreover, Lisa Su’s team even managed to get Intel to revise its pricing strategy, but first things first.
Raven Ridge Refresh
For Advanced Micro Devices, this year began with the debut of Picasso APUs. The new APUs migrated to the Zen + microarchitecture, which is a “refresh” of the original Zen, and are produced using the 12-nm process technology at the facilities of GlobalFoundries. The graphics subsystem has not changed and is still represented by the Radeon Vega video core with up to 11 Compute Units.
Roughly speaking, the new APUs are an improved version of the 14nm Raven Ridge processors. Architecture improvements, coupled with thinner production standards, brought a tangible increase in operating frequencies, and by the time they were released, AMD engineers managed to get rid of most of the «childhood diseases» that owners of the first generation Ryzen APUs faced. Thanks to this, Picasso was perceived with less apprehension not only by ordinary users, but also by large manufacturers of finished PCs and laptops.
As with Raven Ridge chips, the company started with Ryzen Mobile notebook solutions. True, this time the lineup was represented not only by 15-watt solutions, but also by more “gluttonous” H-series APUs with a 35-watt heat pack. AMD created them with the idea of gaming laptops, wanting to break the hegemony of the Core + GeForce bundles, which, in general, it succeeded. We will touch on the topic of laptops in the third part of our story.
At the beginning of summer, AMD Picasso hybrid processors made their way to desktop computers. The Ryzen 3 3200G and Ryzen 5 3400G models offer higher frequencies than their predecessors and occupy lower places in the Ryzen 3000 chip hierarchy. At the same time, the older APU boasts solder under the lid, while the rest use a plastic thermal interface. Already in the fall, the company released a budget dual-core Athlon 3000G processor based on the 14-nm Zen architecture, the main feature of which is the official support for overclocking.
Intel over the past twelve months has been able to bring to market a whole host of new products manufactured to the good old 14nm standards. In January, the 9th generation Core processors saw the light of day with an “F” mark indicating a deactivated graphics core. In all other respects, including the recommended cost, these CPUs repeated full-fledged Coffee Lake-S Refresh models. Intel only corrected the pricing misunderstanding in the fourth quarter, providing a $25 discount on all F-series processors.
In April, the planned expansion of the range of Coffee Lake Refresh processors took place. Mobile CPUs with four to eight cores, designed for use in gaming laptops, as well as a huge number of desktop models, including Celeron, Pentium Gold and 35-watt products with the “T” suffix, joined the previously released desktop chips.
Toward the end of autumn, the Intel Core i9-9900KS went on sale, which is a hastily organized answer to the AMD Ryzen 3000 line. The chipmaker proudly calls it «the world’s best gaming CPU.»
The main difference from the Core i9-9900K released a year earlier lies in the operating frequencies: at the maximum boost-mode frequency of 5.0 GHz, all cores operate at once, and not one or two like their predecessor. This provides the Intel Core i9-9900KS with an advantage out of the box, but the owners of the Core i9-9900K can achieve similar performance manually — just overclock the CPU.
Along with products for the mass market, the blue giant regularly releases new 14nm processors for other segments. For example, at the beginning of 2019, the 28-core “stone” Xeon W-3175X entered the shelves, designed to compete with the older 2nd generation AMD Ryzen Threadripper. It runs on monstrous LGA3647 motherboards like the ASUS ROG Dominus Extreme, Gigabyte C621 Aorus Xtreme, and EVGA SR-3 Dark, and the CPU and motherboard bundle will cost several thousand dollars.
«The most expensive processor I’ve ever scalped» — Der8auer
The never-released EVGA SR-3 Dark motherboard
Another interesting release was the server Cascade Lake-AP (Xeon Platinum 9200). They contain up to 56 cores, have a BGA5903 design and are soldered directly to the motherboard. It is curious that until recently, Intel made fun of «glued» AMD EPYC processors, and in Cascade Lake-AP she herself decided to use two crystals.
Cooking pancakes on Intel Core i9-9980XE. Photo by Tom’s Hardware
Speaking of 14-nm Intel processors, one cannot help but recall the shortage of production capacities that the chipmaker faced last year. The growth in the number of cores in mainstream CPUs has a negative impact on the number of finished processors per silicon wafer and at the same time stimulates consumer demand. Over the past year, the corporation has increased the production of 14-nm silicon «wafers» by a quarter, but it still cannot cope with all orders.
AMD announced the era of multi-core
Perhaps the main processor event of 2019 can be safely called the release of AMD Ryzen 3000 series chips. In them, the company applied the so-called chiplet design, bringing together a 12-nm I / O chip and one or two 7-nm chiplets on a textolite, each of which contains eight physical cores of Zen 2. The Infinity Fabric bus is responsible for the connection of individual crystals, already familiar from past generations of «red» CPUs. As a result, Lisa Su’s team has come up with a product that can compete not only with mainstream Intel processors, but also with solutions for the LGA2066 HEDT platform.
The flagship of the AMD Matisse family is the 16-core Ryzen 9 3950X. In terms of raw processing power, it outperforms its predecessor Ryzen 7 2700X by more than two times, and in games it is not far behind the top Intel Core. To date, it is the most expensive CPU for a mass platform, which, however, does not scare PC enthusiasts with a tight wallet. Some of them are ready to stand in long lines to get hold of an outstanding 16-core.
The queue in Japan for the AMD Ryzen 9 3950X. Photo AKIBA PC Hotline
Now let’s move on to the «fly in the ointment». If AMD Ryzen processors of the first generations were not famous for their great overclocking potential, then in the case of the 7nm Ryzen 3000, there is no frequency margin at all. With traditional types of cooling, it will not be possible to overclock all the cores to the maximum mark of the boost mode. This fact is recognized by AMD representatives themselves. It is also worth mentioning the unsatisfactory performance of dynamic overclocking algorithms, which enthusiasts are forced to correct on their own.
Together with the Ryzen 3000 series processors, the AMD X570 chipset debuted. It brought support for the PCI Express 4.0 interface to the mainstream desktop segment and, roughly speaking, is the same I / O chiplet that is under the cover of the new CPUs. As you know, previous hubs for Socket AM4 boards were created by ASMedia. The main disadvantage of the new logic was high heat dissipation (around 10-15 W), which is why most motherboards are equipped with active cooling of the chipset.
Race for nanometers
At the end of the summer, Intel released Ice Lake-U/Y mobile processors, manufactured using a 10-nanometer process technology. Formally, this is the second generation of 10-nm CPU. Last year, the Cannon Lake-U family debuted, or rather, one single dual-core Core i3-8121U chip. The processor giant tries not to remember its existence and even stopped the production of NUC Crimson Canyon mini-computers based on it, although the life cycle of such systems is several years.
But back to Ice Lake processors. In them, Intel used the Sunny Cove microarchitecture, which carries an 18% increase in IPC (the number of instructions executed per clock) relative to Skylake, and the Gen11 video core. In terms of integrated graphics performance, Intel has finally managed to catch up with AMD with its Ryzen Mobile hybrid chips. True, for this it was necessary to use high-speed LPDDR4X-3733 memory, while rivals are content with DDR4-2400.
The production volumes of 10-nanometer products do not allow such CPUs to become a mass phenomenon, and there is no need to talk about the release of desktop chips yet. Today, Ice Lake chips can only be found in laptops and tablets from the top shelf.
For relatively inexpensive devices, Intel offers Comet Lake-U/Y processors, which are produced on a 14nm pipeline. It is noteworthy that without significant changes in technological standards, the chipmaker managed to increase the number of cores in the flagship Core i7-10710U CPU to 6 pieces, however, in order to fit into the 15-watt TDP, the nominal frequency of this «stone» had to be reduced to 1.1 GHz. In a single-threaded load, it can dynamically overclock up to 4.7 GHz.
The chip design that AMD has used in the Ryzen 3000 desktop processors has shown itself in all its glory in the 2nd generation EPYC chips. With the move to the Zen 2 architecture, the number of cores in the “red” CPUs for the server market has exactly doubled: from 32 to 64 units.
In top-end products, under the lid, there are eight crystals with x86 cores and a large 12nm I/O chip, which contains an 8-channel DDR4-3200 RAM controller and provides 128 PCI Express 4.0 interface lines. The amount of cache memory in the third level in them is an impressive 256 MB.
Photo by OC_Burner
Of course, the very outstanding characteristics of the 2nd generation AMD EPYC chips did not go unnoticed by both corporate customers and other organizations, and overclockers. The latter began to update world records in those disciplines that take into account the total processing power of the processor, for example, Cinebench test suites.
Another area where the new AMD EPYC perform well is supercomputing. Over the past six months, the «red» managed to conclude many contracts for the supply of CPUs for systems of this class. In particular, 7nm EPYC processors will form the basis of the most powerful supercomputer in Europe: by May next year, the Archer2 supercomputer with a peak performance of 28 petaflops will be launched in Scotland.
Archer supercomputer powered by Intel Xeon processors will be shut down in February 2020
Meanwhile, in China, the production of server (and not only) Hygon Dhyana chips, related to the first generation EPYC processors, is in full swing. They use the Zen microarchitecture, accessed by AMD’s Chinese partners through a joint venture. These CPUs are intended for the domestic market and do not officially leave the borders of the Celestial Empire.
Hygon Dhyana server processor based on Zen architecture. Photo AnandTech
The new king of the high-end segment
Knowing about the upcoming release of the competitor’s new HEDT platform, as well as assessing the threat from the Ryzen 9 3950X, Intel decided to reconsider its pricing policy in this market segment. If earlier the 18-core flagship LGA2066 was estimated at almost $ 2,000, then in the case of the recently released Cascade Lake-X processors, the top chip is available for «only» $ 979.
The recommended cost has become the main and almost the only difference between the 10th generation Intel Core HEDT processors and their predecessors. Other innovations include increased Turbo Boost frequencies, four additional PCI-E 3.0 lanes, and intra-silicon patches from Meltdown and Specter. By the way, even this turned out to be enough for the first batches of new CPUs to sell like hot cakes.
Simultaneously with Intel Cascade Lake-X, the high-end AMD Socket sTRX4 platform entered the market. It consists of motherboards with a TRX40 logic set and 3rd generation Ryzen Threadripper processors based on the 7nm Zen 2 microarchitecture.
Today, the range of CPU data is represented by only two models: Ryzen Threadripper 3960X (24 cores / 48 threads) and Ryzen Threadripper 3970X (32 cores / 64 threads). If you look under the lid of the new processors, you can find four 7nm chipsets with Zen 2 cores and a large 12nm I/O die inherited from the 2nd generation EPYC. However, this does not prevent AMD and its partners from teasing those who are not indifferent to the hardware with the 64-core Ryzen Threadripper 3990X, which is expected to be released in January.
Scalped AMD Ryzen Threadripper 3960X. Der8auer Photos
Along with the third generation of Ryzen Threadripper chips, the arrival of the PCI Express 4.0 interface to the HEDT segment took place. The controller integrated into the new CPUs provides 64 PCI-E 4.0 lanes, of which eight are used to communicate with the AMD TRX40 hub. It is worth noting that, like the massive X570 for the AM4 platform, the TRX40 chipset needs active cooling.
Unfortunately, backward compatibility with TR4 motherboards and first generation Ryzen Threadripper processors has been lost. AMD explained this by the need to increase the interface bandwidth between the CPU and the logic set, as well as the insufficient scalability of the old platform, hinting at the upcoming release of the 64-core flagship.
In the next part, we will find out what was remarkable about 2019 in the graphics card market.
Overview of the main events of 2019. Video cards
Overview of the main events of 2019. Memory, overclocking and new trends