Has anyone been having any issues with performance with the September 26 image?
I wiped both of mine for different reasons and downloaded the newest one and although I can't give any hard and fast numbers it just feels like there was a performance hit.
I am wondering if they did something to help with the heat issue.
No help on the RPi forums so I figured I would ask here.
Thanks,
Dave
Topic: Raspberry Pi 4 performance - page 2. (Read 1307 times)
Would you please tell us which “cheap China case” did you buy?
Would you recommend it, given you had to do so much DIY to make it operational?
It's this one https://www.aliexpress.com/item/4000056606252.htmlWould you recommend it, given you had to do so much DIY to make it operational?
If you are comfortable adding a small piece of aluminum/copper in between the gap, then it's great. But if you like something that just works then stay away from it.
It's really odd those Chinese manufacturers don't check these kind of things before manufacturing by the thousands...
I think you didn’t mentioned exactly the mode of the case.
Would you please tell us which “cheap China case” did you buy?
Would you recommend it, given you had to do so much DIY to make it operational?
Thanks!
Would you please tell us which “cheap China case” did you buy?
Would you recommend it, given you had to do so much DIY to make it operational?
Thanks!
What is the ambient around the case? Any real airflow? That's a 18C temp drop which is a fairly large percentage.
Tried leaving it running with 100% load on all cores (cpuburn-a53). It stabilizes at 71C after about an hour. Damn hot to the touch but no clock throttling.
If 100% load was a 24/7/365 thing, then I would definitely go with a fan but being an intermediate condition it is fine.
bitcoind and lnd doesn't even seem to mind the cpu being kept busy :-)
Thanks to the Bitcoin I can play with my new toy, thanks Satoshi 
Have fun playing :-)
I need to get a second one now that I allocated mine to run a bitcoin node
Woo-hoo look what I just got in my mail today:
I want to play a bit with it first and later I'll move my full node to the Pi.
Have other projects on the way so I'll need to wait with the testing and benchmarking, but when that time comes I'll share my experience with you guys
Thanks to the Bitcoin I can play with my new toy, thanks Satoshi
I want to play a bit with it first and later I'll move my full node to the Pi.
Have other projects on the way so I'll need to wait with the testing and benchmarking, but when that time comes I'll share my experience with you guys
Thanks to the Bitcoin I can play with my new toy, thanks Satoshi
What is the ambient around the case? Any real airflow? That's a 18C temp drop which is a fairly large percentage.
There's 22C ambient. No airflow but it's in 'the open' (i.e. unrestricted convection) in the corner of an office:
http://too-hot.dk/storage/temp/raspi_case.jpg
Showing 52C after a couple of hours - haven't seen it higher than that yet. Dropped to 51C after taking the image.
You can even tell I'm currently using a rotary disc on the thermal image
http://too-hot.dk/storage/temp/thermal.jpg
I was expecting about 50C given a sizeable heatsink with good thermal interface. A bit of forced convection will really help, but I want a completely quiet solution and 50C is totally fine :-)
Oh, by the way. I removed the anodized surface of the area of the heatsink that evenually contact the cpu. Anodized aluminum is a really good insulator of both electricity and heat. Probably doesn't matter much at this relatively low thermal densitity though. But we recently had an issue at work with a heatsinks delivery that was inadvertently anodized at the contact area where we transfer in excess of 100W/cm2 from a power tranasistor. This caused an increase in die temperature of more than 10C so we had to rework those heatsinks by milling the surface.
Make sure that there is actual contact between the case and CPU in the proper area. Cheap cases suck.
Received one of the cheap cases, the one with ribs. And as DaveF suspected, there were no contact between cpu and heatsink. There were about 1mm air gap 😕.I cut a small piece of 1mm aluminum and fitted it with heatsink compound, which turned it into an ON solution 🙂.
Raspiblitz running at about 50C now.
What is the ambient around the case? Any real airflow? That's a 18C temp drop which is a fairly large percentage.
-Dave
Make sure that there is actual contact between the case and CPU in the proper area. Cheap cases suck.
Received one of the cheap cases, the one with ribs. And as DaveF suspected, there were no contact between cpu and heatsink. There were about 1mm air gap 😕.I cut a small piece of 1mm aluminum and fitted it with heatsink compound, which turned it into an
Raspiblitz running at about 50C now.
Edit: Fixed typo.
Can you mention how much time did it took to finish the sync?
I can't say for sure as I had some troubles with the last part of the sync and did some experiments etc. But at least, the first half of the blockchain (i.e. +100GB) was downloaded within the first day.
With optimum conditions (fast network and SSD drive) I believe a clean sync should easily complete within two days on the RPI4B.
I now also got raspiblitz (Bitcoin + LND) up and running :-)
With the small passive heatsink temperature settles at about 68C with some open connections...
RaspiBlitz v1.3 Blitz99
bitcoin Fullnode + Lightning Network
,/ -------------------------------------------
,'/ CPU load 3.85, 3.75, 2.71, temp 68°C 154°F
,' / Free Mem 3392M / 3906M HDDuse 262G (95%)
,' /_____, ssh [email protected] d8.3MiB u25.8MiB
.'____ ,'
/ ,' bitcoin v0.18.1 mainnet Sync OK 100.00%
/ ,' Public xx.xxx.xxx.xx:8333 17 connections
/,'
/' LND 0.7.1-beta wallet 0 sat (+400000)
0/0 Channels 0 sat 4 peers
I don't think you can extrapolate the remaining half of the blockchain by what amount of time it took to get the first half downloaded. Consider this:
https://jochen-hoenicke.de/queue/#0,all
Unfortunately there isn't older charts that collect this data that I know of, but by looking at that you can see the all time peaks of 2017 during the $20k bubble, that mix of legit traffic + Ver and Bitmain spamming makes that section of the blockchain specially slow to synch in my experience. Maybe it was because it's an HDD and you notice this section less with SSD. When I find some free time I will try for myself and monitor the full sync time and report back.
Can you mention how much time did it took to finish the sync?
I can't say for sure as I had some troubles with the last part of the sync and did some experiments etc. But at least, the first half of the blockchain (i.e. +100GB) was downloaded within the first day.
With optimum conditions (fast network and SSD drive) I believe a clean sync should easily complete within two days on the RPI4B.
I now also got raspiblitz (Bitcoin + LND) up and running :-)
With the small passive heatsink temperature settles at about 68C with some open connections...
RaspiBlitz v1.3 Blitz99
bitcoin Fullnode + Lightning Network
,/ -------------------------------------------
,'/ CPU load 3.85, 3.75, 2.71, temp 68°C 154°F
,' / Free Mem 3392M / 3906M HDDuse 262G (95%)
,' /_____, ssh [email protected] d8.3MiB u25.8MiB
.'____ ,'
/ ,' bitcoin v0.18.1 mainnet Sync OK 100.00%
/ ,' Public xx.xxx.xxx.xx:8333 17 connections
/,'
/' LND 0.7.1-beta wallet 0 sat (+400000)
0/0 Channels 0 sat 4 peers
Did the sync ever finish OK?
Yes it finished. Not without problems though
I naively thought the blockchain would still just fit on my 250GB SDD but turned out I was about 20 gig short.
Then I moved the data to a larger 5400 rpm HDD which of course is awfully slow at sync'ing.
Then then project turned into a filesystem experiment. I re-formatted the 250GB SSD to a btrfs filesystem with zlib compression enabled and copied the blockchain back.
Even though the blockchain is poorly compressible it still compresses to about 80% i.e. freeing up sufficient space with some 10GB to spare after sync'ing was done.
Naturally, the compressed SSD is also significantly slower than uncompressed it is still much faster than the HDD. I guess I just need to order a new disk..
Now I'm fiddling around with raspiblitz to get some hands-on with lightning
I ordered one of those china aluminium boxes - will see how it fits and whether the thermal interface can be done properly. But as mentioned, it is not really a huge problem. It will throttle down if it gets too hot and normally it won't even get that hot when just staying in sync.
Can you mention how much time did it took to finish the sync? Does Bitcoin Core keep such a statistic? this would be interesting to have so you could keep track of it even if you close the device you are using. Just count the time it takes until you reach the first ever sync, this would be extremely useful for the database im doing. Right now im running a bunch of different devices and keeping track of it all with chronometers because as far as I know the client doesn't have anything built in to do this.
I wonder if there is a massive difference Linux vs Windows but that would be x2 the time it will take for this to finish... I think it's best to put this effort in Linux for now.
Very interested in temperature and eventual thermal throttling.
Heard of many instances of thermal throttling of R4 just being sitting idle (probably due to very poor air circulation/high ambient temp).
Very interesting thread here:
https://www.raspberrypi.org/forums/viewtopic.php?f=28&t=243500&sid=6c9afc7c1a072ffb808168ebaa95a7ea&start=525#p1536192
Wondering if just running bitcoind + LN hit the 80 C° threshold.
Heard of many instances of thermal throttling of R4 just being sitting idle (probably due to very poor air circulation/high ambient temp).
Very interesting thread here:
https://www.raspberrypi.org/forums/viewtopic.php?f=28&t=243500&sid=6c9afc7c1a072ffb808168ebaa95a7ea&start=525#p1536192
Wondering if just running bitcoind + LN hit the 80 C° threshold.
Did the sync ever finish OK?
Yes it finished. Not without problems though
I naively thought the blockchain would still just fit on my 250GB SDD but turned out I was about 20 gig short.
Then I moved the data to a larger 5400 rpm HDD which of course is awfully slow at sync'ing.
Then then project turned into a filesystem experiment. I re-formatted the 250GB SSD to a btrfs filesystem with zlib compression enabled and copied the blockchain back.
Even though the blockchain is poorly compressible it still compresses to about 80% i.e. freeing up sufficient space with some 10GB to spare after sync'ing was done.
Naturally, the compressed SSD is also significantly slower than uncompressed it is still much faster than the HDD. I guess I just need to order a new disk..
Now I'm fiddling around with raspiblitz to get some hands-on with lightning
I ordered one of those china aluminium boxes - will see how it fits and whether the thermal interface can be done properly. But as mentioned, it is not really a huge problem. It will throttle down if it gets too hot and normally it won't even get that hot when just staying in sync.
Did the sync ever finish OK?
Also wondering if you ordered the cases or are doing anything else with the cooling. I'm still kicking around ideas for mine, none look great.
-Dave
Also wondering if you ordered the cases or are doing anything else with the cooling. I'm still kicking around ideas for mine, none look great.
-Dave
There's also a great overview on cases/cooling here: https://www.martinrowan.co.uk/2019/09/raspberry-pi-4-cases-temperature-and-cpu-throttling-under-load/
Passive cooling i.e. natural convection does indeed require a really huge heatsink compared to forced air-flow cooling. To be effective you basically need the entire case to be a heatsink.
If you have the data it is fairly easy to calculate the temperature rise of a heatsink given the required power dissipation. Or at least getting a rough estimate.
It might be difficult finding data for a Chinese raspberry case, but let's say we want a heatsink to be 30C above ambient then you would need one with a thermal resistance of 3 degree C per watt at natural convection.
A random example of a heatsink specified for 3 C/W @ natural (not suitable for RPI but just to provide a hint about size) : https://www.digikey.com/product-detail/en/wakefield-vette/401K/345-1048-ND/340344
On top of this comes the thermal resistance in the interface between the chip and the heatsink, which adds to the temperature rise of the silicon. On the other hand, a lot of the power dissipation is caused by other chips on the RPI board than the CPU.
Sorry for blabbering. Just happened to be an electrical engineer doing stuff like that for a living
Btw. I've plugged the RPI to my own LAN. It's sync'ing much faster again. Should be done sometime tomorrow :-)
Passive cooling i.e. natural convection does indeed require a really huge heatsink compared to forced air-flow cooling. To be effective you basically need the entire case to be a heatsink.
If you have the data it is fairly easy to calculate the temperature rise of a heatsink given the required power dissipation. Or at least getting a rough estimate.
It might be difficult finding data for a Chinese raspberry case, but let's say we want a heatsink to be 30C above ambient then you would need one with a thermal resistance of 3 degree C per watt at natural convection.
A random example of a heatsink specified for 3 C/W @ natural (not suitable for RPI but just to provide a hint about size) : https://www.digikey.com/product-detail/en/wakefield-vette/401K/345-1048-ND/340344
On top of this comes the thermal resistance in the interface between the chip and the heatsink, which adds to the temperature rise of the silicon. On the other hand, a lot of the power dissipation is caused by other chips on the RPI board than the CPU.
Sorry for blabbering. Just happened to be an electrical engineer doing stuff like that for a living
Btw. I've plugged the RPI to my own LAN. It's sync'ing much faster again. Should be done sometime tomorrow :-)
There's plenty of bandwidth to the external SSD through the super-speed USB connection. Also the gigabit LAN is no bottleneck.
Yes, so long as you don't use a crap drive or one that is using a crap USB -> SATA chip.
I work on too much low end hardware, and some of the cost cutting things are really irksome.
Some of the newer low end external SSDs have worse performance then an older spinning drive.
Will try to order one or both of these 'cool' 
cases from china. There are also options with fans but I really prefer passive cooling for something that is intended for 24/7 use.
https://www.aliexpress.com/item/4000056606252.html
https://www.aliexpress.com/item/4000095452880.html
cases from china. There are also options with fans but I really prefer passive cooling for something that is intended for 24/7 use.https://www.aliexpress.com/item/4000056606252.html
https://www.aliexpress.com/item/4000095452880.html
Make sure that there is actual contact between the case and CPU in the proper area. Cheap cases suck.
Just scrapped a bunch of aluminum cases for the RPi3 that had enough space between the heatsinnk that was part of the case and the CPU itself that the thermal pad I put on the CPU was untouched on top
-Dave
There's plenty of bandwidth to the external SSD through the super-speed USB connection. Also the gigabit LAN is no bottleneck.
other performance bottlenecks are:
- the dbcache setting. With 4GB RAM, setting this to 1GB (i.e. dbcache=1024) should be fine
- Despite all the RAM, setup a new swapfile (2-4GB) on the SSD, then after turning that new swapfile on, turn off the old swapfile on the slow SD card
- Use bitcoin 0.18.0+ to get better block processing performance (0.18.0 had a double digit performance increase in the database handling code)
Will try to order one or both of these 'cool' cases from china. There are also options with fans but I really prefer passive cooling for something that is intended for 24/7 use.
https://www.aliexpress.com/item/4000056606252.html
https://www.aliexpress.com/item/4000095452880.html
cases from china. There are also options with fans but I really prefer passive cooling for something that is intended for 24/7 use.https://www.aliexpress.com/item/4000056606252.html
https://www.aliexpress.com/item/4000095452880.html
We indeed need cool cases.
Sorry, couldn't resist a silly comment.
Sorry, couldn't resist a silly comment.
Jump to: