ChanServ changed the topic of #linux-rockchip to: Rockchip development discussion |
hipboi has joined #linux-rockchip
hipboi has quit [Ping timeout: 248 seconds]
hipboi has joined #linux-rockchip
Astralix has joined #linux-rockchip
Astralix1 has quit [Ping timeout: 248 seconds]
<Tsvetan> ausxxh_ according to the datasheet brief its -10 +85C
Astralix1 has joined #linux-rockchip
hipboi has quit [Ping timeout: 268 seconds]
<Astralix1> good morning
<hramrach__> good morning
<Astralix1> any news in here?
<hramrach__> by sane environment I mean linux distribution which gives you full access to the device. not some crippled andriod
<Astralix1> ??
<hramrach__> if the bootloader is closed source that's not nice but so long as it works that's ok
<Tsvetan> android is linux running dalvik :)
<Astralix1> right
<hramrach__> and usually hiding most of hte system from you
<Astralix1> bootloader does load linux kernel, regardless what flavour it is
hipboi has joined #linux-rockchip
<Astralix1> good morning
<hramrach__> morning :)
<Astralix1> no, I meant hipboy :)
<hramrach__> I know you did :)
<Astralix1> guess we already had that ... I mean I am getting older but I hope it ins't that far already
<hramrach__> the problem with android is it's not self-hosted
<hramrach__> opening a file is second class operation on nadriod. no tools for that in base system. if that does not work nobody minds, rigt? - wrong
<Astralix1> if you have an problem with android, why don't u use ubuntu touch, picuntu or any other ARM based dist?
<hramrach__> the sticks come with andriod
<Astralix1> On a RK3188 stick it looks like an Ubuntu ARM (Desktop) runs really smooth
<hramrach__> so the question is how do you bootstrap something sane
<hramrach__> from what I saw the answer is scattered among numerous forum threads
<Astralix1> Yes they come with Android, but it's the same as with my Linksys routers... The original firmare survived less than 2 min after unpacking
<hramrach__> there is like 40 page thread and the first post saying how to do it is maybe in hte middle and maybe ther are some updates and corrections in the remaining pages or elsewhere ..
<Astralix1> 1) Call the windows tool for flashing and run erase IDB
<Astralix1> 2) flash latest bootloader
<Astralix1> 3) flash a kernel having rootfs set up for SD-Card
<Astralix1> 4) put rootfs on (fast) sdcard
<Astralix1> 5) have fun
<Astralix1> lol
<hramrach__> can you flash without running windows?
<Astralix1> yes
<Astralix1> you can take (my) rkflashtool
<hramrach__> nice
<hramrach__> I guess if olimex picks up rk we will finally have some nice board rather than tons of lame sticks
<Astralix1> The sticks are pretty fine, but I'd like to do a mainline support. So I need as many available pins as possible to verify functionality of any component
<Astralix1> for hobby use, the sticks are cool enough
<Astralix1> On the other hand, bringing kernel mainline is hobby for me too...
<Tsvetan> sticks do not use all features of the chip
<hramrach__> if there are more pins it means there could be more stuff
<Astralix1> May be I like boards where I can attach some probes and things
<hramrach__> it depends on the features that are unused and available if the board is worth it
<Tsvetan> not only the GPIOs but also Ethernet, I2S, SPI, Camera IF
<hramrach__> ethernet sounds good
<hramrach__> possibly more USB ports too
<Tsvetan> high speed ADC 10 bit is also in the specs with MSPS
<Astralix1> Yes but for watching some youtube and writing some email, WiFi, BT, HDMI, USB that's it
<hramrach__> I use wires
<hramrach__> reliable
<Astralix1> no, only if I do HF measurements, the I disable all readios
<Tsvetan> RK3188 + Camera chip + OpenCV could be great for hobby robot apps
<Astralix1> That is true
<Astralix1> but then use 2 cams
<Astralix1> for stereo 3D object recognition or separate near field / far recognition
<Tsvetan> it have plenty of power to do this
<Astralix1> I am thinking mor on the small little helper part. I'd like to have WiFi ac, USB3.0 and / or SATA ports with it
<Astralix1> A small 20W system doing file/print server and WiFi Router
<Astralix1> like the black knight from asus but with high speed hdd access
<Tsvetan> RK3188 have no SATA
<Astralix1> Funny... I was searching for it the whole night....
<Astralix1> I try again :)
<Astralix1> If I have a wish for a component that olimex should add to the baseboard, it is a SATA or USB3 controller
<Astralix1> USB2.0 is too slow and a memory coupled USB3.0 controller is faster. That would make it an option to bunde a small PCB to a HDD/SSD and have your linux box always going with you.
hipboi has quit [Remote host closed the connection]
<Tsvetan> A20 seems better in this regard as have SATA and Gigabit Ethernet
<Astralix1> Is there a board featuring these options?
<hramrach__> it's in the works
<hramrach__> but the driver for the ethernet aw put in the soc is somewhat broken
<hramrach__> so not sure you could get ane decent speed ou of that, anyway
<Astralix1> Are you olimex too?
<hramrach__> no affliation whatsoever
<Astralix1> you mean the eth IP is broken or the software driver around it?
<hramrach__> you can't really tell
<hramrach__> it's probably sw issue but without docs you cannot tell how it is *supposed* to work
<hramrach__> people working on hte dirver say that there is probably some hw buffer overrun which is not noticed by the driver
<hramrach__> tuning the driver results in good spped either for local connections or remote (larger latency) connections
<hramrach__> but not both
<Astralix1> Ah, ok. I had that one in my former life... Guess it was a davicom chip that went nuts once in a while without reason...
<hramrach__> somebody suspected that the ehternet might be davicom, actually. the origin is still unknown afaik
<hramrach__> if you want to know all the details there are #linux-sunxi logs somewhere. they probably have the location in the topic
<Astralix1> lol
<Astralix1> I'll read that later. I am at work and have to do other things. If you start crawling code... you know... you can really get lost in it
<hramrach__> whitequark too :)
<hramrach__> good luck crawling :)
<Astralix1> tonight after RK3188 mainline kernel timer system is working
<hramrach__> I guess there will be good incentive to look into it once cubietruck is out so somebody likely will ;-)
<Astralix1> ??
<hramrach__> the board is *made* to work as NAS so there will be people bitching if it does not
<Astralix1> Yes, for sure
<hramrach__> and at least trying to debug it
<Astralix1> in this case I'd prefer to have 4 SATA cons
<hramrach__> I use the ethernet for git push/pull which is abysmally slow with any card
<hramrach__> yes, or at least PMP support in the ahci
<Astralix1> yes that would be fine
<Astralix1> Is the SATA IP known?
<hramrach__> but not present :/
<hramrach__> not sure
<hramrach__> but does not do PMP
mmind00 has joined #linux-rockchip
ausxxh_ has quit [*.net *.split]
mmind00 has quit [*.net *.split]
Tsvetan has quit [*.net *.split]
mmind00 has joined #linux-rockchip
Tsvetan has joined #linux-rockchip
ausxxh_ has joined #linux-rockchip
<hramrach__> anybody did the memory benchmarks on rk?
<Astralix1> nope, not yet
<Astralix1> on the rk kernel I didn't do it and on the mainline kernel there is no scaling support for now. Just use everything as it is set up by the bootloader.
<Astralix1> would not result in a fair comparision
<Tsvetan> Astralix how the plls are initialized in RK3188 without documentation?
<Tsvetan> hramrach__ the link which you send is with DDR3 set to 480Mhz which is not safe for A20
<mmind00> Tsvetan ... the bootloader initialises them to a working but slow state
<Tsvetan> there is no document which to state safe DDR3 settings, but we found from manufacturing statistic that 80% of A20 processors we got could work on 480Mhz but 20% fail and need to be initialized to 380Mhz
<Tsvetan> actually even these which work on 480Mhz is not sure if they will not freeze after 5-10 hours of work
<Tsvetan> we just cant test so long the boards during the production
<Tsvetan> speaking with Tom he observe same with his boards, the problem is exactly in A20 as A10 works on 480Mhz safely
<Tsvetan> so if you want to be on safe settings your DDR3 should be 380Mhz for A20
<Tsvetan> if you want to risk no problem to clock to 480 Mhz in most of the cases it may work
<Tsvetan> this is the fun to work with non documented SoCs :-) you find the parameters in practice
<Astralix1> Oh, a catalouge of questions :)
<Astralix1> Tsvetan, the clock code of rk3188 is available so reverse engineering of the code is an option. But even with the TRM at hand you have no luck, als clocks are part of PMU and CRU which have separate manuals.
<Astralix1> Tsvetan, the DDR speed is limited by DDR I/F of CPU, DDR memory capability and PCB layout.
<Astralix1> And then it makes difference if you use reliable DDR memory with guaranteed specs or parts from over-production stocks
<Astralix1> On some android stick I bet there are re-labeled DDR chips that failed tests somewhere lese but is you keep them 30% below specs they work fine.
<Astralix1> mmind00? Have we lost the other chat server?
<hramrach__> Tsvetan: it's in the same ballpark with 320MHz
<hramrach__> and safe is relative
<hramrach__> there are no hard specs for the chip. it's chinese :/
<Astralix1> lol
<hramrach__> eihter way if the rk is in the same ballpark the extra cores can't help that much
<hramrach__> a10 is very constrained by memory bandwidth and a20 does not have much more. it's subject to tuning still but meh
<mmind00> Astralix why?
<Astralix1> cannot access it anymore
Astralix1 has quit [Quit: Leaving.]
Astralix1 has joined #linux-rockchip
<mmind00> yep, seems to be dead
<hramrach__> Astralix1: the memory runs way below spec even on 480MHz. the issue is in the memory controller and PCB design
<Astralix1> Yes, the bottleneck is the controller side
<Astralix1> but I think all components have to play together
<mmind00> according to the rk3066 trm the supported datarates are 400mhz for ddr3/lpddr2 and 200mhz for lpddr
<mmind00> will probably be the same for the 3188
<hramrach__> the question is what is the actual speed
<hramrach__> while the ram chip has the same speed as on a10/a20 theoretically the memory controller and chache management in the SoC can affect the user available speed a lot
<hramrach__> also what is the bus width used to acces the ram? it differs on different SoCs
<Astralix1> afaik the rk3188 is 533mhz for ddr3/lpddr2 so it has good chance to win against 3066
<hramrach__> eg. the Chromebook with Exynos 5k is about twice as fast as an Atom 450 or Core2 while those a10/a20 are about half the speed of those
<mmind00> Exynos5 also is a Cortex-A15
<mmind00> for which ARM claims a speed improvement of 40% over a same-clocked A9
<Astralix1> And you don't know how optimized the tollchain is for the system
<hramrach__> it's probably hand-crafted assembly most of the time
<Astralix1> we had 10..20% speed improvement by just recompiling a kernel for RK2918
<hramrach__> not memcpy improvement I would guess
<Astralix1> instead of using ARM gcc 4.4.3 we used ARM-SIMD Cortex A8 optimized gcc 4.7.3
<hramrach__> faster syscalls are neat but this is pure userspace benchmark
<Astralix1> The benchamrk was done on android layer by playing and watching videos and such
<hramrach__> video players tend to do a lot of kernle work, yes
<hramrach__> some benchmarks show eg. tons of time spent in gettimeofday()
<Astralix1> lol
<hramrach__> it's cheap on x86 but expensive on arm
<hramrach__> because the hw counter is inside and outside of the CPU core, respectively
<Tsvetan> Astralix_ same PCB layout and memory chips are used for A10 and A20
<Tsvetan> so this is not related to DDR3 memory or PCB layout
<hramrach__> yes, the a20 did not come out that well. the max stable cpu clock is lower than a10 as well
<Astralix1> ok, that there are tolerances somwhere else :)
* hramrach__ is starting to like the chromebook
<Astralix1> Yep, thought about buying one and push Ubuntu desktop onto it
<hramrach__> I should benchmark the AMD e-450 board as well
<hramrach__> would be good comparison
hipboi has joined #linux-rockchip
hipboi has quit [Quit: Leaving]
Astralix1 has quit [Quit: Leaving.]
mmind00 has quit [Quit: KVIrc 4.1.3 Equilibrium]
julien_ has joined #linux-rockchip
julien_ is now known as Guest94853
Guest94853 has quit [Quit: Page closed]