One of the most basic principles that AMD has followed over the past years is: «Let’s add more cores.» It is under this motto that both the Ryzen desktop processors, along with their Threadripper HEDT counterparts, and the EPYC family of server chips lead their triumphant march. For clarity, we can recall in more detail AMD’s mass-segment offerings that have appeared on the market over the past years: in the first generation of Ryzen against quad-core Kaby Lake, it put up processors with eight cores, then eight-core Ryzen fought with six-core Coffee Lake, and this year to massacre with the eight-core Coffee Lake Refresh, the twelve-core Ryzen 9 3900X was thrown. But this is the story about how AMD was imbued with the idea of screeaThe number of core users is far from over, because the company decided to make an even more monstrous processor with sixteen cores the crown of the Ryzen 3000 family.
However, despite the continued desire to increase the number of cores at every opportunity, AMD’s current tactics are still noticeably different from how they acted before. If in 2017 and 2018 the additional cores in Ryzen acted as a kind of compensation for their lower specific performance and frequency than the competitor, then with the Zen 2 architecture, the «red» made up for the gap in the IPC indicator and began to claim to redraw the entire processor market. As it turned out four months ago, when the 12-core Ryzen 9 3900X came out, users are willing to pay significantly more than $ 500 typical for this market segment for mass AMD processors with exceptional characteristics. You don’t need to go far for examples: when problems with the production of the Ryzen 9 3900X turned into a severe shortage and the price of 12-cores jumped to $900 at the peak, this didn’t stop buyers at all — they continued to methodically sweep them off the shelves.
Now, AMD wants to finally fix the mass platform Socket AM4 in a higher position, proving that it is not forbidden to cost one and a half times more than the price that has been established for years. The new flagship Ryzen 9 3950X will prove this, in which the number of processing cores has increased by another step — up to 16. Such a processor is estimated by the manufacturer at $ 749, allowing the Socket AM4 platform to make a shameless foray into HEDT territory, and, judging by what was already known about the Ryzen 9 3950X until today, not only to conduct reconnaissance in battle, but also to securely settle there. Confidence in the viability of the novelty is given by its characteristics. With the help of TSMC’s state-of-the-art 7nm process and innovative chiplet design, AMD has been able to ensure that increasing the number of cores in the Ryzen 9 3950X does not reduce operating frequencies or reduce power efficiency. As a result, the 16-core flagship should adequately prove itself in the entire range of existing tasks, and in resource-intensive multi-threaded loads, one can expect a truly breakthrough performance for the mass segment from it.
In other words, the Ryzen 9 3950X, to which this review is dedicated, claims to be a kind of uber processor for mainstream platforms, which are characterized by the use of dual-channel memory and have limitations in the number of supported PCI Express lanes. But this is where the main intrigue of today’s test lies: when getting acquainted with the characteristics of the Ryzen 9 3950X, many may have the feeling that for a conventional commonly used platform this is “already too much”, because it turns out that, along with the heroic 16-core, the manufacturer allows you to install and dual-core processor plugs starting at $50. Will the promising Ryzen 9 3950X succeed in spreading its wings, or will its potential be unrealized due to limitations in memory bandwidth and power supply? We will talk about this further.
⇡#Ryzen 9 3950X in detail
In terms of topology, the new 16-core Ryzen 9 3950X is very similar to the Ryzen 9 3900X we reviewed earlier. Before us is the second mass AMD processor, which is based not on two, but on three semiconductor crystals at once, called chiplets in modern terminology. One of these chips is a universal 12nm cIOD chiplet responsible for I / O functions and containing a memory controller, a PCI Express controller, and SoC elements. The other two are 7nm CCD chiplets, which contain eight computing cores. All this is connected into a single whole through the Infinity Fabric bus, which connects each of the CCD chipsets to the cIOD chip. At the same time, CCDs do not have a mutual connection with each other, but this does not entail any negative consequences, since the entire Infinity Fabric logic is located in the I / O chiplet, which equalizes all cores in rights. In other words, the Ryzen 9 3950X, unlike previous generation multi-core Threadripper processors, does not have any NUMA nodes and, from a logical point of view, has an absolutely monolithic design in which the delays in working with memory and in mutual data exchange are the same for all cores.
To learn more about how AMD is pursuing a multi-chip approach to Ryzen flagship processors, check out our review of the 12-core Ryzen 9 3900X. The 16-core Ryzen 9 3950X is built in exactly the same way, but uses fully functional CCD dies that don’t have locked cores.
It would seem that everything is very simple, and the only thing that is completely incomprehensible is why AMD did not introduce the Ryzen 9 3950X earlier, simultaneously with other Socket AM4 processors. However, this delay has a completely logical explanation. The fact is that, having received the progressive Zen 2 microarchitecture, AMD has set itself an ambitious goal not only to bring a processor with 16 cores to the mass segment, but to do it as effectively as possible. To do this, according to the original plan, the Ryzen 9 3950X had to receive not only support for the maximum number of threads for the desktop segment, but also high clock frequencies, and so that all this at the same time did not lead to a noticeable increase in heat dissipation and power consumption.
The multi-chip design solves a variety of problems inherent in large monolithic chips, due to the fact that small-area chips are both easier and cheaper to produce. But it does not offer direct methods for neutralizing the growth in power consumption and heat dissipation when additional cores are added to the processor. Therefore, in the end, it took AMD some more time for the Ryzen 9 3950X to be able to acquire the desired characteristics: 16 cores, the maximum turbo frequencies in the entire 3rd generation Ryzen model range and an ordinary thermal package for mass CPUs. This is achieved by a very simple in description, but rather dreary in implementation method — the choice for such processors of the highest quality semiconductor crystals.
In fact, the same approach has already been used in the Ryzen 9 3900X, and using this processor as an example, one could see how difficult it is to select CCD crystals for three-chip processors, even if a quarter of the cores are then blocked in them. AMD was unable to keep up with demand and supply 12 cores for an entire quarter, leaving the Ryzen 9 3900X in short supply for a long time. Choosing the right chips for the Ryzen 9 3950X is even more difficult: two full-fledged eight-core full-featured CCD chipsets, along with a cIOD die, should fit into a 105-watt thermal package, while providing approximately the same frequencies around 4.0 GHz at full load as other processors of the Ryzen 3000 family.
The pause taken by AMD was supposed to allow the company, together with a manufacturing partner, the TSMC semiconductor forge, to optimize the technical process and create some stock of suitable crystals so that the Ryzen 9 3950X turns out to be a real, not a paper novelty. But we will be able to check how effective all this preparation turned out to be only after November 25 — that’s when sales of the massive 16-core AMD will start. For now, we can only evaluate the performance and nuances of the Ryzen 9 3950X, but we cannot vouch for its wide availability on sale at the price that AMD promised for it — $ 749.
Passport characteristics of the 16-core Ryzen 9 3950X in comparison with other third-generation Ryzen are as follows.
|Cores/Threads||Base frequency, MHz||Turbo frequency, MHz||L3 cache, MB||TDP, W||Chiplets||Price|
|Ryzen 9 3950X||16/32||3.5||4.7||64||105||2×CCD + I/O||$749|
|Ryzen 9 3900X||12/24||3.8||4.6||64||105||2×CCD + I/O||$499|
|Ryzen 7 3800X||8/16||3.9||4.5||32||105||CCD+I/O||$399|
|Ryzen 7 3700X||8/16||3.6||4.4||32||65||CCD+I/O||$329|
|Ryzen 5 3600X||6/12||3.8||4.4||32||95||CCD+I/O||$249|
|Ryzen 5 3600||6/12||3.6||4.2||32||65||CCD+I/O||$199|
|Ryzen 5 3500X||6/6||3.6||4.1||32||65||CCD+I/O||~$160|
|Ryzen 5 3500||6/6||3.6||4.1||sixteen||65||CCD+I/O||~$150|
The base frequency of the Ryzen 9 3950X is set at 3.5 GHz, which is slightly lower than other processors in the lineup, but the frequency in turbo mode can reach 4.7 GHz, and this is the maximum auto-overclocking for any third-generation Ryzen. At the same time, the processor should not go beyond the TDP of 105 W, that is, it should not be more gluttonous and not hotter than the older models for Socket AM4 that were available until now.
However, there is no particular reason to worry about the heat dissipation of the Ryzen 9 3950X. AMD’s thermal and power requirements are easily met through Precision Boost 2.0 technology, which sets the Ryzen 9 3950X to the same power and current limits that were previously set for the Ryzen 9 3900X or even the Ryzen 7 3800X. The PPT (Package Power Tracking) value, which limits the actual consumption of the CPU, is set to 142 W, the maximum current consumption during long-term loads TDC (Thermal Design Current) is limited to 95 A, and during short-term bursts its growth to 140 A is allowed — this limit is laid down in constant EDC (Electrical Design Current). If in real conditions the processor tries to go beyond these limits, its frequency is forcibly reduced, which just determines its frequency formula in reality.
To evaluate how the Ryzen 9 3950X can be clocked in practice, we ran a traditional experiment — we ran the popular Cinebench R20 rendering test with different thread counts and recorded the frequency that was observed in each case. The result is the following relationship.
Unfortunately, in reality, the frequencies of the Ryzen 9 3950X do not look as great as one might expect. Firstly, this processor does not deliver the 4.7 GHz frequency promised by the specifications under single-threaded load. AMD has been struggling with this problem for a long time, but for the new 16-core it pops up again. Despite the fact that a BIOS based on the latest AGESA 184.108.40.206B libraries is already available for the ASRock X570 Taichi test motherboard, the maximum recorded frequency under a single-threaded load in Cinebench R20 was 4.625 GHz, and the average was 4.6 GHz, that is 100 MHz below the declared value.
Secondly, with an increase in the number of active threads (and the number of cores involved), this frequency goes down quite sharply. For example, with a load on two cores, it drops to 4.475 GHz, and on four — already up to 4.325 GHz. Obviously, AMD and its partners will still have to spend some time fixing the annoying problem with turbo frequencies, but now with new sixteen-cores.
It is also impossible not to notice that the full load in Cinebench R20 on all cores drives the processor frequency noticeably below the 4 GHz bar. Thus, the Ryzen 9 3950X is the first member of the 3rd generation Ryzen family that falls short of 4.0 GHz with all cores active. Our instance of the processor had a frequency of 3.875 GHz, while, for example, the 12-core Ryzen 9 3900X could run at 4.05 GHz when rendering in Cinebench R20. The decrease in frequency occurs in this case due to the fact that the Ryzen 9 3950X begins to run into the established power consumption limits. And, obviously, with a more complex load than rendering, the situation can easily get worse. For example, in the Prime95 29.8 stress test with AVX instructions, the frequency of the Ryzen 9 3950X drops even below the nominal value — up to 3.3 GHz.
The underlying reasons for the low frequency of the Ryzen 9 3950X under multi-threaded workloads lie in the way AMD approached the selection of crystals for such processors. Unfortunately, high-quality CCDs capable of operating at high frequency at moderate voltage and with controlled heat dissipation are extremely rare. Therefore, in order to ensure the release of the Ryzen 9 3950X in some noticeable quantities, AMD decided to use two types of CCD chips of fundamentally different quality in them. The principle is simple: the first CCD crystal is responsible for the high frequencies in turbo mode, but it can also have high leakage currents and higher power consumption and heat dissipation. The second chiplet should offer good energy efficiency: it may not pick up high frequencies, but its efficiency should definitely be better than the first one.
The viability of this scheme in the case of the Ryzen 9 3950X is determined by the fact that the cores in the third generation Ryzen processors are not only physically heterogeneous, but also logically. At the production stage, they are prioritized, and the operating system task scheduler always tries to load the more pliable cores first (in Windows 10, for example, this has been working since the May 2019 Update). As a result, AMD really manages to win by dividing tasks between chiplets. In particular, it is enough to guarantee the performance of the turbo mode only for selected successful chipset cores with good frequency scalability. The rest of the cores never reach a high frequency and are involved in calculations only with multi-threaded loads, when the processor frequency is much lower.
All of the above is very easy to illustrate. The following table shows the maximum turbo frequencies achievable by the cores of our Ryzen 9 3950X instance with the voltages that they request to operate in this mode. These data were obtained in a single-threaded Cinebench R20 test with a forced direction of the load on different cores. Along with monitoring frequencies and voltages, we also measured the power consumption of running cores.
It is clearly seen here that the first chiplet (CCD0) is selective in terms of frequency, significantly surpassing the second chiplet (CCD1) in terms of performance scaling capabilities. So, the cores of the first CCD chiplet can operate at frequencies above 4.475 GHz, without requiring voltages in excess of 1.466 V. The maximum frequencies of the cores of the second chiplet are 100-200 MHz worse, but despite this, to operate at maximum frequencies, they need more high voltages that can reach 1.481 V. At the same time, a characteristic feature of the second CCD is that even at the highest possible frequency, its cores consume less electricity under load due to lower leakage currents.
Differences in energy efficiency of CCD crystals are visible even when the processor is fully loaded. In this case, all the cores run at the same frequency and with the same voltage, but show different power consumption. Here’s how it looks when multithreaded in Cinebench R20.
Our copy of the Ryzen 9 3950X consumption of the second chiplet was not much belowthan the first, but the general principle is still clear: the first CCD chiplet in the processor is high-frequency, and the second is slower, but with reduced energy appetites.
We will not dwell on a detailed analysis of the rest of the characteristics of the Ryzen 9 3950X: they are exactly the same as those of the Ryzen 9 3900X. So, the processor has a 64 MB L3 cache, which consists of four 16 MB parts distributed over four CCX (Core Complex), and also offers 24 PCI Express 4.0 lanes, 16 of which are usually allocated to the video card, 4 to NVMe -drive and 4 more — for connection with a set of system logic. All this can be seen in more detail in the CPU-Z screenshot.
Although the Ryzen 9 3950X is a processor for the mainstream Socket AM4 platform, AMD clearly states in all its marketing materials that it has no competitors in the LGA1151 ecosystem. The Ryzen 9 3950X’s $749 price tag sets it apart from offerings for the LGA2066 HEDT platform. In the official presentation, the new 16-core processor is called a competitor to the 12-core Core i9-9920X, which, according to the official price list, costs $1,189. X), the cost of which will be reduced. And soon, not a 12-core, but a 14-core Core i9-10940X with a recommended price of $784 will compete against the Ryzen 9 3950X.
How it will turn out with performance, we’ll see, but it’s worth having in mind and the fact that the Ryzen 9 3950X, unlike processors for LGA2066, works with dual-channel rather than quad-channel DDR4 memory, and also offers half as many PCI Express lanes (but the fourth generation).
Such dramatic changes that have taken place this year with Socket AM4 systems are staggering. But it is doubly surprising that by doubling the number of cores in the older Ryzen series processors, AMD continues to maintain end-to-end compatibility throughout the entire ecosystem. In other words, the 16-core Ryzen 9 3950X can run in Socket AM4 motherboards released this year, last year, and even the year before. True, with some reservations.
AMD itself recommends motherboards based on the X570 chipset for the Ryzen 9 3950X, because they tend to have more powerful power circuitry, but this is not a necessary condition. It is only necessary to use the latest firmware for the 16-core processor to work in Socket AM4 boards: correct and stable operation of the Ryzen 9 3950X is guaranteed only with those BIOSes that are built based on the AGESA 220.127.116.11B (18.104.22.168 Patch B) libraries. For the most common boards based on the X570, X470 and B450 chipsets, the necessary updates should be released before the end of this month, at least in the form of beta versions. At the same time, in these updates, the operation of the turbo mode will once again be corrected, the maximum practical frequencies in which, with each update, are systematically approaching their passport values.
The second important point concerns the fact that the Ryzen 9 3950X is the first member of the family to ship without a bundled cooler, even in the boxed version. AMD simply did not have standard options for cooling systems in its assortment that could provide sufficient heat dissipation for a 16-core processor. Despite the progressive 7nm chipsets and die-soldered heat dissipation cover, the Ryzen 9 3950X needs more efficient cooling than AMD’s older Wraith PRISM cooler can offer. Therefore, AMD decided to leave the issue of choosing a cooling system to the user, giving only a vague recommendation that the Ryzen 9 3950X requires liquid cooling with a 280mm radiator. In fairness, it should be noted that AMD itself tested the Ryzen 9 3950X with a Noctua NH-D15S air cooler for its marketing materials. Therefore, we can hope that high-performance supercoolers for the Ryzen 9 3950X are also quite suitable.
But at the same time, you need to keep in mind that the automatic overclocking algorithm Precision Boost Override takes into account, among other things, the temperature regime of the processor. Therefore, more efficient cooling can be useful to get the best out of the Ryzen 9 3950X.
⇡ # Overclocking
Since the Ryzen 9 3950X is made up of two diverse CCD chips, one of which is not about high frequencies at all, overclocking this processor by choosing a single fixed frequency for all cores is a strange idea and, most likely, doomed to failure. Nevertheless, for the sake of curiosity, we still decided to see how high the frequency of this processor can be raised if we act head-on.
The best processors of the Ryzen 3000 family that visited our laboratory overclocked synchronously across all cores up to 4.3 GHz. The sixteen-core Ryzen 9 3950X did not break this record for at least two reasons. First, because of the high heat dissipation. Three adjacent semiconductor chips will always heat up more than two, and therefore, despite the use of the NZXT Kraken X72 high-performance liquid cooling system in the test platform, high temperatures became a serious obstacle. Secondly, do not forget about the second CCD chip, which in the Ryzen 9 3950X is selected with an eye to lower heat dissipation, which in some cases is converted into a noticeably worse frequency potential.
Ultimately, the maximum frequency at which the Ryzen 9 3950X was able to work stably with a high load on all cores was 4.1 GHz. At a supply voltage of 1.2 V, the processor passed an hour-long test at this frequency in Prime95 29.8 with active AVX2 instructions.
It is worth noting that approximately the same results were obtained when overclocking the Ryzen 9 3900X, from which we can conclude that three-chip processors do not tolerate a simultaneous increase in the frequency of all cores at once. Indeed, the resulting frequency with the simplest approach to overclocking turns out to be significantly lower than the maximum frequency in turbo mode, giving a clear hint that this is not necessary.
The correct method is to use the Precision Boost Overdrive function, which allows you to remove all the inherent restrictions on processor consumption and its supply currents. In this case, the processor, left to itself, will independently overclock to those milestones that will be determined by the quality of its silicon crystal and the potential of the cooling system used. The huge advantage of overclocking with Precision Boost Overdrive is that this feature leaves the opportunistic auto-tuning algorithms active and the processor keeps the frequency variability depending on the load. And therefore, the CPU frequencies at low load after overclocking will be at least no worse than in the nominal mode.
However, when Precision Boost Overdrive was enabled, the test Ryzen 9 3950X proved to be a typical representative of the series. AMD itself overclocked its 16-core processor almost to the limit, leaving very little space for user experiments. With a single-threaded load in Cinebench R20, the processor increased the frequency by only 25 MHz — up to 4.65 GHz.
And in a multi-threaded load, the frequency of the Ryzen 9 3950X with activated Precision Boost Overdrive increased by 125 MHz to 4.0 GHz.
Naturally, the meaningfulness of such overclocking raises serious doubts: in the best case, it can increase performance by 1-3%. But this is not the final result. We must not forget that while the Ryzen 9 3950X does not develop the prescribed frequencies even in the nominal mode, therefore, as AMD and partners make corrections to the BIOS of motherboards, the effectiveness of Precision Boost Overdrive may also increase.
⇡#Description of test systems and testing methods
By and large, the Ryzen 9 3950X is such an unusual processor that there are no normal rivals for it in the same weight category. For the mass segment, it is too heavy, and in order to be called a HEDT processor, it at least lacks four-channel memory. As a result, we will have to compare the Ryzen 9 3950X with two different product categories at the same time: flagships for the LGA1151 and Socket AM4 platforms, as well as with existing 16-core processors for the LGA2066 and Socket TR4 platforms. In addition, the comparison also had to include the 12-core Core i9-9920X, which, for some reasons that are not entirely clear to us, is considered by AMD to be the main competitor of the Ryzen 9 3950X.
In the end, the list of components involved in testing turned out to be as follows:
- AMD Ryzen 9 3950X (Matisse, 16 cores + SMT, 3.5-4.6 GHz, 64 MB L3);
- AMD Ryzen 9 3900X (Matisse, 12 cores + SMT, 3.8-4.6 GHz, 64 MB L3);
- AMD Ryzen 7 3800X (Matisse, 8 cores + SMT, 3.9-4.5 GHz, 32 MB L3);
- AMD Ryzen Threadripper 2950X (Colfax, 16 cores + SMT, 3.5-4.4 GHz, 32 MB L3);
- Intel Core i9-9960X (Skylake-X, 16 cores + HT, 3.1-4.4 GHz, 22 MB L3);
- Intel Core i9-9920X (Skylake-X, 12 cores + HT, 3.5-4.4 GHz, 19.25 MB L3);
- Intel Core i9-9900K (Coffee Lake Refresh, 8 cores + HT, 3.6-5.0 GHz, 16 MB L3).
- CPU cooler: NZXT Kraken X72.
- ASRock X570 Taichi (Socket AM4, AMD X570);
- ASRock Z390 Taichi (LGA1151v2, Intel Z390);
- ASUS ROG Strix X299-E Gaming II (LGA2066, Intel X299);
- MSI MEG X399 Creation (Socket TR4, AMD X399).
- 2 x 8 GB DDR4-3600 SDRAM, 16-16-16-36 (G.Skill Trident Z RGB F4-3600C16D-16GTZR);
- 4 × 8 GB DDR4-3600 SDRAM, 16-16-16-36 (G.Skill Trident Z RGB F4-360016Q-32GTZR).
- Video Card: NVIDIA GeForce RTX 2080 Ti (TU102, 1350/14000 MHz, 11 GB GDDR6 352-bit).
- Disk subsystem: Samsung 960 EVO Plus 2TB (MZ-V7S2T0BW).
- Power Supply: Thermaltake Toughpower DPS G RGB 1000W Titanium (80 Plus Titanium, 1000W).
All compared processors were tested with the settings accepted by the motherboard manufacturers «by default». This means that for Intel platforms, the power consumption limits indicated in the specifications are ignored and the maximum possible frequencies are used in order to obtain maximum performance. It is worth emphasizing that the vast majority of users operate processors in this mode, since enabling limits on heat dissipation and power consumption requires special BIOS settings.
All compared processors were tested with memory operating in DDR4-3600 mode with XMP timings, with the exception of the Ryzen Threadripper 2950X, which does not work in this mode with the kit we use due to memory controller limitations. For this processor, a slower DDR4-3200 mode was used with timings of 14-14-14-32.
You also need to be aware that the processors participating in the tests have slightly different prices. For reference, in the following table, we provide either the manufacturer’s recommended or the real (if it differs significantly from the official) cost of the CPU.
|CPU||Number of cores/threads||Price|
|Intel Core i9-9960X||16/32||$1,475 (retail)|
|Intel Core i9-9920X||12/24||$1,050 (retail)|
|AMD Ryzen 9 3950X||16/32||$749|
|AMD Ryzen Threadripper 2950X||16/32||$650 (retail)|
|AMD Ryzen 9 3900X||12/24||$499|
|Intel Core i9-9900K||8/16||$488|
|AMD Ryzen 7 3800X||8/16||$399|
Testing was performed on the Microsoft Windows 10 Pro (v1903) Build 18362.175 operating system using the following driver set:
- AMD Chipset Driver 22.214.171.1243;
- Intel Chipset Driver 10.1.1.45;
- Intel Management Engine Interface Driver 126.96.36.1997;
- NVIDIA GeForce 441.08 Driver.
Description of the tools used to measure computing performance:
- Futuremark PCMark 10 Professional Edition 2.0.2144 — testing in Essentials scenarios (typical work of the average user: launching applications, surfing the Internet, video conferencing), Productivity (office work with a word processor and spreadsheets), Digital Content Creation (digital content creation: editing photographs, non-linear video editing, rendering and visualization of 3D models). OpenCL hardware acceleration is disabled.
- 3DMark Professional Edition 2.10.6799 — testing in the Time Spy Extreme 1.0 scene.
- 7-zip 19.00 — archiving speed testing. The time taken by the archiver to compress a directory with various files with a total volume of 3.1 GB is measured. The LZMA2 algorithm and the maximum compression ratio are used.
- Adobe After Effects CC 2019 16.1.1 — Animation video rendering speed test. The time taken by the system to render a pre-prepared video in 1920 × 1080@30fps resolution is measured.
- Adobe Photoshop CC 2019 20.0.6 — Graphics performance testing. The average execution time of the Puget Systems Adobe Photoshop CC Benchmark 18.10 test script, which simulates the typical processing of an image taken by a digital camera, is measured.
- Adobe Photoshop Lightroom Classic CC 8.4.1 — performance testing for batch processing of a series of images in RAW format. The test scenario includes post-processing and export to JPEG at 1920 × 1080 resolution and maximum quality of two hundred 16-megapixel RAW images taken with a Fujifilm X-T1 digital camera.
- Adobe Premiere Pro CC 2019 13.1.5 — performance testing for non-linear video editing. It measures the rendering time to YouTube 4K format of a project containing HDV 2160p30 footage with various effects applied.
- Blender 2.80 — testing the speed of the final rendering in one of the popular free packages for creating three-dimensional graphics. The duration of building the final bmw27 model from Blender Benchmark is measured.
- Corona 1.3 — testing the rendering speed using the renderer of the same name. To measure performance, the standard application Corona 1.3 Benchmark is used.
- Microsoft Visual Studio 2017 (15.9.17) — measuring the compilation time of a large MSVC project — a professional package for creating three-dimensional graphics Blender version 2.79b.
- OBS Studio 24.0.3 — testing the performance and smoothness of game streaming. The following video stream settings are used: x264 encoder, resolution 1080p@60fps, bit rate 6 Mbps, CPU Usage Preset = slow or slower.
- Stockfish 10 — testing the speed of the popular chess engine. The speed of enumeration of variants in the position «1q6/1r2k1p1/4pp1p/1P1b1P2/3Q4/7P/4B1P1/2R3K1 w» is measured.
- V-Ray 4.10.03 — testing the performance of a popular rendering system using the standard V-Ray Benchmark Next application;
- x265 3.2+9 10bpp — testing the speed of video transcoding to the promising H.265/HEVC format. To evaluate performance, we use the original 2160p@24FPS AVC video file with a bitrate of about 42 Mbps.
- XMRig 4.6.2 — mining performance testing using the RandomX algorithm.
- Assassin’s Creed Odyssey. Resolution 1920 × 1080: Graphics Quality = Ultra High. Resolution 3840 × 2160: Graphics Quality = Ultra High.
- Far Cry 5. Resolution 1920 × 1080: Graphics Quality = Ultra, HD Textures = On, Anti-Aliasing = TAA, Motion Blur = On. Resolution 3840 × 2160: Graphics Quality = Ultra, Anti-Aliasing = Off, Motion Blur = On.
- Gears 5. Resolution 1920 × 1080: Default Quality = Ultra. Resolution 3840 × 2160: Default Quality = Ultra.
- Kingdom Come: Deliverance. Resolution 1920 × 1080: Overall Image Quality = Ultra High. Resolution 3840 × 2160: Overall Image Quality = Ultra High.
- Shadow of the Tomb Raider. Resolution 1920×1080: DirectX12, Preset=Highest, Anti-Aliasing=TAA. Resolution 3840 × 2160: DirectX12, Preset = Highest, Anti-Aliasing = Off.
- Total War: Three Kingdoms. Resolution 1920 × 1080: DirectX 12, Quality = Ultra, Unit Size = Extreme. Resolution 3840 × 2160: DirectX 12, Quality = Ultra, Unit Size = Extreme.
- Watch Dogs 2. Resolution 1920 × 1080: Field of View = 70°, Pixel Density = 1.00, Graphics Quality = Ultra, Extra Details = 100%. Resolution 3840 × 2160: Field of View = 70°, Pixel Density = 1.00, Graphics Quality = Ultra, Extra Details = 100%.
- World War Z. Resolution 1920 × 1080: DirectX11, Visual Quality Preset = Ultra. Resolution 3840 × 2160: DirectX11, Visual Quality Preset = Ultra.
In all gaming tests, the results are the average number of frames per second, as well as the 0.01-quantile (first percentile) for FPS values. The use of the 0.01-quantile instead of the minimum FPS is due to the desire to clean up the results from random bursts of performance that were provoked by reasons not directly related to the operation of the main components of the platform.
⇡#Performance in complex tests
Surprisingly, in PCMark 10, AMD’s 16-core new product manages to show leading performance in all three fundamentally different scenarios. From normal home PC use to heavy office applications and even digital content, the Ryzen 9 3950X proves to be the best choice among the tested CPUs. It looks like AMD really managed to create a processor without obvious weaknesses, at least when it comes to complex performance metrics.
PCMark 10 is echoed by another comprehensive benchmark, 3DMark, which evaluates hypothetical gaming performance if the software engine is well-optimized for modern technologies: multithreading and vector instruction sets. The Ryzen 9 3950X is second only to the 16-core Core i9-9960X, which is a HEDT processor in a different weight category. It seems that the performance of the 16-core Zen 2 here is somewhat constrained by the lack of memory bandwidth.
⇡#Performance in applications
At the moment, the Ryzen 9 3950X seems to be a very worthy processor for use in resource-intensive tasks. Its performance is about 15% better than the Ryzen 9 3900X and 66% better than the Core i9-9900K. With this potential, the Ryzen 9 3950X almost always manages to noticeably outperform the 12-core HEDT processor Intel Core i9-9920X and almost compete on equal terms with the 16-core Core i9-9960X.
Almost all concerns about whether AMD has chosen the right environment for its 16-core processor turn out to be groundless. The weakness of the Socket AM4 platform with dual-channel memory hinders the performance of the Ryzen 9 3950X in a relatively small number of cases: when archiving or when processing photos and videos. But in computational tasks, such as rendering, chess, software development or mining, the new AMD flagship is very good. But even if we talk about the few applications where the Ryzen 9 3950X lags behind the Core i9-9920X and Core i9-9960X, there is no reason to be dissatisfied with its performance. Firstly, we are most often talking about a relatively small loss in speed. Secondly, do not forget that the Ryzen 9 3950X is a mainstream processor running in a mainstream platform, and the cost of a configuration based on it is noticeably lower than that of a full-fledged workstation.
In other words, if we talk about the ratio of performance and price, then the Ryzen 9 3950X looks like an absolute leader, at least until AMD and Intel updated their HEDT platforms.
⇡ # Performance in games. Tests in 1080p
The situation with the performance of the Ryzen 9 3950X in games is quite expected. This processor provides about the same frame rate as other older members of the Ryzen 3000 family. All of them are inferior in terms of gaming performance to the leader in this discipline, the Core i9-9900K, but nevertheless, against the background of Intel HEDT processors, they look quite decent.
With the Ryzen 9 3950X, AMD has been able to increase the number of cores without compromising performance when working with a small number of threads and without spoiling anything in the speed of interaction with the cache and system memory. Therefore, those users who want to use the new 16-core AMD processor not only for work, but also for entertainment, are likely to be satisfied, because other options are equally versatile, capable of coping with both resource-intensive workloads and games just as well. , are significantly more expensive.
⇡ # Performance in games. Tests in 2160p
As the resolution increases, the gap between processors is expected to shrink, as the main load begins to fall on the video card. Therefore, the gap in the speed of various processors becomes very small. And this means that gaming performance cannot be the factor that can affect the conclusions about the superiority of certain multi-core processors. The only thing to keep in mind is that the best performance in games is provided by the Core i9-9900K anyway. The addition of cores in Ryzen processors, like those in the Intel Core X family, does not help them achieve the same level of gaming performance. The strength of the Core i9-9900K is in its efficient ring bus and fast memory controller, which neither the Ryzen 3000 nor the Core X have any alternative solutions with similar latency characteristics.
Many gamers choose powerful processors based on the desire to stream. Therefore, we added another game scenario to testing — streaming using the processor. This time, the game Far Cry 5 was used for streaming tests. The popular Open Broadcasting System (OBS) Studio application was responsible for encoding the video stream. It used x264 software encoder. The broadcast was carried out in a resolution of 1920 × 1080 at a frame rate of 60 FPS and a fixed bitrate of 6 Mbps. In the encoding settings, the slow and slowest quality settings profile was selected.
The slow profile is generally pulled out by all the tested processors, however, the eight-core Core i9-9900K and Ryzen 7 3800X do it “on the verge”, without having enough performance margin. So if you really intend to stream games at high quality, it is better to choose processors with 12 cores or, for example, the new Ryzen 9 3950X. It provides a slightly lower fps compared to Intel’s HEDT solutions, but it also costs much less.
As for the slower profile, today’s test participants do not have enough power to provide streaming in this mode. However, by the share of frames delivered to the receiving side, one can judge which of the CPUs have the best streaming potential.
The highest score here is shown by the new 16-core Ryzen 9 3950X, which Not only confidently bypasses all 12-core processors, but it also turns out to be significantly better than the 16-core member of the Threadripper family and the 16-core Core i9-9960X.
In fact, power consumption tests cannot provide any additional knowledge about the properties of the Ryzen 9 3950X. As already mentioned, in this processor AMD artificially limited the electrical parameters — consumption and maximum currents. For this reason, in real conditions, when the load increases, the Ryzen 9 3950X does not start to increase its appetites, but simply resets the clock speed. That is why the processor obtained from AMD boasts, among other things, good energy efficiency.
This can be illustrated by measurement results. The digital power supply of the Thermaltake Toughpower DPS G series, which we use in the test system, allows you to control the consumed and output electrical power, which we use for measurements. The graphs below show the total consumption of systems (without a monitor), measured «after» the power supply, which is the sum of the power consumption of all components involved in the system. The efficiency of the power supply itself in this case is not taken into account.
The situation when the Ryzen 9 3950X tries not to go beyond the limits set for it limitsconsumption, sacrificing clock speed and performance, we see in the last graph, because with a high AVX2 load, its frequency drops to 3.3 GHz. But if we talk about more realistic work scenarios, such as rendering, then in terms of its energy appetites, the Ryzen 9 3950X still turns out to be more economical than Intel processors with more than eight cores. In fact, we can say that the new 16-core AMD thermal and electrical characteristics are more similar to the eight-core Core i9-9900K, and not to any multi-core HEDT processors.
By and large, in the conclusions about the new 16-core Ryzen 9 3950X, one could repeat everything that we said about the 12-core Ryzen 9 3900X, but in superlatives. The fact is that back in July, when AMD decided to launch the first three-chip Ryzen 3000 on the market, it no longer had real competitors. The older LGA1151 processor Core i9-9900K could not fully compete with the Ryzen 9 3900X due to one and a half times fewer computing cores, and the 12-core Core i9-9920X for the LGA2066 platform was incomparably more expensive.
The new Ryzen 9 3950X has an even more difficult situation with rivals. In fact, it’s a one-of-a-kind product that balances between desktops and workstations, but doesn’t really fall into either category. Sixteen cores for a mass desktop is too much, and the Ryzen 9 3950X cannot become a full-fledged HEDT product, at least due to dual-channel memory and an insufficient number of PCI Express lanes, which limits the possibilities of equipping platforms based on it.
Therefore, it is best to consider the Ryzen 9 3950X as an experiment and a kind of demonstration of the possibilities, in which AMD clearly and convincingly shows all the advantages of its chosen chiplet approach. The building blocks of Zen 2 are eight-core 7nm chips, and if a couple of them fit inside a Socket AM4 package, then why not take advantage of it, AMD thought something like this. And then everything turned out by itself: both the 12-core Ryzen 9 3900X and the 16-core Ryzen 9 3950X. Of course, in order to make a full-featured three-chip processor that would fit into the 105-watt thermal package, AMD had to work hard and figure out how to organize the selection of suitable chips. But in the end, a very inventive solution was found for this task, and now we have to get used to the new reality, in which you can install four times as many cores in a typical consumer computer than two or three years ago.
AMD’s rapid increase in desktop processor power is sure to make the Ryzen 9 3950X something like a trolleybus out of a loaf in the mass consciousness. Indeed, it’s hard to imagine that the average user today would be able to appreciate the ability to equip a regular Socket AM4 system with a 16-core CPU, especially considering the complete uselessness of such a number of cores for gaming applications. But that doesn’t mean the Ryzen 9 3950X doesn’t have market prospects. On the contrary, such a price has been chosen for this processor that among enthusiasts there will surely be those who will be guided by the logic of “why not” or even “there are not many cores”. In the end, AMD offers to get twice the number of productive computing cores compared to the Core i9-9900K, overpaying only one and a half times, and this argument can certainly work for a sufficient number of buyers.
Also, it’s undeniable that the Ryzen 9 3950X is really good at CPU-heavy content and very good at rendering. In terms of performance in such applications, it can be safely compared with the twice as expensive Core i9-9960X, which until recently was one of the flagship processors in the HEDT segment. Now Ryzen 9 3950X allows you to get the same thing in a regular desktop.
To be honest, the Ryzen 9 3950X could have become a much more popular solution if AMD had not promised to introduce the third generation of Ryzen Threadripper in a week. Now, those who really need high performance and uncompromising multithreading will look to the new Socket sTRX4 platform, which, judging by the available data, will set new standards among HEDT systems. All this further complicates the formulation of specific recommendations regarding the Ryzen 9 3950X. But if you feel that 8 or 12 cores are too few for you, and 24 and 32 are expensive and generally redundant, then take a closer look at the Ryzen 9 3950X. This processor is clearly worth the money and certainly will not disappoint with performance.