Question Overclocking AMD Ryzen 5500 on Gigabyte motherboard
Question Overclocking AMD Ryzen 5500 on Gigabyte motherboard
Hello, I'm just starting with Ryzen family CPUs and planning to boost their performance. My current setup includes more CPU power needed for 25 virtual machines, which is a big requirement. I’ve already tried overclocking AMD FX processors in UEFI but felt overwhelmed by all the options. I’m confident I can learn, but time constraints mean I might not have enough to finish everything in a couple of days.
I’d appreciate any advice or tutorials on overclocking for the AM4 platform, especially for the 5th generation and Gigabyte B550M DS3H motherboard. If anyone has similar setups and experience, please share how they approached it.
MOTHERBOARD: Gigabyte B550M DS3H
What BIOS version are you using for your motherboard?
RAM: 2 x 32 GB Kingston KF3200C16D4/32GX and 2 x 8 GB Kingston KHX3333C16D4/8GX
Consider using a single RAM kit instead of mixing them. Ideally, opt for a DDR4-3600MHz dual channel kit. If you choose one kit—like the 2x32GB—make sure to enable X.M.P or enter the sticker information in BIOS (D.O.C.P.).
Standard fixed frequency overclocking for Ryzen CPUs tends to be ineffective due to the boost algorithm's limitations in reaching very high clocks, which also necessitates a higher Vcore (up to 1.5V) for stability. The protection features of the boost algorithm become irrelevant when using fixed frequency, forcing either continuous high voltage use or settling for a clock below the maximum boost value.
There are two approaches to enhance performance. One involves applying PBO2 and Curve Optimizer to individually undervolt each core while maintaining compatibility with the boost algorithm. The other focuses on employing superior cooling solutions that maintain processor temperatures under 80°C during peak loads, such as during Cinebench rendering. Combining these methods yields optimal results.
If you still wish to pursue fixed all-core overclocking, aim for a VCore around 1.25-1.3V maximum and keep the LLC setting at medium levels. Identify the highest stable clock frequency for your specific workload. Closer to 1.2V is safer for heavy tasks; exceeding 1.3V with only slightly stable clocks may lead to rapid degradation during intensive operations like video rendering, unless you manage temperatures in the 60s or low 70s. This is unlikely without advanced sub-ambient cooling.
Note: CPUs from the 5000 series are considered the fourth generation Ryzen, while the 5th generation Ryzen corresponds to the 7000 series running on AM5 platforms.
The bios version is the newest one. I also forgot to mention that the motherboard Rev is 1.4. The X.M.P profile was already launched with 3200 MhZ frequency.
I found some free time to have fun with the stuff. While I was searching for some tutorials I found this guy's video
View: https://www.youtube.com/watch?v=RD51wAbmPvo
. I just followed the same settings as his ( 1.352 Voltage for core, 45x multiplier and discabled core boost option) and started benchmarking and testing the stability. I need to mention that in the beggining I disconnected two 8 GB ram sticks and left these with bigger capacity.
The tests went successfully. At the beggining I launched the performance test from Passmark. The results are in attached links.
Zrzut-ekranu-2023-11-03-133728 hosted at ImgBB
Image Zrzut-ekranu-2023-11-03-133728 hosted in ImgBB
ibb.co
Zrzut-ekranu-2023-11-03-133814 hosted at ImgBB
Image Zrzut-ekranu-2023-11-03-133814 hosted in ImgBB
ibb.co
Zrzut-ekranu-2023-11-03-134402 hosted at ImgBB
Image Zrzut-ekranu-2023-11-03-134402 hosted in ImgBB
ibb.co
It's strange that these memory sticks performed worse. However I made it to achieve 10% performance boost for CPU compared to the Passmark database and 8% compared to CPU-Z benchmark. The disturbing things were the temperatures while benchmarking. In CPU-Z Stress test for 10 minutes the Package and the Cores temperatures didn't reach much higher than 72 Celsius. Whereas while CPU benchmark in Passmark the Core sensor showed 82 Celsius for a few seconds.
After the benchmarking I went to stress testes. I started with only CPU test via OCCT Program.
Zrzut-ekranu-2023-11-03-135907 hosted at ImgBB
Image Zrzut-ekranu-2023-11-03-135907 hosted in ImgBB
ibb.co
Zrzut-ekranu-2023-11-03-142042 hosted at ImgBB
Image Zrzut-ekranu-2023-11-03-142042 hosted in ImgBB
ibb.co
Zrzut-ekranu-2023-11-03-144033 hosted at ImgBB
Image Zrzut-ekranu-2023-11-03-144033 hosted in ImgBB
ibb.co
Everything went well without any errors and the CPU didn't throothled. I noticed that the CPU VDD Voltage was mostly at 1.3V instead of 1.35V ( maybe VRM overloaded? ) and sometimes the Cores sensor showed +10 Celsiuses just for a second, then it dropped down ( maybe sensor error? ). The average temperatures were fine but surely this was a stop border for further voltage/clock increasing.
Then I tried to launch a Power stress test but unfortunately my hardware shutted down immediatelly
🙁
The reasons was the PSU which couldn't handle +5V so I had to remove some USB devices and disconnect DVD drive. I runned it again and temperatures went up to 90 degrees just in few seconds that my fans didn't have enough time to gain speed so I gave up the test.
Zrzut-ekranu-2023-11-03-150742 hosted at ImgBB
Image Zrzut-ekranu-2023-11-03-150742 hosted in ImgBB
ibb.co
I also runned one more test to check RAM performance for mixed sticks. The CPU was 1,5% slower and RAM was around 8% slower than after the previous OC.
Zrzut-ekranu-2023-11-03-151743 hosted at ImgBB
Image Zrzut-ekranu-2023-11-03-151743 hosted in ImgBB
ibb.co
Anyway I tested the hardware in the field I'm going to need and it handles the temperatures well with 80~90% CPU usage but I'm surely gonna try to undervolt it in the future.
The Curve optimizer is essential for Ryzen 5000-7000 series systems. While you can adjust it manually, you'll likely need to spend hours fine-tuning. This advice comes from Ryzem Master. It isn't flawless, but let the software handle the work. Run per-core and all-core tests, noting the outcomes. Adjust BIOS settings according to the recommendations provided. If your results are consistently 20 for all cores, aim for around 18. Modern processors behave differently, so abandon traditional overclocking methods. Maintaining proper temperature and power management is crucial for optimal performance.
Well, you're trying out fixed OC which is fine but do also try out PBO2 and Curve Optimizer is strongly suggested before settling on anything for long-term use. There are several youtube videos on how it's done. It doesn't have to take hours of tweaking (there's a process) to find a lower voltage for each core but it certainly can if you want to find absolute limits.
But whatever you do don't compare based on clocks alone; use benchmark results. Passmark is probably quite good but it's not as widely used as something much more simple: Cinebench. It's fairly short to run and yet has highly repeatable results. But what's more it's not a purely synthetic bench but much more reflective of real-world performance since it uses a production rendering application to render a scene. Also, it's widely used so you can compare results with others.
If you find bottom-line performance is the same or very little difference between fixed clock OC and PBO2/CO then you'll have to decide if it's worth burning up your processor using the fixed or going for longer life letting the boost algorithm do it's thing to keep it within FIT projections.
And when letting the CPU use it's boost algorithm (PBO2/Curve Optimzer) it's going to be very dynamic with constant and frequent clock changes on a per-core basis. It will only boost a core when it's needed so you might not think clock is high with PBO2/CO but it still performs very well in BM's.
And don't think of it as "throttling" when clocks are low; that occurrs only when something very bad happens such as when the CPU overheats significantly (around 105C if I'm not wrong) or the VRM overheats (usually around 115C on the FET's). You have to forget about anything you've known before as this is nothing like Intel CPU's, nor previous AMD such as Bulldozer or Phenom series.
And lastly: don't search for a stable memory overclock at the same time you're looking for a stable CPU OC. Do one at a time. This part of good OC technique is really no different from any generation or make of CPU, of course.