• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Ryzen 5 3600 FCLK OC confusion

Joined
Oct 9, 2010
Messages
1,381 (0.35/day)
Location
Kolkata, India
System Name Ryzen | Kaby
Processor AMD Ryzen 5 3600 @ 4.5 GHz | Intel Core i5 7600K @ 4.9 GHz
Motherboard ASUS B550 ROG Strix F Gaming | ASUS Maximus VIII Ranger
Cooling Cooler Master ML120L RGB with Kryonaut Extreme | Cooler Master Seidon 120V with Kryonaut
Memory 4*8 GB (32 GB) Kingston HyperX Fury @3266 MHz 16-16-16-35 | 2*8 GB DDR4 @ 3100 MHz CL16
Video Card(s) Galax RTX 2060 Super 8 GB (+170 MHz core, +850 MHz memory) | Zotac GTX 1660Ti AMP 6 GB
Storage Crucial P5 1 TB, Kingston A2000 500 GB, WD10EARS, ST2000DM006 | HyperX Fury 240 GB + 500 GB HDD
Display(s) Dell S2418H, 1080p 75 Hz, HDR | Samsung FHD TV
Case Lianli Lancool 215 | Corsair 100R
Audio Device(s) Realtek S1220 (Supreme FX) | Realtek ALC1150 (Supreme FX)
Power Supply ASUS ROG Strix 750G (80+ Gold) | Cooler Master MWE 650 V2 (80+ Bronze)
Mouse Cooler Master Storm Devastator
Keyboard Cooler Master Storm Devastator
Software Windows 10 Pro x64 |Windows 10 Pro x64
Benchmark Scores Cinebench R20 MT - 4200, ST 537 | R20 MT - 1979, ST - 516
Hello everyone, this is regarding overclocking on my primary system as on specs. As far as I read everywhere, it is recommended that the MCLK and FCLK should be in the 1:1 ratio for best performance. However, I tried playing with the FCLK and have set it at 1900 MHz at stock SOC voltage (anything above that results in failed POST) and for some reason, I have seen a bump in performance across benchmarks especially memory bandwidth intensive benchmarks (verified with AIDA 64). It is also worth noting that my RAMs are quite old from 2015, dual ranked with stock XMP at 2666 MHz. I have been able to overclock it 3266 MHz rock stable at 1.36V.

With the FCLK set to 1900 MHz, the latency creeps up by about 4 ns but at the same time, the memory write speed in AIDA also bumps up to 30400 MB/s from 25599 MB/s. The CPU Photoworxx benchmark which relies on memory bandwidth is also about 2500 points higher.

So my question is why is it that everyone recommends putting the FCLK and MCLK as 1:1? I am confused because my experience suggests otherwise. Any inputs are appreciated.
 
Joined
Feb 6, 2021
Messages
446 (3.46/day)
Location
Germany
Processor AMD Ryzen 9 5950X (and a LOT of other CPUs)
Motherboard MSI B550 Tomahawk
Cooling Noctua NH-D15 Chromax. Black
Memory G.Skill Trident Z Neo 3600 Mhz CL16 (2x16GB)
Video Card(s) Several 6900XTs, 3090s,80s,70s and 60s, almost all Turing and Pascal Cards + a few older ones.
Storage Corsair MP600 1TB, Samsung 970 Evo Plus 500GB, 2x Samsung 870 QVO 4TB
Display(s) Samsung Odyssey G7 32"
Case Fractal Design Meshify S2 White TG / Noctua NF-A14 iPPC-3000RPM Fans
Audio Device(s) Bose Companion Series 2 III, Sennheiser GSP600 running through a SoundblasterX G6
Power Supply bequiet! Dark Power Pro 12 1200w Titanium
Mouse Glorious PC Gaming Race Model O, G903 Hero, G502 Hero SE, Deathadder V2, EC2-A
Keyboard ASUS ROG Strix Scope TKL DLX, Corsair K95, Logitech G815
VR HMD Oculus Rift S
because your scenario is the exact opposite.

normally people run ram higher than the max stable FCLK and then it gets cut in half.
and faster RAM with a slower FCLK that is not 1:1 skips clocks and the overall performs falls off a cliff.
 
Joined
Oct 9, 2010
Messages
1,381 (0.35/day)
Location
Kolkata, India
System Name Ryzen | Kaby
Processor AMD Ryzen 5 3600 @ 4.5 GHz | Intel Core i5 7600K @ 4.9 GHz
Motherboard ASUS B550 ROG Strix F Gaming | ASUS Maximus VIII Ranger
Cooling Cooler Master ML120L RGB with Kryonaut Extreme | Cooler Master Seidon 120V with Kryonaut
Memory 4*8 GB (32 GB) Kingston HyperX Fury @3266 MHz 16-16-16-35 | 2*8 GB DDR4 @ 3100 MHz CL16
Video Card(s) Galax RTX 2060 Super 8 GB (+170 MHz core, +850 MHz memory) | Zotac GTX 1660Ti AMP 6 GB
Storage Crucial P5 1 TB, Kingston A2000 500 GB, WD10EARS, ST2000DM006 | HyperX Fury 240 GB + 500 GB HDD
Display(s) Dell S2418H, 1080p 75 Hz, HDR | Samsung FHD TV
Case Lianli Lancool 215 | Corsair 100R
Audio Device(s) Realtek S1220 (Supreme FX) | Realtek ALC1150 (Supreme FX)
Power Supply ASUS ROG Strix 750G (80+ Gold) | Cooler Master MWE 650 V2 (80+ Bronze)
Mouse Cooler Master Storm Devastator
Keyboard Cooler Master Storm Devastator
Software Windows 10 Pro x64 |Windows 10 Pro x64
Benchmark Scores Cinebench R20 MT - 4200, ST 537 | R20 MT - 1979, ST - 516
because your scenario is the exact opposite.

normally people run ram higher than the max stable FCLK and then it gets cut in half.
and faster RAM with a slower FCLK that is not 1:1 skips clocks and the overall performs falls off a cliff.
I did not get you. I have seen people running 3200 MHz RAMs with 1600 MHz FCLK or just left it at auto (which essentially sets itself to half of whatever the effective DDR frequency is or equal to the value of MCLK). Even AMD Ryzen Master does not allow the FCLK to be set higher than the MCLK. But since the BIOS has the flexibility, I tried it and it has been a win.
While it's true that increasing the MCLK (if RAMs are capable) and setting the highest possible FCLK is guaranteed to increase performance, the other way round does not have much exposure and I am just wondering why since it has done wonders for my exact system at least.
I did try setting 1900 MHz FCLK as well on my friend's Ryzen 5 3600 based system with an MSI B450 motherboard and 3200 MHz TridentZ and he has got similar performance gains as well.
 
Top