Initial Raspberry Pi 4 Performance Benchmarks

[AndyChow]

Only if you use one of the newer smart-cables that can autodetect the device. Use a dumb cable and the USB port is no longer a problem. Still a faulty design though.

There is no such thing as a dumb USB-c cable, at least not according to the specs. Sure, you can use a non-compliant USB-c cable which, if you try to charge your cellphone from accidentally, may or may not fry it.

[brent]

Yeah, the designers skimped on a resister that they shouldn't have. Companies taking liberties with specification circuit diagrams and making assumptions about optionals (none of the reference circuit for USB-C is optional in reality) in the name of saving half a cent per product built is rife in consumer electronics. Nothing new there.

A resistor basically costs nothing in mass production, so that's really unlikely. It's most likely a legitimate oversight, due to the complicated nature of the USB-C standard. It still should have been caught in QA though.

[woife]

I have benchmarked the RPi 3B and RPi 3B+ quite a lot when I had problems with my NESPi-Case. Here are my findings:

• For ongoing load (as in game emulation) you have two enemies: heat and under voltage, both leading to throttling
• Throttling leads to a significant loss in processing power. Any Raspberry Pi benchmark without information about throttling is useless. Sorry Michael, I appreciate your effort, but we need that data.
• The common small passive heat sinks make no difference for ongoing load. You reach the same temperate (and throttling) as without them, just a little later.They might be useful for burst loads.
• You could use huge passive heat sinks. There are a few Youtube videos on that topic. Use a PC-style heat sink on your Pi and it will stay cool and silent. The drawback is that this won't fit in a typical Pi case.
• Active heat sinks are a completely different story: even small ones cool the Pi down quite impressively. Personally I went with the "GeeekPi Raspberry Pi Fan Kit" in the end. That one works with 5V and 3V. Even at 3V it keeps my Pi cool during gaming, and with 3V it is so silent you can't hear it outside the case.
• Be careful with external circuitry on the power supply path (remember, the Pi has no real power button, so custom circuits are quite common on that path). The Pi has a very narrow input voltage range. The voltage output for typical power supplies is 5V or slightly above, but if you have a small circuit between the power supply and the Pi you can loose e.g. 0.5V and that could already lead to throttling because of under voltage. I had to heavily modify my NesPi+ case to make it work. The way it comes out of the box leads to throttling because of under voltage.
• The RPi 3B+ is a lot better than the RPi 3B in terms of thermal behavior. I don't know the details, but the SoC has a metal housing, so I guess that is partly the reason for this.
• When investigating throttling, the command vcgencmd get_throttled is your friend. Documentation about this tool is rather limited, but this post sums it up quite well.
• You can force the Pi to ignore under voltage. Add avoid_warnings=2 to /boot/config.txt. With this setting, the Pi will keep going full speed even in an under voltage condition. The risk you take is that the system could go down immediately, meaning you can loose data. In my tests it has worked quite well: I could drive my Pi with a weak power supply without throttling. You have been warned, use it at your own risk.

// Edit: fixed typos/wording

[LoveRPi]

The problem with the Raspberry Pi 3 Model B was that the packaging (black plastic) doesn't transfer heat well. When the CPU die area got hot, it was significantly far enough away from the thermal sensor next to the GPU to overheat. While the GPU area maxed out at 85C, the CPU area exceeded 100C. They addressed this in the Raspberry Pi 3 Model B+ with a metal heatspreader which distributed the heat generated by the CPU so that the firmware will downclock when the CPU properly.

The Raspberry Pi 4 uses very hot OoO A72 cores which is fine when you use mostly single threaded workloads. It will consume around 100% more than before. The problem becomes when you use all 4 cores. It will start to consume around 200%+ of the A53 cores but the 28nm helps a little here.

In stock configuration, the Raspberry Pi 4 Model B will run about 10 seconds at full performance (1.5GHz). Then runs for 30 seconds with 33% less performance (1GHz). Then runs for 60 seconds with 60% less performance (600MHz).

With heatsink and active cooling, it will maintain 1.5GHz.

[blackshard]

[*]The RPi 3B+ is a lot better than the RPi 3B in terms of thermal behavior. I don't know the details, but the SoC has a metal housing, so I guess that is partly the reason for this.

The 3B+ has also a metal foil inside the PCB which spreads heat away from the SoC to the whole board. In fact the 3B weights 10g and the 3B+ weights 50g due to the metal foil.

[F.Ultra]

There is no such thing as a dumb USB-c cable, at least not according to the specs. Sure, you can use a non-compliant USB-c cable which, if you try to charge your cellphone from accidentally, may or may not fry it.

Ah ok, I see it's not needed to have your cables certified to be USB-C compliant, what a major loop hole they created there.

[woife]

The problem with the Raspberry Pi 3 Model B was that the packaging (black plastic) doesn't transfer heat well. When the CPU die area got hot, it was significantly far enough away from the thermal sensor [...]

The 3B+ has also a metal foil inside the PCB which spreads heat away from the SoC to the whole board. In fact the 3B weights 10g and the 3B+ weights 50g due to the metal foil.

Thank you both for the detailed explanations, that makes sense to me now.

With heatsink and active cooling, it will maintain 1.5GHz.

Exactly, this is my main conclusion: if you consider an RPi for any performance-related task, use a heatsink and active cooling.

[nuetzel]

Negative press about the USB-C implementation on the Pi4 is starting to grow. Word on the street is its a foobar.

The USB-C cable guru at Google has declared it a problem already, which is not a good sign.

Look for a new port design out of the foundation soon.

Also some people have already hacked the Pi4 and wired a PCIe 1x slot to it. Pretty wild.

Sorry in German:
Raspberry Pi 4: Zukünftige Board-Revision soll Probleme mit USB-C-Kabeln lösen

https://www.heise.de/newsticker/meldung/Raspberry-Pi-4-Zukuenftige-Board-Revision-soll-Probleme-mit-USB-C-Kabeln-loesen-4467303.html

[elatllat]

/(compared to various other) (ARM SBCs)/$1 slower $2/

unlike the N2, the pi-4 has thermal problems and can only use encrypted disks at usb2 speed.

[discordian]

Tests are weird, differences between rp4 2gb and rp4 4gb, aswell as the stark difference between tinkerboard and firefly (both use rk3328 and should be reasonable close).

Running without heat sinks, or using different software stacks?