The most efficient cruncher rig possible

Message boards : Number crunching : The most efficient cruncher rig possible

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5

AuthorMessage
Endgame124

Send message
Joined: 19 Mar 20
Posts: 63
Credit: 14,855,841
RAC: 12
Message 95698 - Posted: 1 May 2020, 23:15:02 UTC - in response to Message 95691.  

Yeah, I set it to 3 threads from the start. Happy to bump it up to 4 to see the impact though.

If you keep yours at 3, and I leave my stock one at 4, we should get some good comparison points!
ID: 95698 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
sgaboinc

Send message
Joined: 2 Apr 14
Posts: 282
Credit: 208,966
RAC: 0
Message 95767 - Posted: 2 May 2020, 5:19:03 UTC

ID: 95767 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
sgaboinc

Send message
Joined: 2 Apr 14
Posts: 282
Credit: 208,966
RAC: 0
Message 96596 - Posted: 18 May 2020, 5:41:41 UTC
Last modified: 18 May 2020, 6:21:32 UTC

my Pi4 levelled off at around 954.14 RAC running on 3 threads 1.75ghz. Based on this https://boinc.berkeley.edu/wiki/Computation_credit, gflops = RAC / 200 ~ 4.77 gflops. note that i've been stopping between to give it a 'break' occasionally.

it seemed to come in lower than BCM2835 [Impl 0x41 Arch 8 Variant 0x0 Part 0xd08 Rev 3] entry on https://boinc.bakerlab.org/rosetta/cpu_list.php which states 8.50 gflops. it is likely due to the breaks in between, running on 3 cores and another possible reason is that in 'real world' context, there are codes such as memory and structures handling that do not contribute to gflops alone. so RAC looks lower.

my Pi4 954.14 RAC running on 3 threads 1.75ghz. active sink fan cooling
https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=4235245
it is quite interesting when compared vs Endgame's
stock Pi4 1,231.96 RAC Flirc Aluminum case
https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=4215281
Pi 4 4GB @2015 mhz 1,175.42 RAC Flirc Aluminum case
https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=4269102
from dcdc looks oddly lower at 490.40 RAC accordingly 3 threads stock speeds passive heat sink
https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=4021327

it would seem that running for different periods and different wu done could have made a difference as well.
i've seen wu running for 5 hours and gives 50 credits and wu running for 5 hours giving 100 credits. i'd guess it is also partly 'luck' based on the wu received that the RAC varies.

RAC per core in my case running on 3 threads, 954.14 / 3 ~ 318 RAC per core ~ 1.59 gflops / core

RAC per core based on Endgame's stock running on 4 threads, 1,231.96 / 4 ~ 308 RAC per core ~ 1.54 gflops / core

it seemed to pull rather close, the slight bump may be due to the overclock @ 1.75ghz and possibly less onchip cache contention but otherwise they are pretty much levelled off. statistically i don't think they are very different as different wu crunched returns different RAC. it seem sink fan cooling may make a difference, but this isn't a fair comparison as if we take 'placebo' i.e. no heat sink at all, chances are that the Pi4 would run much hotter nearer to 80 deg C. a good heat sink or good sink fan is very much necessary to run high loads (including r@h) on the Pi4
ID: 96596 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Endgame124

Send message
Joined: 19 Mar 20
Posts: 63
Credit: 14,855,841
RAC: 12
Message 96608 - Posted: 18 May 2020, 15:05:53 UTC

On my overclocked pi 4, I had misconfigured swap sizes, and it was occasionally failing when it picked up more than 5gb of Rosetta processes. I also Bumped the over voltage to 5 just to be 100% certain I wasn’t hitting voltage problems.

After correcting the issue, it has run perfectly for around 3 days. It will take some time for the RAC to stabilize though. It’s still going up every time I check it.
ID: 96608 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Endgame124

Send message
Joined: 19 Mar 20
Posts: 63
Credit: 14,855,841
RAC: 12
Message 96609 - Posted: 18 May 2020, 15:05:56 UTC
Last modified: 18 May 2020, 15:09:41 UTC

In regards to my pi3, it is no longer consistently receiving Rosetta work units. I’m wondering if boinc / Rosetta is getting wise to adjusting the amount of memory in the config file. At this point, I’m likely to pull the pi 3 off Rosetta and assign it fully to other projects.
ID: 96609 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
bkil
Avatar

Send message
Joined: 11 Jan 20
Posts: 97
Credit: 4,432,035
RAC: 37
Message 96767 - Posted: 24 May 2020, 18:15:25 UTC - in response to Message 95129.  
Last modified: 24 May 2020, 18:25:59 UTC

Just a small update. After assembling the $SUBJECT PC, it consumes 30.8 W when idling.

The PSU of choice at the end was an FSP HEXA 400W (because it could be had for a 50% discount): https://www.jonnyguru.com/blog/2013/05/20/fsp-hexa-he-400-400w-power-supply/

It beeps without VGA, but nevertheless boots correctly and is accessible over ssh. Burn-in will continue in the coming days.
ID: 96767 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
bkil
Avatar

Send message
Joined: 11 Jan 20
Posts: 97
Credit: 4,432,035
RAC: 37
Message 96768 - Posted: 24 May 2020, 18:24:28 UTC - in response to Message 96596.  
Last modified: 24 May 2020, 18:29:41 UTC

What you have measured just highlights what I have mentioned above: you can't trick physics. If you are bumping up the the voltage and disabling a core, you are losing efficiency.

Comparing top scores doesn't make sense.

I've recommended to compare a RAC over many weeks (>2), preferably from the same time interval. ~1000 RAC does sound realistic for a Raspberry Pi 4. dcdc's system doesn't seem to be crunching since long and it's not in continuous operation:

https://www.boincstats.com/stats/-1/host/detail/346676938/lastDays

If it is not cooled properly, it will throttle itself even above 60C, but otherwise the values should match closely between well-cooled devices.

You also didn't seem to have picked up my hint that if you use a titanium ATX PSU to power multiple Pi's then it will vastly increase the performance/watt ratio (compared to ~60-70% wall warts) and that if you purchase 100 Pi devices they will give you a huge discount - I've tried this myself.
ID: 96768 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Endgame124

Send message
Joined: 19 Mar 20
Posts: 63
Credit: 14,855,841
RAC: 12
Message 96769 - Posted: 24 May 2020, 19:01:37 UTC - in response to Message 96768.  

What you have measured just highlights what I have mentioned above: you can't trick physics. If you are bumping up the the voltage and disabling a core, you are losing efficiency.

Comparing top scores doesn't make sense.

I've recommended to compare a RAC over many weeks (>2), preferably from the same time interval. ~1000 RAC does sound realistic for a Raspberry Pi 4. dcdc's system doesn't seem to be crunching since long and it's not in continuous operation:

https://www.boincstats.com/stats/-1/host/detail/346676938/lastDays

If it is not cooled properly, it will throttle itself even above 60C, but otherwise the values should match closely between well-cooled devices.

You also didn't seem to have picked up my hint that if you use a titanium ATX PSU to power multiple Pi's then it will vastly increase the performance/watt ratio (compared to ~60-70% wall warts) and that if you purchase 100 Pi devices they will give you a huge discount - I've tried this myself.


Stock pi 4 stabilizes at roughly 1,200 average credit. I think running at 2gh and a slight over volt is the sweet spot, but I need my over clocked pi’s credit to stabilize for another week to verify.

As for running 100 pis off of a single titanium psu, how are you wiring it to create 100 usb c cables? That is a pretty important bit :)
ID: 96769 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
bkil
Avatar

Send message
Joined: 11 Jan 20
Posts: 97
Credit: 4,432,035
RAC: 37
Message 96783 - Posted: 25 May 2020, 21:30:43 UTC - in response to Message 96767.  

Testing with an 80GB HDD (FSP Hexa, B450M DS3H, AMD 3700X, 16GB DDR4) consumes 116-118W full load power from the wall under BOINC. Running from a RAM disk could shave off some more.

With a 22C ambient temperature, the CPU core heats up to 82C.

The cores seem to stay boosted to around 3.88GHz.
ID: 96783 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
bkil
Avatar

Send message
Joined: 11 Jan 20
Posts: 97
Credit: 4,432,035
RAC: 37
Message 96784 - Posted: 25 May 2020, 21:35:06 UTC - in response to Message 96769.  

It's just a matter of good soldering skills and preferably also using some 12V-5V DC-DC conversion if the power supply isn't connected to anything else (depends on exact kind).
ID: 96784 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Previous · 1 · 2 · 3 · 4 · 5

Message boards : Number crunching : The most efficient cruncher rig possible



©2022 University of Washington
https://www.bakerlab.org