Often, owners of AMD Phenom II processors with an unlocked multiplier overclock their chips using only the previously voiced instrument, bypassing the system bus, which means they ignore the speed of the L3 cache and the built-in RAM controller. And given the fact that the L3 cache and memory controller of these CPUs operate at a fixed frequency of 2000 MHz, overclocking by raising the core frequency (only by a multiplier) will bring very few dividends.
In this article, we will conduct a “correct” study of the overclocking potential of an old quad-core AMD Phenom II X4 955 BE processor.
Welcome to the sixth article in the Retro Overclocking series.
processor
Test sample marked HDZ955FBK4DGM CACAC, the processor die made in Germany and packaged in Malaysia.
The codename for the AMD Phenom II X4 955 Black Edition chip is Deneb (K10.5 architecture). Under the heat spreader cover of this CPU is one 45nm quad-core monolithic Deneb die in 6M modification.
The revision of our test copy is C3, which is good news since C3 is the best possible revision (Here we are talking about the overclocking potential and the power-hungry of the processor. In addition to revision C3, there is also C2 – which is based on extremely unsuccessful and hot chips.
The Phenom II X4 955 processor carries four cores with a base frequency of 3200 MHz (multiplier 16, bus frequency 200 MHz, frequency of the integrated memory/L3 cache controller, and 2000 MHz HT bus). The chip has 512KB of L2 cache per core and a shared 6MB L3 cache buffer. The base voltage of the processor cores is set at 1.400 volts. TDP 955 “Fena” does not exceed 125 watts.
Due to AMD’s loyal policy, the Phenom II X4 955 chip supports both DDR2 and DDR3 memory. This is because this CPU is designed to be installed both in outdated motherboards with the AM2 socket, as well as in relatively current motherboards with the AM3/AM3+ socket. Thus, the RAM controller of this processor, which, by the way, is located inside the chip, is banally obliged to work with both types of RAM. Which it successfully does.
In our case, the processor was installed in a motherboard with an AM3+ socket, which means that a pair of Phenom II X4 955 will be DDR3 memory.
Of the obvious disadvantages of the X4 955 processor, it is worth noting the lack of support for the SSE4.1/SSE4.2 instruction, the minimum required at the moment. And believe me, this is a very serious disadvantage, because due to this feature, this chip may simply not run the program or game you need (and in the case of the Phenom II X4 955, this is really extremely offensive, since the processor could still please its owners)/
Test setup:
- Processors — Phenom II X4 955;
- Cooling — Cooler Master Hyper 212 Black Edition (RR-212S-20PK-R1);
- RAM for AM3+ — 2 x 8GB HyperX Genesis Na’Vi Edition (KHX16C9C2K2/8) with a total volume of 16GB for the final overclocking with an increase in voltage, as well as 2 x 2GB Hynix HMT325U7EFR8C-RD strips for the rest of the study;
- Motherboard AM3+ — ASUS M5A97 LE R2.0;
- Video card — KFA2 GeForce RTX 2060 SUPER 8GB (~1950/14000MHz, Power Limit 112%);
- SSD — KINGSTON 120GB SA400S37120G (Windows 10/Apps);
- Hardware — Seagate 2TB ST2000DM008-2FR102 (Games);
- Power supply — Chieftec GPS-1250C.
overclocking Phenom II X4 955 Black Edition
To begin with, let’s figure out what our Phenom II X4 955 chip is capable of with the base voltages of cores, RAM, memory controller, and L3 cache.
If you do not understand technical processes, processor generations, and so on, you should not despair. In order to find out what your CPU model is capable of in theory, you just need to find out the code name of its core (in our case, it is Deneb), after which it is banal to “google” the top model based on the same kernel.
The fastest processor codenamed Deneb is Phenom II X4 980. Its frequency is 3700 MHz, which means that in theory (I repeat, in theory), our chip is also capable of stably operating at 3.7 GHz. Let’s check it out. But we will do this not in a simple way, by increasing the multiplier, but by overclocking the system bus:
In order to overclock our “Fenom” to 3700MHz, you need to set the system bus frequency to 232MHz and make sure that the installed RAM can work at 1600MHz (if not, it’s okay, just lower its multiplier to 1066MHz and you’ll end up with ~ 1333MHz).
Reboot, test in LinX, and voila, the processor is stable at 3724MHz:
Let me remind you that I did not raise any of the supply voltages (they were rigidly fixed in the BIOS in manual mode in order to prevent the board from raising them automatically). In fact, such overclocking should not harm the processor, motherboard, or memory, even in the long run. However, remember that any overclocking will void the warranty for this or that product. Thus, you do all the manipulations with your iron solely at your own peril and risk!
Further, without raising any of the supply voltages, we increase the system bus frequency by 5MHz, reboot, test the chip in LinX for stability, and if the test succeeds, add 5MHz to the bus frequency and test it again.
The result of the study of the overclocking potential of our Phenom II X4 955 without raising the voltages was the frequency 3890 MHz:
Overclocking settings for AMD Phenom II X4 955 up to 3890MHz in BIOS:
- CPU Core voltage – 1.400V (+0.000 to the base);
- System bus frequency (BCLK) – 235MHz;
- L3 cache/memory controller voltage (CPU-NB voltage) – 1.000V (+0.000 to the base);
- RAM multiplier in BIOS – 1333MHz (taking into account overclocking, the effective memory frequency was 1570MHz);
- RAM voltage (DDR3 voltage) – 1.500V (+0.000 to the base).
In general, this is far from a record result, since the results of other enthusiasts can be found on the network, where the frequency of the cores without raising the voltage turns out to be close to 4000 MHz. But there’s nothing we can do about it because overclocking is primarily a lottery.
Now let’s move on to the most interesting thing, namely, overclocking with raising several supply voltages.
All that needs to be done here is also to increase the system bus frequency by 5MHz, reboot, test the chip in LinX for stability, and if the test succeeds, add 5MHz to the bus frequency and test it again. If the application gives an error, or, the system freezes – first you need to increase the voltage on the cores by 0.05 volts and reboot again. If the system is again unstable, then there is not enough CPU-NB (memory controller/L3-cache) voltage. Raise it by 0.05 volts, reboot, and test again.
However, I think, for obvious reasons, this is not the most efficient way to find out the maximum stable frequency.
After your processor becomes unstable at a certain frequency without raising the voltage, reduce the core multiplier by 2-3 points in order to eliminate their influence on the final indicator and check if the processor is stable after that. Thus, you will be able to calculate the stable frequency of the most important processor module of the AM3 platform, namely CPU-NB.
So, in my case, lowering the multiplier by 2 points, I got the final frequency of 3400 MHz cores. That, in fact, is significantly lower than what I achieved at the base voltage of the cores so that their frequency will not become a limiter in the future.
After that, to get the final CPU-NB frequency around 2500MHz, I set the BCLK FSB frequency to 250MHz. At the same time, the speed of the cores was 3625 MHz, which means that I still had a margin for their frequency without increasing the voltage. But the system, as expected, did not start with such settings.
Next, I raised the CPU-NB voltage to 1,200 volts. The system started, but after a few minutes it fell into the blue “screen of death”. It was only at a CPU-NB voltage of 1.350 volts that I was able to successfully complete an hour-long benchmark in LinX.
Well, we figured out the frequency of the memory controller and L3 cache. Now it’s time to switch to overclocking the cores.
After CPU-NB overclocking, the space for overclocking the rest of the processor’s execution units is quite narrowed. In fact, all that remains is to raise the previously reduced CPU Core multiplier while simultaneously increasing the vCore voltage after each failed LinX test attempt.
The first stable point was a frequency of 4000 MHz with a core voltage of 1.475 volts, which is not bad at all. However, I did not stop there and decided to raise the voltage of the cores to the maximum, regulated by AMD itself, at 1.550 volts.
The result of overclocking the AMD Phenom II X4 955 chip with an increase in supply voltages was the frequency of 4138 MHz:
Overclocking settings for AMD Phenom II X4 955 up to 4138MHz in BIOS:
- CPU Core voltage — 1.550V (+0.150 to the base);
- System bus frequency (BCLK) — 250MHz;
- CPU multiplier – 16.5;
- L3 cache/memory controller voltage (CPU-NB voltage) – 1.350V (+0.350 to the base);
- RAM multiplier in BIOS – 1333MHz (taking into account overclocking, the effective memory frequency was 1672MHz);
- DDR3 voltage — 1.600V (+0.100 to the base).
- NB HT voltage – 1.250V (+0.150 to the base)
It should be clarified that with such a voltage of cores and CPU-NB, the power consumption of the processor exceeded 200 watts. If you want to repeat my experiment, you’d better take care of decent cooling of the chip and the space around the socket (power supply circuits). And remember: you do all the manipulations with your hardware solely at your own peril and risk!
P.S.:
Roughly, a full-fledged performance test of the AMD Phenom II X4 955 processor will be released by the end of this spring (if nothing changes in my plans). However, you already have the opportunity to evaluate the overclocking gain of this chip in the article “Testing 16 Budget Processors in The Witcher 3: Wild Hunt [February 2021]“, which presents the results of the Phenom II X4 955 CPU at 3200MHz and 4138MHz.
Thank you for your attention!
If this article turned out to be useful to you and you would like to see more similar tests on the UmTale Lab resource, then please support our site on Patreon! The main goal of fundraising is to expand the component park and improve the quality of testing: replacing the bench drive with a larger SSD, buying a capture card to reduce the impact of gameplay recording using ShadowPlay on the final results, and so on).