[ad_1]
Welcome to 2021 or 2020 New Game+. How about dusting off the ancient forgotten practice of processor binning? As many as ten copies of the Ryzen 3 3100, well, how can you not indulge then ?!
Let’s start by listing the participants, this is not a whole party:
- BF 2019PGS 9HZ8038500010;
- BF 2019PGS 9HZ8038S00002;
- BF 2019PGS 9HZ8038S00006;
- BF 2019PGS 9HZ8038S00009;
- BF 2019PGS 9HZ8038S00011;
- BF 2019PGS 4515368S00001;
- BF 2019PGS 4515368S00002;
- BF 2019PGS 4515368S00030;
- BF 2022PGS 9HZ8070S00001;
- BF 2022PGS 9HZ8070S00002.
Three different batches. Five, three and two representatives, respectively. Eight produced during Week 19 (May 4 – May 10) 2020 and two during Week 22 (May 25 – May 31) 2020. Absolutely all subjects are from Penang, Malaysia.
Test stand:
- cooling No. 1: CBO circuit (HAILEA HC-300A chiller, two Hydor Seltz 1200 pumps, 15 l water tank, EK Supremacy Nickel + Acetal water block);
- cooling #2: AMD Wraith Prism;
- thermal interface: Arctic Cooling MX-4;
- motherboard: MSI B450I Gaming Plus AC (UEFI A.C0, AGESA 1.0.0.6);
- memory: HyperX Predator HX441C19PB3/8 x2 (2×8 GB, 4266 MHz, CL19-26-26-45-2T);
- video card: GIGABYTE GeForce GT 1030 (GV-N1030D5-2GL);
- system storage: Kingston KC2000 250GB (SKC2000M8250G, M.2 NVMe);
- power supply: Rosewill Hercules-1600S (1600 W).
Temperature indicators in the case of cooling No. 1 were ignored. The water temperature is set at 20 degrees. The room temperature is 20 degrees. Binning criteria: FCLK, frequency potential, voltage curve. Ask why IMC was not tested? Yes, because it is pointless, everyone mastered the DDR4-5000 mode of operation, as well as the aggressive DDR4-3600 CL14 profile, so the power of the built-in memory controller is enough for any user’s whims … from a rational sports mode to eye-pleasing corn megahertz. And this is great, because if the RAM does not work stably at high frequencies / low timings, the memory itself is to blame, at worst the unfortunate tweaker itself, but not IMC. And this significantly saves time in achieving the best results.
Someone is already stretching their fingers to proudly write the post “Tests are not tests! But what about automation? Where is the turbo test in the factory? Relax and keep reading. This processor has already been tightened as much as possible so that it does not create much competition for the neighboring Ryzen 3 3300X, and also does not spoil sales of the remnants of stocks of predecessors, like the Ryzen 3 1200. Even on a factory cooler (AMD Wraith Stealth), any subject will statically keep the frequency of 3 .9 GHz without a hint of drawdown, as long as temperatures do not exceed 70 degrees.
As soon as the temperature becomes 70.1 and higher, the frequencies begin to drop to the 3.7 GHz line in steps of 25 MHz, and these same 3.7 GHz are kept in the range of 80–94.9 degrees.
When the processor warms up to 95 degrees, it starts throttling mercilessly, the temperature does not rise, but after 10 minutes of the Cinebench R23 rehearsal run, the processor slowed down to 1.4 GHz.
The author with the facial expression of Matthew McConaughey in that famous scene from “Interstellar” could not stand it, and stopped bullying after 11 minutes. Of course, each motherboard manufacturer can redefine the principle of operation of the acceleration and throttling algorithms, but in this particular case there was nothing to binning.
To find the maximum FCLK frequency, the processor frequency was fixed at 4000 MHz for all cores with SMT enabled, the SOC voltage was 1.1 V, and the CPU was 1.5 V. The RAM frequency was twice the FCLK frequency (for ease of understanding and settings), JEDEC timings.
During experiments with voltages, the XMP profile of DDR4-3600 memory was used, FCLK was 1800 MHz, UCLK was 1800 MHz, SOC voltage was 1.1 V, BCLK corresponded to 100 MHz. The voltage step is 0.025 V. The frequency limit was found by applying voltage up to 1.6 V. The criterion for confirming the stability of operation at a certain frequency was the successful passage of: HWBOT x265 2.3 Benchmark – 4k, y-cruncher 0.7.8.9507 – Pi-1b preset, and also LinX AMD 0.7.0 – Problem size = 36585 with 10 repetitions.
Let’s start with FCLK. Five mastered 1900 MHz, four 1866 MHz, and even one 1933 MHz. As further testing will show, the FCLK limit is a lottery. The 9HZ8038S00009 instance, which broke through 1933 MHz, was by no means the most successful in terms of frequency potential and voltage curve, like the quite typical 4515368S00002, it took only 1866 MHz, when other processors of its level took 1900 MHz.
Only one thing looks suspicious – all the test subjects of the 4515368S000 (xx) series conquered only 1866 MHz FCLK.
Let’s move on to the frequency limit.
All, without exception, are able to conquer the 4.4 GHz line. More than half (seven) of the copies master 4.5 GHz. But 4.6 GHz was taken by only one subject – 9HZ8038S00011. He has the best stress curve, so this was not a surprise.
The previously tested Ryzen 3 3100 (BF 2014SUT 9JB0078P00342) did not take part in this event, because it has been working correctly in someone’s PC for more than six months, but it also mastered 4.6 GHz on all cores, even at a lower voltage (1, 5V porting 1.55V) than our current favorite. Do you know what that means? That’s right, you need to go even further, filling it with liquid nitrogen! But this will be dessert, but for now let’s evaluate the stress curve:
Let’s take the line of 4.4 GHz for analysis, 1.275 V is enough for someone, and 1.475 V is enough for someone. However, as practice shows, it is better not to apply voltage above 1.5 V for 24/7/365 operation on an ongoing basis (hello, high-speed degradation), some motherboard manufacturers do not recommend voltages above 1.45 V, but whether it is necessary or not, let everyone decides for himself.
Freeze! For better contrast, let’s take as a basis the 5798.65 MHz success of the previously mentioned Ryzen 3 3100 (9JB0078P00342) as a benchmark, another participant in that test in the face of the Ryzen 3 3300X (9JB7896P00300), the best example today (9HZ8038S00011), as well as a typical “potato chip” (9HZ8038S00009). The meanings of XOC abbreviations have already been written more than once, so let’s see how the reader remembers them.
And then the author was covered with melancholy and despondency. The best of the sample was able to squeeze out the “pathetic” (by XOC standards, of course) 5250 MHz frequencies for all cores, although it was able to do it at 1.5 V voltage, but it was not possible to move further than 5300, even though you applied 1.65 V to it . And it seems that CB is also not bad, it starts to suffer from nonsense only in the region of -150 degrees. CBB around -100 degrees. As it turned out, the best instance in the air turned into a “potato” on liquid nitrogen, for example, 5.3 GHz – this was the milestone of the previously tested Ryzen 3 3300X.
For obvious reasons, the bench stand was immediately disassembled, dried, and prepared for the next test subject – an unremarkable Ryzen 3 3100 with traditional cooling. And again, sadness and melancholy, CB is already at the turn of -120 degrees, and CBB is somewhere around 70-80 degrees, this is some kind of APU … Not only can the temperature not be lowered, but also 1.625 V 5125 MHz, and at 5150 MHz the system freezes tightly.
Now again a short summary:
- Summer tested Ryzen 3 3100 (BF 2014SUT 9JB0078P00342) = 5.8 GHz.
- Summer tested Ryzen 3 3300X (BF 2013SUS 9JB7896P00300) = 5.3 GHz.
- Best of Sample Ryzen 3 3100 (BF 2019PGS 9HZ8038S00011) = 5.25 GHz.
- Random sampled Ryzen 3 3100 (BF 2019PGS 9HZ8038S00009) = 5.125 GHz.
In general, you have to quote the popular meme: “This is a fiasco, bro!”.
As a conclusion to this whole event, I would like to once again remind Intel that it was stupid to take the unlocked multiplier from the Core i3 series, ingo then there would be binning Core i3-10300k after all, as the binning showed, the Ryzen 3 3100 can easily be heated up to the 4.4 GHz line (well, or 4.2-4.3 GHz, if the user is so worried about the safety of the processor, as for the resource of MLC flash memory chips in SSD), on in which the CPU will be an objectively powerful product among the 4-core solutions of early 2021, even taking into account the overpayment during the current shortage. The RAM controller will not put spokes in the wheels in the study of the potential of any modules. Unless with FCLK it will be “as lucky” regardless of the success of the instance. And whatever one may say, whatever one may say, even the weakest instance will respond with an improvement in speed performance, if it is “engaged”. I hope all readers have remembered the existence of a twilight sorcery called “binning”!
[ad_2]