Looking through the forum thread, not only our resource dedicated to processors, you can find very different overclocking results achieved on seemingly identical CPUs. Of course, the spread is due to the efficiency of different cooling systems, advanced overclocking capabilities of motherboards, and user skills. But what about the «silicon lottery»? — you ask. After all, there were times when only a certain batch was considered successful and enthusiasts were looking for it. And that’s how big the scatter of results for different processors is, and we decided to check it.
In the study, we have a small group of so-called APUs from AMD. We will have the opportunity to test the conventional wisdom that any Ryzen is capable of conquering the 4 GHz mark and it will be very interesting to observe the behavior of the integrated Vega graphics core. Not all users need to overclock to the limit at the limit, conditionally safe, voltage. Many enthusiasts prefer low voltages with moderate heating of the processor and motherboard power system elements. There are fans of silent or conditionally silent systems who strive for silence and low temperatures when using passive or semi-passive cooling. Someone wants to build a versatile multimedia PC in a small package and is looking at solutions based on Raven Ridge as a basis. In connection with all of the above, we will check the ability of processors to operate at voltages from conditionally low to conditionally high for computing and graphics cores.
Getting to know the victim
So, the test subjects are five AMD Ryzen 5 2400G. We have already reviewed this model before and this review can be considered an addition to the previous material, so there is no point in giving the main characteristics of the processor, I will only note the standard frequencies of 3600 and 1250 MHz for the CPU and GPU, respectively.
First of all, we determine the date and serial number of the subjects. This can be found in the second line of the marking of the form AN 1802SUT, where AN is the batch number, 1802 is the year and week. In our case, our CPU is released on the 2nd week of 2018. Further, SUT stand for «Suzhou» (Suzhou, China — processor assembly plant) + «Texas» (GlobalFoundries semiconductor production in Austin, Texas, USA).
Data for five processors are summarized in a table.
We have one copy from the first batches under consideration, released last year, and four APUs that rolled off the assembly line this year. Of particular interest is a trio produced in the same week and having close serial numbers.
The search for maximum values for the graphics and processor parts was carried out according to a similar technique.
The grafical part
The base frequency is fixed in UEFI at 100 MHz, the RAM frequency is 2933 MHz with delays of the form 14-14-14 with a voltage of 1.35 V. The processor frequency for all modes is the same and is 3.6 GHz. Due to the peculiarity of the formation of SOC and iGPU voltages with the active graphic part, their voltages were equal during the test. The search for stable frequencies took place in the range from 1 to 1.2 V. in steps of 0.1 V. The stability of the system was checked by passing the 3DMark06 benchmark.
The base frequency is fixed in UEFI at 100 MHz, the RAM frequency is 2933 with delays of the form 16-16-16-32, 1.35 V. The SOC voltage is fixed at 1.05 V. Search for the maximum frequency in the voltage range from 1.0 up to 1.5 V in steps of 0.1 V. The accuracy of setting and voltage drop was checked using a multimeter. By enumeration of the multiplier, the maximum frequency of passing 10 cycles of the LinX v0.7.0 (2017.0.21) test for AMD with a task size of 28326 is searched.
This, as in the graphical part, does not mean that this is enough to ensure 100% stability for 24/7 mode, and it is possible that if the test is run for several hours or days, the system will show instability, but a certain idea of the capabilities of the test participants can be folded.
For graphics tests, the stand was based on the MSI B350I PRO AC motherboard, which showed fantastic overclocking capabilities for integrated graphics cores.
To test the stability of the computing part of Raven Ridge, the choice fell on the ASUS ROG CROSSHAIR VI HERO, which has a redundant power subsystem for quad-core processors.
To reduce the impact on the results of the cooling system, a liquid cooling system loop was used.
The test bench configuration is as follows:
- processors: AMD Ryzen 5 2400G;
- cooling: LSS EK-Supremacy EVO — Full Copper, Laing DDC + Top Phobya Nickel Edition, EK-CoolStream RAD XTC (420);
- motherboard #1: MSI MSI B350I PRO AC (AMD B350);
- motherboard #2: ASUS ROG CROSSHAIR VI HERO (AMD X370);
- memory: PATRIOT PV416G373C7K (2×8 GB, DDR4-2933, CL16-16-16-32 / 14-14-14-28);
- system disk: Samsung 850 EVO (120 GB, SATA 6Gb/s);
- graphics adapter: MSI N750TI-2GD5TLP (GeForce GTX 750 Ti 2GB);
- power supply: Raidmax Cobra RX-850AE-B (850 W);
- monitor: LG E2242 (1920×1080, 22″);
- operating system: Windows 10 Pro x64;
- Radeon driver: AMD Adrenalin Edition 18.5.1;
- GeForce driver: NVIDIA GeForce 388.13.
iGPU test results
|Serial number||Marking||1.0 V (0.986)||1.1 V (1.096)||1.2V (1.194-1.208)|
All samples demonstrated the ability to pass the test at 1183 MHz, which is only 5% less than the nominal frequency. At a voltage of 1.1 V, a pair of leaders stood out, capable of operating at a frequency of 1319 MHz, which is the same 5% higher than the nominal value. They also obeyed an impressive frequency of 1625 MHz at 1.2 V. This is higher than the nominal value by a significant 30%. The spread of results within 100 MHz, which is 8% of the nominal frequency. It is amazing that the best and worst results are obtained on processors with a difference in serial number of a couple of units.
CPU test results
|Serial number||Marking||1.0 V (0.986)||1.1 V (1.096)||1.2V (1.194-1.208)||1.3 V (1.308)||1.4 V (1.392)||1.5 V (1.492)|
The minimum available frequency for passing the test at a voltage of 1 V turned out to be 3200 MHz, which is 11% lower than the base frequency. One copy at this value mastered 3325, he also conquered 3925 MHz at 1.3 V — plus 9%. For the other three processors, this frequency succumbed only when the voltage was increased by 0.1 V. In the champions, we see all the same samples that showed themselves well when testing the graphic part. They can do the coveted 4000 MHz at 1.4 V and 4050 MHz at 1.5 V.
The maximum difference in the achieved frequencies is about 125 MHz, at relatively low voltages: 1.0, 1.1, 1.2 V. The minimum spread was noted at 1.4 V and amounted to 75 MHz.
All instances have demonstrated the ability to work with low voltages at frequencies that correspond to the standard ones for some other processors of the Zen architecture, for example, the base frequency of the Ryzen 7 1700 is declared at 3000 MHz. No dependence on the production date was found, rather, on the contrary, processors released in the same week and having a difference in serial number by two units showed the worst and best results. For statistical information, the sample is too small, but apart from sports overclocking, the selection of copies seems of little interest, at a reasonable voltage of 1.4 V, the difference between 3925 and 4000 MHz will be barely noticeable in the vast majority of applications. The same is true for the graphics part, when the limiting factor in the performance of Vega 11 is the memory bandwidth. I would like to note that the CPUs, which demonstrated energy efficiency in the processor part, turned out to be effective in the graphic part as well. This statement is also true in reverse.
4.0 GHz turned out to be tough for not all instances at a reasonable voltage. The transition from 1.4 to 1.5 V computing modules gives a completely meager increase in frequency. On the contrary, when the voltage rises to 1.2 V, the graphics part receives a significant increase in frequency.
You should not consider the frequency results obtained in the course of this study as absolutely stable for 24/7 mode under any load. Especially with the use of less productive cooling systems, especially a boxed cooler.
Instead of an epilogue
There are some peculiarities in working with memory. RAM frequencies up to 3200 MHz are stable in a variety of synthetic memory tests, higher frequencies up to 3600 MHz sometimes passed complex memory stability tests without problems. But after rebooting or re-launching the test, they began to throw errors at the very start, or demonstrated instability when starting the 3D mode. The maximum memory frequency at which it was possible to pass CPU-Z validation is 3733 MHz.