How to overclock the Raspberry Pi 5 beyond 3 GHz!

Pi 5 Extreme Overclock — (Image credit: Pexels)

The Raspberry Pi 5 is well known to be the fastest Raspberry Pi, it is the new flagship after all.Originally we managed to overclock the Raspberry Pi 5 to 3 GHz, a great boost over the 2.4 GHz stock speed. Many alleged that they overclocked to much higher speeds, but we confirmed with Raspberry Pi that 3 GHz was the limit.

Well it was, but a new firmware seems to break that speed limit. Can we push the Raspberry Pi faster? We can, but how fast is in the hands of the silicon gods so your mileage may vary.

Latest Videos FromTom's Hardware

Watch full video here:

sudo apt update && sudo apt dist-upgrade

sudo nano /boot/config.txt

arm_freq=3100

force_turbo=1

over_voltage_delta=50000

watch -n 1 vcgencmd measure_clock arm

Les Pounder is an associate editor at Tom's Hardware. He is a creative technologist and for seven years has created projects to educate and inspire minds both young and old. He has worked with the Raspberry Pi Foundation to write and deliver their teacher training program "Picademy".

7 Comments Comment from the forums

Pierce2623

Realistically, 3GHz to 3.2GHz on an a76 core will produce results inside the margin of error.
Reply
oofthebeatenpath

"YouTuber and Raspberry Pi Expert Jeff Geerling managed 3.14 GHz"

A Raspberry Pi reaching a speed of 3.14GHz. Will it get to 3.14159GHz ?
Reply
bit_user

oofthebeatenpath said:
A Raspberry Pi reaching a speed of 3.14GHz. Will it get to 3.14159GHz ?
That reminds me of how NTSC relied on a chroma subcarrier that was specified at exactly 3.579545 MHz. Receivers could sync to a frequency slightly different than that, but the standard really does specify it down to the Hz!
Reply
bit_user

abufrejoval said:
I've finally fallen to the temptation of buying an RP5.
Thanks, again, for taking the time to post your thoughts and experiences!

abufrejoval said:
I got a passive case for €12 which adds some heft but not size yet cools the RP5 even at the max 2900MHz my RP5 permits without overheating or throttling.
For how long did you test it, though? I presume you ran it until it reached a steady state?

abufrejoval said:
I got an Orange PI 5+ for comparison, unfortunately I could only get the 16GB model. And no passive case yet
Yeah, that's one thing about the true Raspberry Pi machines - they have far & away the widest selection of cases. With other boards, you're lucky if you find one good aftermarket case for them. KKSB made a nice aluminum case for my ODROID-N2, but I'm glad I bought it when I did, because I don't fancy the revised version.

abufrejoval said:
the OP5 throttles as nicely as a Ryzen and only synthetic loads get it there.

The OP5 is very nearly the same in CPU as both the RP5 and N6005, but also very near the N6005 (32EU) in terms of iGPU and media performance at somewhat less power: the RP5 simply hasn't kept up with the competition there.
The Orange Pi 5 is made on Samsung 8 nm, IIRC, while I believe the Raspberry Pi 5 is made on something in the range of 14 nm to 20 nm? It does seem the Raspberry Pi folks keep plowing most of their transistor budget into bigger, better CPU cores, rather than GPU. Since they tend to use trailing nodes, they don't have as much transistor budget as their main competitors. For instance, the Amlogic SoC in my ODROID-N2 came out in like 2019 and is made on a 12 nm node, IIRC. Meanwhile, RP4 launched around the same time, using 28 nm.

abufrejoval said:
on my 4k screens, both the Intel N6005 and the OP5 deliver very good Youtube performance without stutter or dropped frames.
How sure are you it's using the full hardware codec? What's the CPU utilization, during playback? If it's not negligible, then what about if you use mplayer with the null video output device?

abufrejoval said:
The quality is quite obviously much worse and even annoying on the RP5: Youtube scales down to 720p yet painting the video on the 4k screen stutters with the VPU, even overclocked to 1000Mhz (CPU at 2900 MHz). No throttling etc., I monitored carefully, just not enough VPU horsepower.
Do you know if it's even using the hardware codec? Are you sure?

abufrejoval said:
Google Maps 3D globe view on Jasper Lake and OP5 are also quite impressive at 4k, while the RP5 is usable but not nearly as smooth there: I don't see a large improvement to the RP4 for GPU/VPU.
Are you aware of specs on their GPUs (i.e. how many shader pipelines, GFLOPS, etc.)?

abufrejoval said:
In short, don't waste your time on overclock, buy the right hardware if you need extra speed!
Eh, but there are other reasons to buy the Raspberry Pi. I tend to agree about overclocking and never do it myself, as I value stability more than a few % more performance. However, if someone has a Raspberry Pi 5, for whatever reason, and they need a little more juice out of it, I guess it's nice to have the option.

abufrejoval said:
Neither the RP5 nor the OP5 get cool on idle;
That's a shame. My ODROID-N2+ stays pretty cool. Right now, the SoC is idling at 29 C and the DRAM is idling at 31 C. Those are steady state temps, as I tend to leave that machine turned on. The cooling setup isn't even ideal, as the SoC is bottom-mounted and the case I have provides some air circulation but definitely less than if you just had the bare board sitting on a table.
Reply
bit_user

abufrejoval said:
I've been trying several CPU hogs like stress and stress-ng,
I like stress-ng, due to the diversity of different methods it has. Hardkernel used the following, in their thermal testing of the N2:
stress-ng –cpu 6 –cpu-method matrixprod && glmark2-es2-fbdev –off-screen –run-forever

https://www.hardkernel.com/shop/odroid-n2-with-4gbyte-ram-2/
IIRC, there's like a 3D matrix product that is more stressful on some CPUs, but I really haven't spent a ton of time exploring its different options.

For a more real-world workload, I find that povray is an easy option.

abufrejoval said:
I've got a bigger heat-sink on order that would eliminate that space but move the throttle point a little further.
Where do your source heat sinks? I've found it difficult to locate good quality copper heatsinks of any substantial size. For my Gemini Lake, I found a guy selling some ancient server heatsinks on ebay that fit with a custom hold-down I rigged. But, I've generally been disappointed with the selection of heatsinks I could find either on ebay or even some electronics component suppliers.

If the Alder-N board I end up getting has enough space, I might try one of the server heatsinks with a vapor chamber.

abufrejoval said:
My only worry would be for the Samsung NVMe drive, as flash doesn't really like high temperatures.
I got some SK Hynix P31 Gold M.2 drives to use on SBCs, due to their excellent power efficiency. However, I take your point that the real issue might not be the amount of heat generated by the drive, but rather how much it picks up from the CPU.

abufrejoval said:
As with my x86 NUCs, I'm ok with throttling vs. noisy fans when it's finely regulated and it only occurs on synthetic workloads I won't be doing in normal operations anyway.
True. I would tend to use power limits to avoid throttling, however.

abufrejoval said:
I used numactl to control which cluster was used by firefox.
taskset is what I use on the Alder Lake i9 that I use at work. I forget if I've tried it on the ODROID N2...

abufrejoval said:
Pretty much, yes, otherwise CPU utlization would max out immediately. And you need quite a big more CPU power than any of these can deliver to decode and draw even 1080p material, let alone 4k with any modern codec.
That's not great info, since only part of it could be accelerated, or perhaps they're using the GPU's shader array instead of its hardware codec engine.

abufrejoval said:
AFAIK the RP5 VPU is a very distinct architecture and not at all a classic GPU, more of a DSP design extended to deal with graphics and media. I fear that it was a super smart choice for the original Raspberry PI, but struggles to scale with multiplying transistor budgets, while maintaining backward software compatibility.
Why would they need to maintain backward compatibility? You know there are different drivers for the VideoCore IV and VI, right? The GPU in Pi 1 to 3 is handled in the vc4 driver, while Pi 4 and 5's GPU is handled in the v3d driver.

abufrejoval said:
The MALI GPUs aren't publically documented to very high details either, which is why on both platforms developers are busy retro-engineering both of them.
They tend to at least publish the base specs, though!

abufrejoval said:
On x86 all sane overclocking potential is typically already explored and consumed by the vendors themselves, so I no longer bother to try.... apart for some initial and 'purely scientific' stability testing.
While I don't overclock x86, I have started fiddling with power limits, after discovering that you can raise them even on non-K Intel CPUs. I now run my "65 W" work PC at a limit (i.e. PL1) of 80 W, which its thermal solution can keep below throttling temperatures in the air conditioned room where it resides. This not only gives me better performance after Tau expires, but also gives me a longer Tau (i.e. boost window). I haven't had to change PL2, since I've never seen it reach the stock limit of 202 W.

abufrejoval said:
I really need to get some precise measurement tool for the 0.5-20 Watt range at the socket...
This one seems pretty good, but I haven't tried to independently verify what it's reporting:
https://www.amazon.com/Suraielec-Calculator-Protection-Electricity-Electrical/dp/B08GSPLZBN/

abufrejoval said:
I don't mind chips running at 90°C internally,
This used to be above the point where Intel CPUs would throttle... I don't love the idea of such high temps. I normally like to keep die temps below 80 C.

IIRC, my Pi v3 I think would would start to soft-throttle around 80 C, but I've seen it go as high as 83 or 84 C.

abufrejoval said:
an SBC like these heating a hefty metal case on idle points to a power consumption that is far above a smartphone without delivering extra value, while those even tend to perform better staying completely cool.
Smartphones are engineered to minimize idle power, though. They really try very hard, since idle power consumption is inversely-proportional to battery life.
Reply
bit_user

abufrejoval said:
Agreed, it's one giant chest filled with benchmarking tools, something for nearly everyone.

What's a little harder to tell (without looking deeper than I've done so far) is how deeply it can go into modern ISA extensions, which a) allow incredible speedups for some use cases b) often need to be hard-coded in assembly, because compilers can't handle them.
Yeah, I think you should just be able to use a profiling tool to see if it's using SIMD extensions. Depending on the code & compiler options used, it's not always necessary to have hand-coded assembly language for decent vector utilization. Compiling with -Ofast implies lossy floating point optimizations, which are sometimes necessary for the compiler to really vectorize something like how you'd expect.

abufrejoval said:
Anything visual has the advantage of potentially catching invalid data, which in my case is much worse than crashes.
We need in-band ECC to become more popular.

abufrejoval said:
For the four passive N5005 Atoms, they came with a heatsink installed. I just made sure the chassis had enough airflow, too. But they are mostly from a generation of Atoms that still stuck with the 10 Watt max TDP.
I think it has more to do with the board than the SoC. My consumer-grade Apollo Lake J4205 mini-ITX board will easily use 25 W, if I set its BIOS to "performance" mode.
https://www.asrock.com/mb/Intel/J4205-ITX/index.us.asp
I got a Corsair PSU with iCUE, last year. I've been meaning to set it up and do some detailed power measurements of that and ma Gemini Lake Refresh board.

abufrejoval said:
I was a bit worried about heat initially, but as far as I can tell the RP5 just won't throttle with the €12 passive case
That's great to hear, especially since it's said to use more power than the Pi 4 and that had throttling problems.

abufrejoval said:
the OP5 throttles very nicely with a pretty small passive aluminum heatsink that's barely bigger than the SoC itsself and about 20mm in height, so it will just fit in the case.
I had a situation with my Pi 3 where I found a good aluminum heatsink and a not so good copper heatsink. They performed similarly, although multi-core workloads very quickly throttled with either. Hence, my frustration. If the copper heatsink had been designed to the same standard as the aluminum one, I'm sure it'd have performed noticeably better.

abufrejoval said:
Copper and vapor chambers are for stuff that burns 200-400 Watts and I guess it's fluid cooling after that.
The first vapor chamber server CPU heatsinks I found came in the Sandybridge era, when the hottest Xeons probably didn't exceed 165 W, if that. They're not limited to those models, though. I'm sure they were also used on CPU models even near 100 W.

abufrejoval said:
I lack experience with just how badly NVMe sticks will deteriorate under heat, but I don't get paid to find out either. So I try to make sure they stay below 60°C under normal operations. If that requires some glued-on heat sink, I'd slightly prefer that over active cooling,
Yeah, I bought a Samsung PM9A3 110 mm M.2 drive for a fileserver machine. It's dual-sided and has no heatsink. I think I found some datasheet indicating typical power usage that caused me to worry about cooling. That sent me on a quest for decent, aftermarket 110 mm M.2 heatsinks, but there's virtually nothing out there which provides adequate coverage of both sides! The best thing I could find was this:
https://www.ebay.com/itm/155880372217
It's a bit flimsy and I wonder how well it will really conduct heat away from the underside, but I'm sure it's better than nothing! I plan to smear heatsink compound along the mating surfaces where the top and bottom half overlap.

abufrejoval said:
but I don't mind what I don't hear.
True, but fans do burn power. And there's dust, as you mentioned.

abufrejoval said:
I bought Project Lasso only to be told that I should use the free taskset... but I found both rather more confusing than helpful. Since 'work' is always on Linux and cpuctl comes included there, I've given up on managing E vs. P on Windows, especially since I then use 'perf' to measure the results.
I'm talking about Linux, so I don't know if Project Lasso is available for it, or if Windows also has something called taskset. On Linux, the only confusing thing about taskset was figuring out which core IDs corresponded to which cores & threads. I think the syntax was also a little weird, but I basically just put my main use cases in shell aliases and now I don't even think about which tool I'm using, anymore. I recently had to check, in fact!

abufrejoval said:
For the Raspberry PI foundation ... the cost of using a vendor off-the-shelve GPU ip block like Mali, could have easily doubled the licensing cost... for the GPU alone, not including the media codec blocks.
How do you know? And how can companies like AMLogic and Rockchip afford not only to license that IP, but also use better fab nodes and still hit price points close to the Raspberry Pi?
Reply
bit_user

abufrejoval said:
the 28nm SoC on the RP5
Source?
Reply

Show more comments

For this project you will need