For almost a year, processors of the Zen microarchitecture have been on the market, on which AMD has high hopes. Five years of development should have finally brought the company out of the position in which it found itself after the release of Intel’s Core family solutions. A timid attempt to rectify the situation back in 2011 by the Bulldozer architecture did not lead to anything — FX processors were highly dependent on software support: showing excellent performance in multi-threaded applications, they were losing ground in single-threaded, i.e. in games. But, despite another failure on the processor front, the chipmaker decided not to stop and set about developing completely new solutions.
The new products promised a redesigned memory controller and support for the DDR4 standard, an increase in the specific performance of each core, a change in cluster multithreading technology (CMT) to simultaneous multithreading (SMT) — such steps were supposed to increase the speed of processors by 40% compared to their predecessors. And the fans began to look forward to another miracle. And, to tell the truth, their expectations were justified. Zen microarchitecture processors were able to occupy almost all market niches, including high-level and server segments, where AMD’s position has long been shaken. This also had an impact on the confrontation with her opponent, who urgently released a new generation of mid-range CPUs and expanded their lineup by adding six-core products, which had never happened before.
In addition to good performance, without exception, all Ryzen processors have an unlocked multiplier, which makes them more attractive in the eyes of enthusiasts than some competing Intel solutions. It is enough to select a suitable multiplier with a couple of movements in the firmware of the motherboard and run the chip at a frequency higher than the nominal one, thereby increasing the system performance. To do the same with the opponent’s products, you need to choose certain processors belonging to the «K» series, the prices of which are much higher than simple solutions, and the choice is limited to several models. And here AMD is clearly out of competition. But is it so easy to overclock Ryzen and what will it give against the background of products of the opposite camp, we will try to find out in a series of our reviews.
So, in the range of available desktop CPUs of the Ryzen family, there are nine models: three Ryzen 7 series, four Ryzen 5 and two Ryzen 3. The older ones differ in the number of cores, which is eight pieces with multithreading support, the middle ones have 4/6 cores, also with SMT , and the younger line is content with only four cores without the ability to process multiple threads per clock. In addition, Ryzen 7 and older Ryzen 5 models have 16 MB of L3 cache on board, while the rest have only 8 MB. More detailed characteristics of the processors are presented in the table:
|Processor||Ryzen 7 1800X||Ryzen 7 1700X||Ryzen 7 1700||Ryzen 5 1600X||Ryzen 5 1600||Ryzen 5 1500X||Ryzen 5 1400||Ryzen 3 1300X||Ryzen 3 1200|
|Core||Summit Ridge||Summit Ridge||Summit Ridge||Summit Ridge||Summit Ridge||Summit Ridge||Summit Ridge||Summit Ridge||Summit Ridge|
|Process technology, nm||14||14||14||14||14||14||14||14||14|
|Number of cores (threads)||8 (16)||8 (16)||8 (16)||6 (12)||6 (12)||4 (8)||4 (8)||4||4|
|Rated frequency, GHz||3,6||3,4||3||3,6||3,2||3,5||3,2||3,5||3,1|
|Boost-mode frequency, GGc||4||3,8||3,7||4,0||3,6||3,7||3,4||3,7||3,4|
|Multiplier unlocked up||+||+||+||+||+||+||+||+||+|
|L1 cache, KB||8 x (32 + 64)||8 x (32 + 64)||8 x (32 + 64)||6 x (32 + 64)||6 x (32 + 64)||4 x (32 + 64)||4 x (32 + 64)||4 x (32 + 64)||4 x (32 + 64)|
|L2 cache, KB||8 x 512||8 x 512||8 x 512||6 x 512||6 x 512||4 x 512||4 x 512||4 x 512||4 x 512|
|L3 cache, MB||16||16||16||16||16||16||8||8||8|
|Supported memory||DDR4-2667 DDR4-2400||DDR4-2667
|DDR4-2667 DDR4-2400||DDR4-2667 DDR4-2400||DDR4-2667 DDR4-2400||DDR4-2667 DDR4-2400||DDR4-2667 DDR4-2400||DDR4-2667 DDR4-2400||DDR4-2667 DDR4-2400|
|Recommended cost, $||349*||309*||299||219*||189||174||169||129||109|
*— without cooling system
We will overclock the Ryzen 7 1800X, Ryzen 5 1600, Ryzen 5 1400, Ryzen 3 1300X and Ryzen 3 1200 processors as representatives of solutions with different numbers of threads, and compare them with each other in terms of performance. And the test will be supplemented by an old man in the face of AMD FX-6100.
- Motherboard #1: MSI B350 Gaming Pro Carbon (UEFI 1.50, AGESA: SummitPI-AM4 18.104.22.168);
- motherboard #2: ASUS Crosshair V Formula (UEFI 1703, AGESA: OrochiPI V22.214.171.124);
- Indicator: Noctua NH-U12P + Nanoxia FX12-2000;
- thermal interface: Noctua NT-H1;
- Memory #1: G.Skill TridentZ F4-3200C16D-16GTZB (2x8GB, 3200MHz, 16-18-18-38-2T, 1.35V);
- memory #2: HyperX Predator HX324C11PB3K4/32 (4×8 GB, 2400 MHz, 11-13-14-32-2T, 1.65 V, only two modules out of four participated);
- video card: MSI GTX 780Ti Gaming 3G (GeForce GTX 780Ti);
- drive: Silicon Power Slim S55 (240 GB, SATA 6 Gb / s, AHCI mode);
- power supply: SilverStone SST-ST65F-PT (650 W);
- operating system: Windows 10 Pro x64 (10.0.16299.192);
- drivers: AMD APP SDK 3.0, AMD Chipset Drivers 17.30, GeForce 381.65 (126.96.36.19965), PhysX 9.17.0329.
All OS updates available from Windows Update have been installed. Third-party anti-virus products were not involved, no fine-tuning of the system was made, the size of the paging file was determined by the system itself.
The following applications were used as tests:
- AIDA64 5.95.4522 (Cache & Memory benchmark, BenchDLL 4.3.770-x64);
- Super PI 1.5 XS;
- wPrime 2.10;
- x265 HD Benchmark;
- MAXON CINEBENCH R15;
- POV-Ray 3.7.0;
- LuxMark v3.1;
- Futuremark 3DMark 13 (2.4.4180);
- DiRT 3 Complete Edition (188.8.131.52);
- Hitman: Absolution (1.0.447.0).
Overclocking Ryzen 7 1800X. Description of the general methodology
The older model of the family has become a participant in several of our comparative tests and in each of them it overclocked. However, the question of the overclocking procedure itself was then touched upon superficially, today I would like to pay more attention to it. In addition, any of the processors based on the Zen architecture will be overclocked according to a similar principle.
We used an instance with a UA 1705SUT batch (hereinafter, for comparison, a snapshot of the FX-6100 processor is shown):
First, let’s remember how the representative of the Ryzen 7 line behaves with the initial system settings. In idle, the frequency is 2.2 GHz when the voltage is less than 0.5 V:
For older motherboards where the X370 hub is soldered, a special CPU operation mode has been prepared, when the frequency can be maximized due to XFR technology, however, I saw it in action on our test MSI B350 Gaming Pro Carbon, which uses not X370 at all, but B350. The load was created by a single-threaded Super PI 1.5 XS, while the voltage increased to 1.438 V (according to the processor sensor, which in AIDA64 is called CPU VDD). The frequency grew up to 4.1 GHz:
With a full-fledged, multi-threaded load, it was no longer so high — growth stopped at 3.7 GHz, the voltage did not exceed 1.219 V. The test task was the “1024M” scenario from wPrime 2.10:
As you can see, the voltage level of 1.45 V is included in the operating scheme by the manufacturer, therefore, during overclocking, a slight excess should not significantly «harm» the processor. An important aspect that requires attention is the behavior of the voltage regulator on the board. There is a separate review for our bench motherboard, where this issue can be studied in detail, I only state a noticeable decrease in CPU VDD (according to AIDA64) against the background of the values set in UEFI and formed on the VRM by the board itself. In this case, I will perceive the “useful” voltage as critical, and not the one located “on the outskirts” of the processor.
CPU overclocking can be done in more than one way, but for a quick search for frequency limits, AMD’s proprietary software called Ryzen Master is quite suitable. In my opinion, it would be appropriate to start testing with a frequency of 3700 MHz and a voltage of 1.4 V. For some users, even this level may seem redundant, here everyone has the right to make their own decision, but today, without resorting to compromises, we will deal with maximum overclocking ( for air cooling).
Running wPrime 2.10 with the 1024M profile created a load on both the processor itself and the VRM system on the motherboard. The step of changing the processor frequency is 25 MHz, which is rather small. You can increase the final frequency every few seconds by moving between the cell with the new value and the Apply button. Relatively stable, for this section of experiments, the system will be considered by me as long as it retains its performance. As a rule, a failure in its operation leads to a reboot, so you need to remember the mark (new, when it happened, or old, when there were no problems yet).
For our board, you should also take into account the difference between the set and operating voltage on the processor, the following picture was typical for the starting test mode: with maximum LLC, 1.416 V formed on the VRM, and only 1.356 V reached the processor. For NB, I chose 0.9 V, it’s a little over the stock mark.
The full load will be if you do not forget to set the number of active threads characteristic of the CPU used in the load utility:
So, when the first mark is found, you can increase the tension and move on. In our sample, the “stable” result of the first cycle turned out to be 4.1 GHz. Let’s make an increase of 0.1 V, thereby reaching the theoretical 1.5 V. Let’s check how the system reacted to this increase by running the same wPrime. We see that 1.528 V is formed on the VRM, and 1.45 V is valid for the processor (I indicate the maximum marks as psychologically important in this part of the measurements):
We continue, with an increased supply voltage, we managed to move up to 4150 MHz. Is 50 MHz worth it? Let everyone decide for himself. For the sake of curiosity, I decided to go even further, slightly exceeding the 1.5 V threshold (for UEFI), adding another 0.025 V to them. Again, we check the current levels. For the stabilizer, it grew to 1.552 V, while the processor had only 1.475 V at its disposal:
Continuing the tests, it turned out to fix the performance of a PC with an effective CPU frequency of 4.2 GHz. However, this mark is far from stable, but only an outlet for benchmarking enthusiasts. However, the frequency potential of our specimen has been studied, we can safely stop at a frequent 4.1 GHz. To facilitate the perception of information, the figures obtained are summarized in a small table:
|Model||Voltage in UEFI, V||CPU VDD (active), V||Frequency before failure wPrime, MHz|
|Ryzen 7 1800X||1,4||≤ 1,356||4100|
|Ryzen 7 1800X||1,5||≤ 1,45||4150|
|Ryzen 7 1800X||1,525||≤ 1,475||4200|
Our processor had no problems when tested at 4.1 GHz and 1.45 V, but it was under light load. To consolidate, I brought in a more “heavy” one — x265 HD Benchmark.
Short tests allowed me to reduce the level to 1.4875 V set in UEFI, on the CPU the fixed mark was 1.438 V. We are not talking about CPU overheating even with an inexpensive cooler, because the temperature did not exceed 80 degrees.
The next step is overclocking the RAM. The question is acute, as for the AM4 platform, but today — in the coming 2018 — almost all chips (from various manufacturers) can already cross the 3000 MHz mark, however, the undisputed leader of recommendations is a set containing Samsung B-die chips. For our test, the kit will be based on a cheaper and very common E-die revision, it is dual-ranked, which is also considered appropriate for a Ryzen CPU. First you need to check the performance of the system with the XMP profile, the set of delays is determined by the manufacturer for a reason, and the frequency formula can often be workable. Further development of events has many scenarios, but I will describe our case. The system was perfectly capable of running XMP circuitry at 1.35V on the modules. Further experiments (to increase the frequency) were very difficult. Having tested all the firmware that have marks for DRAM greater than 3200 MHz (these are only later versions of UEFI, in the early versions “3200” is the maximum), I settled on UEFI 1.50. With it, our memory ran at 3333 MHz, but required stabilization, while with the latest version, at the time of testing (1.60), there were already problems even getting into the UEFI environment. The voltage on the modules did not exceed 1.5 V, therefore, perhaps, this was precisely the limiting factor (here we are faced with another psychological barrier, but it is also important that this level is the maximum for the motherboard model used). The result of experiments with the main group of delays was the following scheme, which we used for all Ryzen processors in this review (here I confirm its performance for each processor that took part in our tests):
The search for memory frequency limits took place with the initial settings for the processor, and now it’s time to combine them in order to achieve the maximum increase in PC performance. The high frequency of DRAM was not slow to affect the greater load of the processor, so its operation scheme had to be adjusted, namely, the frequency dropped to 4050 MHz at the same voltage level. But he worked properly at 4.1 GHz, but with a slow memory.
Here I had to be a bit cunning, because all our test applications worked without issues, but others, specializing specifically in identifying instabilities in work, could sooner or later declare an error in the calculations. That is why in motherboard tests I use a frequency equal to 4025 MHz, where there are no such problems, but for the consolidated test, the 4050 MHz mark looks more solid. In addition, the RAM overclocked to 3333 MHz turned the “cold start” stage of the system into a real roulette: the PC could turn on the first time, or the second (double start), the third, or even report errors in the settings, then you had to reload profile with them. It would be appropriate to store it not only in the UEFI environment, but somewhere else on removable media. That is, the selected mode has a slight touch of the benchmarking approach, but, with a strong desire, it could be stabilized by choosing a larger effective value for the CPU and adjusting the minor memory delays. Everything will depend on the desires of the owner of this system, but we want to collect one-time information about its performance. The list of changes made to the UEFI settings looked like this:
It should be noted that MSI motherboards do not allow adjusting Pstates, that is, the processor frequency is increased for all cases of PC operation, so it was not necessary to use special Power Plans in the Windows environment. The voltage also behaves, there is only one mode available — the exact fixation of the value, on other motherboards there may be several options.