• Introducing XDA Computing: Discussion zones for Hardware, Software, and more!    Check it out!
  • Fill out your device list and let everyone know which phones you have!    Edit Your Device Inventory

[KERNEL][CM7][Froyo] Dalingrin's OC kernel w/fixed IO perf (quadrant) & 1.2Ghz test

Search This thread

Shawn_230

Senior Member
How much testing have the people who are saying 1190MHz works perfectly fine for them done?

I've got this feeling that you just haven't messed around with it enough? I've gotten it to run stable at 1190MHz but a simple reboot or a tweak in the settings can completely ruin things.

My Nook was fine on n31 and 1190MHz with interactive governor last night, but when I woke up today it had shut itself off and the only way it would turn on was holding the power button for 7 seconds. And when it loaded all the widgets I had set up said "problem loading widget" and the system was overall too sluggish to be usable.

Try and do a bunch of reboots or change the settings a bit in SetCPU (raise the minimum speed or change the governor type) and you just might see things mess up on you. I hope some people really did get better quality 'wafers' in their Nook but personally my Nook has at times seemed fine and others not on this kernel.
 

poofyhairguy

Senior Member
Feb 19, 2011
501
101
Instead of dialing it up maybe tweak it down some? I don't even know, but is it possible to maybe run at 1180 instead of 1190? 1170? 1160, 50, etc?

It seems logical that at some level there is a point where most Nooks stop working. All we know for sure is that most peoples Nooks run fine at 1100, but not most at 1190. So maybe if possible all together we can test different ranges to find the optimal spot between those two points (if it is possible to change/add overclock values).
 
Last edited:

SCWells72

Senior Member
Feb 12, 2011
558
100
I just installed the 1100 no-net kernel and got ~1800 in Quadrant, so this is definitely a nice improvement. Overall interactivity seems smoother, too. I might try the 1190 kernel in a bit, but just the fact that it makes the CM7 settings lock up makes me a little nervous. GREAT work, guys! Between this and the Honeycomb-like usability features, CM7 is becoming one heck of an amazing experience for the NC!!!
 

altimax98

Senior Member
Nov 7, 2008
2,227
676
Florida
im doing some tests and the 1190 kernal runs impressive numbers on both 1100 and 800. but sadly it seems that my nook doesnt care for the 1190 speed :( its weird too because the 20 nightly ran it fine, im on 31 now
 

Shawn_230

Senior Member
So I gave up on 1190MHz and went to the 1100MHz no-net kernel.

At first it didn't seem too much worse than 1190MHz, Quadrant was a few hundred less, say about ~1800 or so... this one doesn't crash like the max kernel but I've noticed it slows down alot.

If I use it for say 20 minutes and then run Quadrant it's only about ~800 or so and the system is very sluggish. This is max CPU set to 1100MHz with interactive.
 

glenairlie

Member
Mar 5, 2011
46
2
So I gave up on 1190MHz and went to the 1100MHz no-net kernel.

At first it didn't seem too much worse than 1190MHz, Quadrant was a few hundred less, say about ~1800 or so... this one doesn't crash like the max kernel but I've noticed it slows down alot.

If I use it for say 20 minutes and then run Quadrant it's only about ~800 or so and the system is very sluggish. This is max CPU set to 1100MHz with interactive.

It seem that the Quadrant benchmark is very flaky.

A benchmark should not return significantly higher scores between the first and (say) third run. That implies it's relying on cached data or some carry-over from a previous run.

If you look at say, -3DMark- on Windows, if you run it multiple times it generally returns the same benchmark result. In other words, it's a true test of your systems baseline capabilities.

Quadrant has some reliability issues imo. Not a great benchmark program...
 

chisleu

Senior Member
Mar 7, 2011
1,042
105
Myrtle Beach SC
It seem that the Quadrant benchmark is very flaky.

A benchmark should not return significantly higher scores between the first and (say) third run. That implies it's relying on cached data or some carry-over from a previous run.

If you look at say, -3DMark- on Windows, if you run it multiple times it generally returns the same benchmark result. In other words, it's a true test of your systems baseline capabilities.

Quadrant has some reliability issues imo. Not a great benchmark program...

Or that the computing power of our device isn't that high, and background/system processes can greatly affect the results.

Plus there is illusion in numbers. The difference between 1950 and 2050 is really big where the difference between 1.95 and 2.05 doesn't see like as much. :D
 

Divine_Madcat

Retired Forum Moderator
Sep 9, 2005
2,940
2,613
USA
There is a reason I've only released a 1.19ghz kernel. 1200mhz+ isn't stable no matter what voltage you push it to. 1.19ghz is really about the max you can get out of our CPU right now. Even 1.19ghz has sketchy stability. I think ~1.19ghz is the ceiling unless we increase the base dpll clock.

No worries.. the "1.2Ghz" is set to 1190 (i figured you had done that for a reason ;) ).

Anyway, thanks for all the continued feedback guys; i am happy to hear that people are having a noticable improvement. As for bugs some people are reporting - i can't say i have an answer for them; this really isn't a new kernel - the only changes were the bug fix, so nothing new should have been introduced (but who am i kidding; as a CS guy, i know the smallest change can have huge results..)

That said, i will note that Dalingrin uses a different toolchain than i do for compiling, which could produce slightly different results. Perhaps i will have to investigate that further.....
 

jtbnet

Senior Member
Oct 28, 2009
568
59
Cochituate
Or that the computing power of our device isn't that high, and background/system processes can greatly affect the results.

Plus there is illusion in numbers. The difference between 1950 and 2050 is really big where the difference between 1.95 and 2.05 doesn't see like as much. :D

Right... There are specific benchmarks for specific subsystems that I would expect to be more consistent... Quadrant tends to be a more 'General Useability' benchmark providing a weighted average of multiple subsytems on each run thus background processing has a sizable affect on the outcome... but all in all I have found that the best of 3 runs to be a good barometer for 'General Useability' overall of the release/version I'm testing...
 

colbur87

Senior Member
Feb 25, 2011
242
92
Oklahoma City
Any advice here. Switched to this kernal from the stock cm7 which i had clocked max at 925.
After trying this out my whole system is lagging.....quad scores from 1500 to 900.
Everything is lagging. Anyone have anything?
 

Shawn_230

Senior Member
It seem that the Quadrant benchmark is very flaky.

A benchmark should not return significantly higher scores between the first and (say) third run. That implies it's relying on cached data or some carry-over from a previous run.

If you look at say, -3DMark- on Windows, if you run it multiple times it generally returns the same benchmark result. In other words, it's a true test of your systems baseline capabilities.

Quadrant has some reliability issues imo. Not a great benchmark program...

So what? Anyone who has used Quadrant knows the first test you do is slightly lower.

I'm only reporting the results as a relative comparison... as long as my system is running smoothly I couldn't care less is Quadrant throws me negative results.

But unfortunately neither the 1.1 or the 1.2GHz version of this kernel seem to work for me beyond just for testing.
 

lschroeder

Senior Member
Dec 1, 2010
372
28
CM7 on Class10 SD, n31, 1190mhz "ondemand".

Best score: 2018
Average: mid-1800's

Nice work!!! and THANK YOU!

Now, if only one of the Gods would figure out the audio-processing issue that's causing the video problems, we'd be all set! :)

1. What Class 10 SD card manufacture and what size?

I have the original verygreen SD card image on SanDisk Class4 4 GB and switched to phiremod nook v5.1 on emmc.

2. Are you running this, his new Size agnostic SD Card image and CM7 installer with updater, or something else?

3. What kernel did you use for the SD card version?

I also need the wizards to figure out the video problems.
 

ADude

Senior Member
Mar 5, 2011
528
59
I have the original verygreen SD card image on SanDisk Class4 4 GB and switched to phiremod nook v5.1 on emmc.

2. Are you running this, his new Size agnostic SD Card image and CM7 installer with updater, or something else?

3. What kernel did you use for the SD card version?

I also need the wizards to figure out the video problems.

2. Those methods only differ in how the installation is updated, the same versions should run the same regardless of how they are installed.

3. There is only one 1190 hz kernel, read back in the thread.

And BTW, I disagree - we don't need wizards to figure out the video problems - they just need to point their wands and say "videosmoothicus !"
 

rahlquist

Senior Member
Dec 19, 2008
555
48
50
Villa Rica, GA
www.southerncorgis.com
It seem that the Quadrant benchmark is very flaky.

A benchmark should not return significantly higher scores between the first and (say) third run. That implies it's relying on cached data or some carry-over from a previous run.

While I will agree to an extent on a multitasking, multithreaded os its not uncommon for differences to occur. The only way to prevent variations is to shut off every other app and code and run with highest priority. Even doing that is not going to be 100% perfect. The only true way to benchmark without fluctuation is single threaded assembly language. And killing all the apps in the task manager doesnt scratch the surface. I just checked my nook with

ps | wc -l

And it returned 142 processes running. Any of those could wake up during a benchmark, eat cpu or IO cycles and cause a variance.

Bottom line is any benchmark can be used as one as long as it consistently does the same thing. Whether its copying a file or performing math. Some excel at single tasks like linpack. Others like quadrant are meant to benchmark the system as a whole and virtually every one of us has a different system because we all use different apps.
 

paleh0rse

Senior Member
Jan 4, 2011
1,942
627
1. What Class 10 SD card manufacture and what size?

I have the original verygreen SD card image on SanDisk Class4 4 GB and switched to phiremod nook v5.1 on emmc.

2. Are you running this, his new Size agnostic SD Card image and CM7 installer with updater, or something else?

3. What kernel did you use for the SD card version?

I also need the wizards to figure out the video problems.
1. This Kingmax card that gets surprisingly nice reviews/speeds:
http://www.amazon.com/Kingmax-MicroSDHC-Class-Performance-Memory/dp/B003TT23QI

2. I'm running Verygreen's size-agnostic SD card image/CM7 installer with the n31 nightly.

3. I used Divine's 1190mhz kernel listed in this OP, used SetCPU with max at 1190, min at 800, "ondemand"

For the record, it did seem to get flaky after a little bit of use, so I dialed the max back to 1100 for everyday use. Other than that, though, it has been smooth sailing all the way! :)
 

ax0r

Senior Member
Jan 17, 2010
170
20
Sacramento
This kernel ran great with on nookie froyo at 1.19 ghz and offered a very silky smooth and responsive ui but had severe wifi looping and would freeze on some programs like when loading images on tapatalk. The short bench on setcpu was 160 Ms which is trademark for a fast system.

Had to revert to a different kernel though because no wifi sucks on the nook :(

Hope someone can fix this problem

Sent from my Nexus One using Tapatalk
 

Shawn_230

Senior Member
I've managed to make the 1190MHz kernel run stable with CM7 n31. It doesn't seem to lag like the 1100MHz did last night.

Battery life is good, Bluetooth works... if video wasn't so laggy the Nook would finally finally have finished converting to a full-fledged tablet. I can't wait until CM7 gets proper video playback.

The only issue is though I must have flashed a new CM7 and this kernel some dozen times before it was stable enough to use. Others who are having issues probably just haven't flashed enough times? :p
 

Top Liked Posts

  • There are no posts matching your filters.
  • 20
    [KERNEL][CM7][Froyo] Dalingrin's OC kernel w/fixed IO perf (quadrant) & 1.2Ghz test

    Disclaimer: As always, flash at your own risk.

    Background
    So, recently there has been alot of debate over why quadrant (and to some extent, system performance) was lower with the OC kernel, as compared to the stock CM7 kernel. (http://forum.xda-developers.com/showthread.php?t=1002504) As part of my examining the problem, i went looking for a solution. Starting from the first OC CM7 kernel, i would run quadrant and see how the performance fared. I finally found the issue appeared in the 3-08a kernel; using that, i rolled back changes until i found the commit remove to fix the issue. Specifically, it seems to be somewhere in the 2.6.29.6 update: https://github.com/dalingrin/nook_kernel/commits/gingerbread-exp

    In short, using his source, i have compiled a kernel with fixed I/O performance, and all his other tweaks (interactive gov, etc).

    So, I want to make this clear; this is still dalingrin's kernel; in no way do i wish to take credit for his work. But, while he is out, work must go on, right?

    Information
    Who is this kernel for?
    eMMC Froyo and CM7 users

    Will this work on SD cards?
    It may if you manually replace you uImage, but i cannot guarantee it (since i cannot test it)

    Is everything working in this release?
    everything like bluetooth and dsp should still be just fine (you may have to do a few hard shutdowns/reboots to kick on BT, like normal).


    Kernel Links -3/25/10
    OC kernels (up to 1100Mhz, Interactive gov, BT and DSP):
    Regular
    http://www.mediafire.com/?k00ruvyulf01y64

    Extra network features (PPP, etc)
    http://www.mediafire.com/?i9h76wn483tk16v

    OC kernel (up to 1190Mhz, Interactive gov, BT and DSP):
    http://www.mediafire.com/?v6d2jpzaynvchde
    For those having problems setting this speed in the CM7 performance settings, give setCPU a try, but don't use profiles. Using set CPU, i am able to get my system stable at 1190, where the CM7 manager instantly locks up.


    Feedback
    After flashing, please make note of OVERALL system response. The dip in quadrant scores prompted the fix, but i feel that the system is overall better; I would like some more input if others are seeing this as well.



    Any and all feedback is welcome, and I will do what i can for support...
    4
    Thanks for taking care of this while I am gone. Much appreciated.
    The update to *.6 is not necessary anyway.
    When I get back I will revert the *.6 update and compile some of the netfilter/iptables and tun as modules as this seems to be more compatible than compiling them directly into the kernel.
    1
    Just started using CM7 last night (w/bottom keys mod, so the nightly is from a few revisions back), but just installed this kernel (the regular version, w/out the Net adds, which I don't need), wiped cache & Dalvik, rebooted, set OC to 1100 and "performance" in CyangenMod Settings and now getting 1918 Quad score...vs. 1372 on the stock kernel with no OC.

    I can definitely live with that score...great work to you and dalingrin!!
    1
    I tried both of these from SD card. The non-net features version boots but posted some terrible #s for me (<700 Quadrant and ~6-7 Linpack); the net features version doesn't boot. Is there something special I need to do with the first kernel to see the benefits? I checked and was clocked to 1.1GHz in CM settings.


    Mine on SD Card, CM7 #30

    @1100 GHz got 1918 Quadrant; 14.01 Linpack
    @800 MHz got 1347 Quadrant; 10.709 Linpack

    uSD 2GB no name, no class :)
    1
    Awesome! 2193 is just insane; i am glad to hear its working for you, since....(

    Well that was fun while it lasted. Picked up the NC after an hour of "sleep" and got past lock screen no problem. Opened up QuickPic, at 1190MHz the folder thumbnails never flew by so quickly (I've got 11000 JPEG, about 250Kb each, spread across 94 folders in the eMMC mount). Next checked eMail, it's been syncing with Exchange server OK (wireless set to never sleep). But then fired up Opera Mobile 11, and touchscreen response started acting wonky. Force killed Opera a few times, same touchscreen behavior and pages never finished loading.

    Nothing actually froze, so I rebooted. First two attempts locked up prior to lockscreen. Third attempt got past lock-screen but locked up again before I could get to Settings to dial back to 1100MHz. At that point I got out CWR 3.0.1.0 SD, booted into recovery, flashed back to madcat-032511, after which NC booted up no problem. Went into Setting/CyanogenMod/Performance and found clock speed defaulted back to 800MHz, promptly overclocked it back to 1100 MHz, and just now ran Quadrant SE three times and got best score 1966.

    I guess 1100MHz is best stable overclock for my NC :(.