• 1
  • 3
  • 4
  • 5(current)
  • 6
  • 7
  • 11
Overclocking Pi 2
#61
avoid_pwm_pll=1 only impact core_freq, h264_freq, v3d_freq and isp_freq

For what I understand, over clocking core_freq without avoid_pwm_pll=1 will also over clock h264_freq, v3d_freq, isp_freq because they share a pll, but not ARM SDRAM and GPU

Quote:The GPU core, H.264, V3D and ISP share a PLL, therefore need to have related frequencies. ARM, SDRAM and GPU each have their own PLLs and can have unrelated frequencies.[9]

The following is not necessary with "avoid_pwm_pll=1".

The point is that over clocking h264_freq, v3d_freq and isp_freq is at best useless because it won't have any effect on the GUI and the default frequency is enough to decode video files without trouble.

By the way, my settings are useless too, it was just to see what I can get from my Pi2, but in practice, I don't really see a difference with a 'softer' over clock or no over clock at all (just using Confluence skin, over clock probably help with heavier skins)
Reply
#62
(2015-02-17, 22:37)Patatra Wrote: avoid_pwm_pll=1 only impact core_freq, h264_freq, v3d_freq and isp_freq

For what I understand, over clocking core_freq without avoid_pwm_pll=1 will also over clock h264_freq, v3d_freq, isp_freq because they share a pll, but not ARM SDRAM and GPU

Not quite. The PLL is shared, but each clock has a separate divisor. If you just set core_freq, the other gpu clocks will use the same PLL, but with a higher divisor that ensures they are not overclocked more than requested. This typically means a slight underclock.

E.g.
core_freq=250 => PLL = 2GHz => v3d_freq = 250
core_freq=500 => PLL = 2GHz => v3d_freq = 250
core_freq=400 => PLL = 2.4GHz => v3d_freq = 240
core_freq=300 => PLL = 2.4GHz => v3d_freq = 240
core_freq=301 => PLL = 1.806GHz => v3d_freq = 225 (I think this is the theoretical worst case)

However underclocking v3d/h264 to 225MHz is unlikely to be measurable as those hardware blocks are rarely stressed.

Adding avoid_pwm_pll=1 means v3d_freq=250 in all those cases.
Reply
#63
Anyone knows how I can increase minimum arm frequency (on demand governor, no force turbo)?

I've tried setting the arm frequency to "arm_freq_min=700" but had no luck.

However, using the same parameter with a lower frequency works - "arm_freq_min=400" - setting the minimum arm frequency to 400 Mhz.

Any suggestions?
Image
Reply
#64
(2015-02-18, 17:25)lysin Wrote: Anyone knows how I can increase minimum arm frequency (on demand governor, no force turbo)?

I've tried setting the arm frequency to "arm_freq_min=700" but had no luck.

However, using the same parameter with a lower frequency works - "arm_freq_min=400" - setting the minimum arm frequency to 400 Mhz.

arm_freq_min doesn't let you overclock - only underclock.
If you want the (minor on Pi2) extra speed then use force_turbo.
Reply
#65
(2015-02-18, 18:41)popcornmix Wrote:
(2015-02-18, 17:25)lysin Wrote: Anyone knows how I can increase minimum arm frequency (on demand governor, no force turbo)?

I've tried setting the arm frequency to "arm_freq_min=700" but had no luck.

However, using the same parameter with a lower frequency works - "arm_freq_min=400" - setting the minimum arm frequency to 400 Mhz.

arm_freq_min doesn't let you overclock - only underclock.
If you want the (minor on Pi2) extra speed then use force_turbo.

Hmm, didn't know that. Thanks popcornmix.

force_turbo is not an option just yet. I don't want that warranty bit added for now Tongue
Image
Reply
#66
Real RPi superheroes just bring on the pain right out of that shiny freshly minted RPi wrapper, and put the hammer down from day 1.

I love the smell of warm Raspberries in the morning ! Wink

Reply
#67
(2015-02-17, 13:44)jimmymcmahon Wrote:
(2015-02-17, 02:43)steve1977 Wrote: Not sure whether this deserves a separate thread, but would like to pick your brain on the very fastest SD card for the RPi2. I don't mind space (4gb, 8gb, 16gb), but looking for the fastest out there. Changing library scan from 10min to 10sec appears to be absolutely outstanding and something I would have never expected to be an effect just by the SD card...

Hi mate I can only comment on the 2 cards I have rested. Read performance reviews of other sd cards in this thread. http://www.raspberrypi.org/forums/viewto...&start=225

As others have mentioned in that thread, for the operations that kodi performs sequential write / read speed is not important.
Kodi performs mainly small random 4k/512k read writes a lot and so if you have an sd card with high 4k / 512k write speeds it will perform very well for Kodi.

You will see cards mentioned in that thread that have like 90MB/s sequential write speeds but only 0.1MB/s 4k / 512k write speeds. As such that would be a very bad card for kodi and you would notice considerable lag in the UI / library scan. On the other hand you will see a cheaper card mentioned with much slower sequential write speeds but with a higher speed in 4k/512k writes. As such that card would be better for kodi.

Rule of thumb. Ignore sequential write speeds (they are only good for use in digital cameras / video recorders etc that need to write large files frequently). For Kodi, look for the card with the highest 4k/512k random write speeds and this will dictate its performance in Kodi.

Lots of people are mentioning the stock noobs card as a favourite as above as random write speeds are above 1 MB/s. But as you can see from my results on the sd card I'm using, (http://www.amazon.co.uk/Samsung-Class-Me...B00D6ENF7M) for about £10 you can get a card with 3-4MB/s 4k/512k write speeds. Trust me it's realy fast and UI does not lag in anyway. Even without overclocking.

I'll also point out that when I said library scan. I was just referring to the scanning in of all the series of 1 particular tv show. It was taking 10 minutes + on the old sd card but less than 10 seconds on the new sd card due to the great 4k / 512k random write speeds.

Thanks for your message. Sounds like you are well experienced. I have two more questions for you:

1) You recommend the following: http://www.amazon.co.uk/Samsung-Class-Me...B00D6ENF7M. I tried to find the same card in Germany and ended up with two similarly looking ones. One sold by amazon, but with different serial. One sold by a 3rd party vendor with same serial. I typically prefer to buy with amazon directly, but not sure whether it is the same card that you made great experience with:
http://www.amazon.de/Samsung-MB-MPAGCAEU...00D6ENF7M/ - sold by 3rd party
http://www.amazon.de/Samsung-microSDHC-C...B00AYC6V2K - sold by amazon (but different serial?)

2) Any thoughts on the EVO Class-10 cards, which are also reviewed in the other thread?
http://www.amazon.de/Samsung-MicroSDHC-S...B00J2973JG


Thanks in advance!!!
Server: Asus Sabertooth Z77 | Intel Core i5 3.4 GHz | 16 GB DDR3 | 128 GB SSD, 82 TB (9 x 6 TB, 7 x 4 TB)
HTPC 1: Raspberry Pi 2 | HTPC 2: Raspberry Pi 2 | HTPC 3: Raspberry Pi
Reply
#68
i have tested the Sunsung pro sd card and the 4k blocks are dubble the speed of the sandisk extreme pro 600x..... i think samsung is the winner Smile
Reply
#69
Thanks. Above are links to 2 or even 3 different Samsung SD cards. None of them is called "pro" and maybe I should add this as a fourth option.

Has someone experience with the different Samsung SDs:

MB-MPAGCAEU - http://www.amazon.de/Samsung-MB-MPAGCAEU...00D6ENF7M/
MB-MPAGCEU - http://www.amazon.de/Samsung-microSDHC-C...B00AYC6V2K
MB-MP16DEU - http://www.amazon.de/Samsung-MicroSDHC-S...B00J2973JG
MB-MG16DEU - http://www.amazon.de/dp/B00J297ZQC/

All four are 16GB, all four are class-10. What's the difference and which one should be best for RPi2?
Server: Asus Sabertooth Z77 | Intel Core i5 3.4 GHz | 16 GB DDR3 | 128 GB SSD, 82 TB (9 x 6 TB, 7 x 4 TB)
HTPC 1: Raspberry Pi 2 | HTPC 2: Raspberry Pi 2 | HTPC 3: Raspberry Pi
Reply
#70
To be Honest most Samsung micro SD cards are pretty good but the one jimmymcmahon posted which looks the same as the first 2 in your links is the fastest results I've seen.
They seem to outperform the evo and pro range.
But please be careful I've brought a fake from Amazon before it was easy to spot as it was made in China, and the writing scratched off.. didn't need to run H2testw or f3 on it.
I'm sure popcornmix posted a link earlier on the benchmark test results over at the raspberry pi fourum.
Here it is again.
http://www.raspberrypi.org/forums/viewto...=63&t=4076
The last few pages are probably more relevant for newer cards
Reply
#71
Thanks. It indeed looks the same as the first two, but those two have different serials (MB-MPAGCAEU vs. MB-MPAGCEU). I am concerned about fakes, so prefer to buy from amazon only and not from 3rd party vendors selling through amazon. Dilemma is that the one with exactly the same serial is on amazon only with a 3rd party vendor.

Any idea whether the two Samsung cards with different serials (MB-MPAGCAEU vs. MB-MPAGCEU) are actually the same?
Server: Asus Sabertooth Z77 | Intel Core i5 3.4 GHz | 16 GB DDR3 | 128 GB SSD, 82 TB (9 x 6 TB, 7 x 4 TB)
HTPC 1: Raspberry Pi 2 | HTPC 2: Raspberry Pi 2 | HTPC 3: Raspberry Pi
Reply
#72
(2015-02-18, 17:25)lysin Wrote: Anyone knows how I can increase minimum arm frequency (on demand governor, no force turbo)?

I've tried setting the arm frequency to "arm_freq_min=700" but had no luck.

However, using the same parameter with a lower frequency works - "arm_freq_min=400" - setting the minimum arm frequency to 400 Mhz.

Any suggestions?

What I did is put the following script called autostart.sh in the configfiles folder.

Code:
#!/bin/sh
(sleep 20;
echo 50 > /sys/devices/system/cpu/cpufreq/ondemand/up_threshold
)&

chmod a+x /storage/.config/autostart.sh

cat /sys/devices/system/cpu/cpufreq/ondemand/up_threshold

sh /storage/.config/autostart.sh

I dunno why but with Openelec 4.xx and 5.xx it slows down the boot time significantly, 5 or so minutes? I actually stopped using it with the R-Pi 2 and just use force turbo.
Reply
#73
(2015-02-18, 17:25)lysin Wrote: Anyone knows how I can increase minimum arm frequency (on demand governor, no force turbo)?

I've tried setting the arm frequency to "arm_freq_min=700" but had no luck.

However, using the same parameter with a lower frequency works - "arm_freq_min=400" - setting the minimum arm frequency to 400 Mhz.

Any suggestions?

Have you tried this?

Code:
echo 700000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq

(2015-02-22, 06:13)Wanderlei Wrote: What I did is put the following script called autostart.sh in the configfiles folder.

Code:
#!/bin/sh
(sleep 20;
echo 50 > /sys/devices/system/cpu/cpufreq/ondemand/up_threshold
)&

chmod a+x /storage/.config/autostart.sh

cat /sys/devices/system/cpu/cpufreq/ondemand/up_threshold

sh /storage/.config/autostart.sh

I dunno why but with Openelec 4.xx and 5.xx it slows down the boot time significantly, 5 or so minutes? I actually stopped using it with the R-Pi 2 and just use force turbo.

You better remove that from your autostart.sh

By default "up_threshold" is "50" so that code will only delay booting time.

This line is enough:

Code:
echo 18 > /sys/devices/system/cpu/cpufreq/ondemand/up_threshold

(I have set it to 18, which seems to be a good balance (for me). While playing music (local/AirPlay) etc CPU will stay at min_freq)
Reply
#74
(2015-02-21, 15:50)steve1977 Wrote: Thanks. It indeed looks the same as the first two, but those two have different serials (MB-MPAGCAEU vs. MB-MPAGCEU). I am concerned about fakes, so prefer to buy from amazon only and not from 3rd party vendors selling through amazon. Dilemma is that the one with exactly the same serial is on amazon only with a 3rd party vendor.

Any idea whether the two Samsung cards with different serials (MB-MPAGCAEU vs. MB-MPAGCEU) are actually the same?
I'm unsure, if I were to guess I would say the serial numbers are slightly different as one of the cards is supplied with a SD adapter, the other is not.
Again not 100% sure, if I were you I'd take the gamble and buy the one from the official Amazon link without the SD adapter.
Reply
#75
I've set up my Raspberry Pi 2 as a WiFi Access Point + VPN Gateway device. The raw CPU power was a huge bottleneck with the Raspberry Pi 1, the Pi 2 is over three times faster in my usage. However it's still far from perfect, and I am trying to increase the performance by overclocking. I am mostly concerned with the speed of the 128-bit Blowfish cipher of OpenSSL since that is what my VPN provider uses.

This is what I started with (the settings from raspi-config):
Quote:arm_freq=1000
#gpu_mem=16
core_freq=500
sdram_freq=500
over_voltage=2

With these settings I benchmarked using "openssl speed bf-cbc" and got "blowfish cbc 19095.24k 21153.30k 21657.60k 21785.94k 21826.22k"
from this I only take the last number 21826.22k for comparison.

I experimented with a bunch of settings. Increasing to arm_freq=1100 worked fine and got me a nice increase, but I didn't want to stop there. arm_freq=1200 would crash all the time, but 1150 seemed to work. After that I increased some of the other values. This is what I am running now:

Quote:arm_freq=1150
#gpu_mem=16
core_freq=530
sdram_freq=530
over_voltage=8

With these settings the openssl speed result is "blowfish cbc 21917.48k 24319.51k 24907.61k 25054.21k 25100.29k"

So, 25100.29 versus 21826.22, a 15% boost for free. I've been running it for a while now and it's not even hot.

I don't think there is much more to be gained by overclocking. I wonder if anything could be done to OpenSSL itself to tune it somehow?

I am writing about my results and experiences here: http://www.tuug.fi/~toni/blog/?p=69
Reply
  • 1
  • 3
  • 4
  • 5(current)
  • 6
  • 7
  • 11

Logout Mark Read Team Forum Stats Members Help
Overclocking Pi 22