Overclockmatic_RK3288 v1.0.6

[Guest]

So not being able to boot is not due to improper patching, and it's rather due to hardware/cpu not being able operating at such high frequency, or it simply doesn't have enough voltage to work stable? Could be true. I might do some more testing over this weekend if I find some time for it. And btw, to be honest I expected better AnTuTu results having cpu clocked 400Mhz per core. Lower frequency would certainly explain that.

[Gormar]

So not being able to boot is not due to improper patching, and it's rather due to hardware/cpu not being able operating at such high frequency, or it simply doesn't have enough voltage to work stable? Could be true. I might do some more testing over this weekend if I find some time for it.

Yes, I think so and my experiments confirm that.
Sadly core voltage of 1.4V is defined as Absolute Maximum Ratings in RK3288 datasheet. You can't set higher value than that or rather you shouldn't. IMHO we are operating to close to it. From 1608MHz frequency all are paired in DVFS with 1.3875V even when you define them in resource file as low as 1.325V in pvtm-operation-points. There is third value called "margin" and it is taken into account when DVFS table is created by kernel code.

[Guest]

I did some more testing today, and I find 1968Mhz pretty stable on my box (@1.4V). I did several AnTuTu stability tests in a row, as well as benchmark tests too, and I didn't experienced any crashes or freezes so far (boot time was 2.5 hours ago). My box had difficulties booting @2112Mhz (duh!), but @2088Mhz I was able to get AnTuTu score of 49k! Now @1.97Ghz I'm getting 48.5k which is still 1k better than before running it @1894Mhz. I will do some more testing tomorrow, and after that I will update my other thread too.

[Jabader]

I did some more testing today, and I find 1968Mhz pretty stable on my box (@1.4V). I did several AnTuTu stability tests in a row, as well as benchmark tests too, and I didn't experienced any crashes or freezes so far (boot time was 2.5 hours ago). My box had difficulties booting @2112Mhz (duh!), but @2088Mhz I was able to get AnTuTu score of 49k! Now @1.97Ghz I'm getting 48.5k which is still 1k better than before running it @1894Mhz. I will do some more testing tomorrow, and after that I will update my other thread too.

That a very interesting but did you overclocked Mali? I tried to change values in mali_kbase_config.c
#define MAX_GPU_ALLOWED_FREQ_KHZ 1000000
#define MIN_GPU_ALLOWED_FREQ_KHZ 1
and compile kernel, modules, change the recource.img but didn't get freq more than 625MHz
Do you know how to receive 700MHz or more?

[Guest]

I've seen Mali-T764 working @700Mhz only on dual-cores, but not on quad-cores. Lower clocks are there probably due to more heat generated, especialy in sealed boxes like HPH. So currently 625Mhz is the maximum we could squeeze out of it. If you somehow find out how to remove GPU frequency limit, please feel free to post your findings here.

[Gormar]

Updated first post with overclockmatic_RK3288 PIE which can work under Lolipop.

At least right now I can modify kernel lock in Wasser's firmware which limit CPU OC to 1.6GHz .

[amclean123]

Updated first post with overclockmatic_RK3288 PIE which can work under Lolipop.

At least right now I can modify kernel lock in Wasser's firmware which limit CPU OC to 1.6GHz .

Awesome! Can you please post a link to the OC Wasser kernel?

[Gormar]

Can you please post a link to the OC Wasser kernel?

Certainly. I am using Lollipop with modded CPU OC kernel limit to 1920MHz. For starters that should do.

https://docs.google.com/uc?id=0B_z6C...xport=download

[cappybob]

Thanks, tried on my Firefly and all went well. Your patch for Lp also worked very well, with no problems.

[sawbones999]

Gormar, have you ever taken a shot at a Pipo RK3288 kernel?

I have been told that no one has unlocked the safety frequency on any Pipo devices because of some sort of unknown compression type on the kernel, or something crazy like that. I've been meaning to try and pry it apart with binwalk and firmware modification kit, but haven't gotten around to it yet since I got my M9S.

I did begin using Burs restool to unpack the resource image, but got some errors. He was very cool to look at it and found that there are folders in my resource image. His tool did not take this into account (because it's probably not that common). Anyway, he fixed it and now I can unpack and pack my resource image! (New restool version is out!).

I had already decompiled my dtb to dts (that file wasn't in a folder), and wanted to just OC DDR and GPU, but the file is a bit different from other reference RK3288 dts files which I had for comparison. Really I couldn't find a guide which explained the process that clearly, especially since my dts has some different stuff in it.

After this I decided to try your overclockmatic. I got some different results and so I wasn't sure if I should proceed.

For instance, overclockmatic_rk3288.elf 3 gives only the result from the cfg file and does not read from the kernel.

Running it with input 5, it says that the dtb leaves for the CPU, DDR, and GPU are found and asks me if I would like to change them. Then either CPU locks not found until 0x2000000 or until 0x1000000.

Running with option 2, just reads the cfg. Same as 3, basically.


Option 1 errors out.

What do you think? Is it safe to run on the DTB leaves? Would you like for me to upload my kernel and resource images to take a look at?

Maybe you could tell me if it will be fine to just o.c. the RAM and GPU with overclockmatic if you saw the dts.

I am pretty sure that I uploaded my resource image in the restool thread, and possibly my dts as well, but give me a few minutes and I will upload them here (including kernel).

One thing which I noticed in kernel logs is that the preferred frequencies are set lower than the max. For instance max DDR maybe 425mhz, but the preference is set for 200mhz. Or gpu max is set at 500mhz but preference seems to be assigned to 350mhz. Something like that, I forgot the exact language. I can post a dmesg if needed.

No matter how hard I try to stress the machine I never seem to push it far enough to go above the preferred frequencies. I'm sure that they do it for battery life, but for me the settings seem a bit low. Yet I always score just under 40K in antutu, which is roughly the same as my R89, even though the clock speeds are set higher on the R89.

Perhaps those only apply when on battery. Maybe I'll do some more testing with the device on power and see if those freqs go up on DDR and GPU.

At least the Pipo kernel doesn't lie about the max frequency. The highest option I can set with 3c tools (and similar) is 1.6Ghz (even though the box clearly says 1.8Ghz!).

Give me a few minutes and I'll post it all up here.
Cheers and THANKS!
-SB

[sawbones999]

That was a LOOOONG few minutes :-)

Here is a link to my resource image, kernel, and dts/dtb (These are from the latest Pipo M9S kernel): https://dl.dropboxusercontent.com/u/...9/pipo_M9S.zip

If anyone has any ideas about it or anything, please post them! I don't know of any RK3288 tablets which have been successfully overclocked via patching or rebuilding resource image so this could potentially help out a lot of people! Let me know if a dmesg or something will help. I'm a little bit confused about this and could use some pointers.

Thanks and cheers!
-SB

[sawbones999]

KMSG

I went ahead and uploaded bc there are some errors here about about CPU freq. Also more frequency related info.

[Gormar]

KMSG

I went ahead and uploaded bc there are some errors here about about CPU freq. Also more frequency related info.

I will look into this but don't set yours hope high. I didn't discovered method to OC RK3288 SoC but only made tool to do it quickly.

[Gormar]

KMSG

I went ahead and uploaded bc there are some errors here about about CPU freq. Also more frequency related info.

I will look into this but don't set yours hope high. I haven't discovered method to OC RK3288 SoC but only made tool to do it quickly.

As for kernel if it is compressed then it wouldn't be possible to make universal tool to patch it on binary level. Decompressing it might work but I tried similar operation with RK3188 Wasser's kernel and failed. Probably because of CRC check. But AFAIK RK3288 kernel doesn't have one then maybe it is possible.

P.S. From quick look into dmesg.txt I can tell that it wouldn't help. Please make one just after devices boot and containing dvfs tree (cat /sys/dvfs/dfvs_tree).

P.S. 2. I've just realised that in Lollipop kernel there is GPU limit lock set to 480MHz .

<4>[21207.133158] |- voltage domain:vd_gpu
<4>[21207.133172] |- current voltage:900000
<4>[21207.133181] |- current regu_mode:UNKNOWN
<4>[21207.133193] |  |
<4>[21207.133193] |  |- power domain:pd_gpu, status = OFF, current volt = 900000, current regu_mode = UNKNOWN
<4>[21207.133210] |  |  |
<4>[21207.133210] |  |  |- clock: clk_gpu current: rate 178571, volt = 900000, enable_dvfs = ENABLE
<4>[21207.133228] |  |  |- clk limit(enable):[179000000, 480000000]; last set rate = 179000
<4>[21207.133238] |  |  |  |- freq = 179000, volt = 900000
<4>[21121.760466] |  |  |  |- freq = 179000, volt = 900000
<4>[21121.760475] |  |  |  |- freq = 297000, volt = 950000
<4>[21121.760484] |  |  |  |- freq = 417000, volt = 1100000
<4>[21121.760493] |  |  |  |- freq = 480000, volt = 1150000
<4>[21121.760502] |  |  |  |- freq = 480000, volt = 1250000
<4>[21121.760513] |  |  |- clock: clk_gpu current: rate 178571, regu_mode = UNKNOWN, regu_mode_en = 0

[sawbones999]

Finally got it attached. It was because I was using the plugin 'phony' for firefox for Android which is for changing the user agent. I had it set to IE

Thanks again!
-SB