No performance difference between 1.5, 1.75 & 2GHz

Moderators: mdrjr, odroid

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby tchiwam » Sun Aug 28, 2016 9:56 am

mlinuxguy wrote:It occurred to me that the default clock sits at the fake "2ghz" meaning its being driven by higher voltages than necessary
I wonder if this translates into any power savings when they reduce the default clock table to max 1.5ghz (or whatever it ends up being).

An easy test would be for us to measure the power consumption of an idle C2 with the frequency set to "2ghz"
# cpufreq-set -u 2016000

# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq
2016000
# cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq
2016000

Measure the power draw of the C2
Keep in mind its being over-driven by higher voltages than necessary for the real 1.5ghz

Redo the power measurement after setting it down to 1.5ghz
# cpufreq-set -u 1536000

Its possible we would be saving a bit of power by driving the CPU at the proper voltages for its real 1.5ghz
It should idle cooler too


Making lemonade out of lemons ... Neat idea !
Anyway even at 1.6GHz the benchmarks will only get better
tchiwam
 
Posts: 83
Joined: Wed Dec 30, 2015 4:53 am
Location: Greenland
languages_spoken: english, French, Finnish, Greenlandic(little)
ODROIDs: 5x XU4, 3x C2, 2 cloudshell, 1 VU7, 1 VU7+, UPS1, UPS2, UPS3, USB/SATA, and many other fine bits from Odroid

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby brad » Sun Aug 28, 2016 10:17 am

odroid wrote:Please don't waste your valuable time.
All the Coretex-A53 based Amlogic SoCs have the same issue. S905, S905 Rev-C(S905H), S905X and even S912.
The kernel clock frequency 1.75 and 2Ghz do not exist in the BL3x blobs.
It must be Amlogic's fault. But we should detect it earlier in our internal development stage. Really sorry about that mistake.

We've been modifying/testing the BL3x code to find a maximum feasible/stable clocks like 1.53Ghz, 1.58Ghz, 1.61Ghz, 1.65Ghz, 1.68Ghz, 1.71Ghz, etc for a couple of weeks.
But 1.65Ghz seems to be the maximum one due to some power and heat issues. We will run the burnA53 on Ubuntu and the Stress app on Android with 10 set of C2 boards in this weekend (for 48 hours stability test) to make sure it.
Once we will have a test result, we will report it on this thread with update package for further sampling test in early next week.

If we disable two cores, the SoC can run up to 1.75Ghz probably. But we need further stability test too.
And yes. I fully agree most people don't like this approach.

Anyway, we know well 1.65Ghz or 1.75Ghz is still very far from the 2Ghz. So we will change the C2 specification in our home page next week as per the test result.
And we must consider some compensation for C2 users. But we need to check what we can do first.

Please accept my sincere apologies for the mistake and kindly understand our situation.


Thanks for your honesty and information on this matter, im very interested in seeing what HK and AML can offer

As it appears 2GHz is out of the question are hardkernel or Amlogic able to help with any of the following to improve performance of processing the board...

- Licence the arm64 crypto extensions
- Provide a driver for the amlogic crypto module
- Create a new governor / power module / policy to shutdown cpu's when not required and up frequency on remaining cpu's. (Will look into this myself maybe)
- Offer a larger heatsink (possibly a fan) for the device.

Im also interested in the following information if are able to provide...

- How much thermal load does mali require and what clock speeds are they running (750MHz I believe) at?
- How accurate are the rests of the clocks, ie what frequencies are the caches, ethernet, memory and USB running at?
- What appears to be the upper limit, 1) to much thermal load or 2) not enough power being able to be supplied to the cores or maybe even the cache?

rooted wrote:What decrease in performance?

The news that the CPU's have been locked at 1.5GHz, this is a big performance decrease over the 2GHz reported in the advertising specifications. Thats 500MHz per CPU and 2Ghz in total across the 4 CPU's or 25% of the boards potential processing power. This is currently a hard limit set in the firmware so no amount of thermal cooling was going to help, I really dont see how you can disagree with other users concerns about this performance issue, this really pisses me off as the facts are solid.

Best Regards,

Brad.
brad
 
Posts: 718
Joined: Tue Mar 29, 2016 1:22 pm
Location: Australia
languages_spoken: english
ODROIDs: C2 N1

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby endecotp » Sun Aug 28, 2016 2:21 pm

Create a new governor / power module / policy to shutdown cpu's when not required and up frequency on remaining cpu's


Isn't that what the hotplug governor already does?
Or do you mean "up" the frequency above 1.65GHz?
endecotp
 
Posts: 230
Joined: Sun May 08, 2016 12:30 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby fvolk » Sun Aug 28, 2016 3:19 pm

odroid wrote:Please accept my sincere apologies for the mistake and kindly understand our situation.


Thank for the open communication and keeping us up to date.

Personally I don't care whether it's 1,5GHz or 2,0GHz, I prefer stable+fanless operation at load, with a reasonable safety margin so the C2 can run 24/7 unattended without fear of overheating or melting the case.

However, I don't like false advertising. As a customer I don't care who screwed up in the chain and I understand sending all C2s back for a full refund or reworking does not make sense financially. However I do expect that you develop a great C3 and everyone who bought their C2 at an official distributor gets a voucher per C2 bought and they then can be redeemed to buy any HK product at a reduced price.
fvolk
 
Posts: 181
Joined: Sun Jun 05, 2016 11:04 pm
languages_spoken: english
ODROIDs: C2, HC1

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby Raybuntu » Sun Aug 28, 2016 6:41 pm

The C2 is an awesome device and complaining it can't do 2ghz is just absurd. Although I think that Amlogic is just cheating. They certainly new about that issue but apperantly they went ahead to get themself some good advertising. For me it doesn't matter.

I'm still gonna buy those devices because they are the best for low powered Kodi Mediacenters.

Whats really bothering me is that those chinese piracy android TV box sellers are gonna keep advertising their boxes with 2ghz and people are gonna buy them.
BTC: 12zLUYC7JzwM7a8cQKekAvZr9kWxjzTfxm
Raybuntu
 
Posts: 1245
Joined: Mon Nov 30, 2015 4:23 pm
languages_spoken: english, german
ODROIDs: C1+, C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby memeka » Sun Aug 28, 2016 7:52 pm

@brad

it's not a performance decrease, because there has never been a true 2Ghz S905 to compare it with. you can't decrease something that was never measured in the beginning.
your board is performing with the same performance as yesterday, as when it was announced, and it has the same performance as in all the benchmarks until now. all the benchmarks/scores were not faked, that's the level of performance you have now. so it's still X% faster than rpi2/3 on one hand, but on the other hand in the @cnx-software benchmarks there were discussions a while ago why single-core performance was marginally better than a 1.2 GHz rk core - now we know :)

of course, what this is however is false advertising from amlogic. but my point is - it's not a DROP in performance, <technically speaking> :)
User avatar
memeka
 
Posts: 4055
Joined: Mon May 20, 2013 10:22 am
languages_spoken: english
ODROIDs: XU rev2 + eMMC + UART
U3 + eMMC + IO Shield + UART

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby graysky » Sun Aug 28, 2016 10:02 pm

So max rate aside.. why are the benchmark results non-linear for the C2 whereas the x86_64 machine is linear?
ODROID:
Image

E5620 (x86_64):
Image
User avatar
graysky
 
Posts: 122
Joined: Wed Jul 13, 2016 9:11 am
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby tchiwam » Sun Aug 28, 2016 11:50 pm

graysky wrote:So max rate aside.. why are the benchmark results non-linear for the C2 whereas the x86_64 machine is linear?
ODROID:
Image

E5620 (x86_64):
Image


Maybe we need to benchmark the caches and memory BW at different settings here. I've seen this before.
tchiwam
 
Posts: 83
Joined: Wed Dec 30, 2015 4:53 am
Location: Greenland
languages_spoken: english, French, Finnish, Greenlandic(little)
ODROIDs: 5x XU4, 3x C2, 2 cloudshell, 1 VU7, 1 VU7+, UPS1, UPS2, UPS3, USB/SATA, and many other fine bits from Odroid

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby endecotp » Mon Aug 29, 2016 1:06 am

Raybuntu wrote:complaining it can't do 2ghz is just absurd..


I really don't understand this attitude.

Looking at all the comparable devices (e.g. Hackerboards' list of 81 SBCs), the stand-out feature of the C2 is that it claimed to have a quad-core, 64-bit, 2 GHz processor. Nothing else came close to that; it was head-and-shoulders above the others. A 1.6 GHz device is much less impressive; it's still good, but it's much closer to the competitors. Of course it's no slower than a month ago, but that's not the point as few of use did side-by-side comparisons to make our choices; we trusted the published specifications. That's why I find it really disappointing that the web sites are still saying 2 GHz.
endecotp
 
Posts: 230
Joined: Sun May 08, 2016 12:30 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby endecotp » Mon Aug 29, 2016 1:11 am

graysky wrote:So max rate aside.. why are the benchmark results non-linear for the C2 whereas the x86_64 machine is linear?
ODROID:
Image

E5620 (x86_64):
Image


Try plotting those results so that both axes always start at zero. (Currently you're dojng this for the ARM graph but not for the x86.). I think they will look more similar. Does you "best fit" tool have a mode for a reciprocal, i.e. y = A/x ?
endecotp
 
Posts: 230
Joined: Sun May 08, 2016 12:30 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby graysky » Mon Aug 29, 2016 1:19 am

tchiwam wrote:Maybe we need to benchmark the caches and memory BW at different settings here. I've seen this before.

I don't think that caches are at play since my data show the same phenotype as the OP's data (his was an orthogonal benchmark).

endecotp wrote:Try plotting those results so that both axes always start at zero. (Currently you're dojng this for the ARM graph but not for the x86.). I think they will look more similar. Does you "best fit" tool have a mode for a reciprocal, i.e. y = A/x ?


Here are the data for you to play with... each is an average of 3 runs.

ODROID-C2:
Code: Select all
Time (s)   Freq (Mhz)
17.08   2016
17.1   1752
16.56   1536
19.30   1296
24.47   1000
43.35   500
82.785   250
201.985   100


E5620:
Code: Select all
Time (s)   Freq (Mhz)
24.03   2395
27.41   2261
29.00   2128
30.38   1995
32.03   1862
34.23   1729
36.71   1596
Last edited by graysky on Mon Aug 29, 2016 2:56 am, edited 1 time in total.
User avatar
graysky
 
Posts: 122
Joined: Wed Jul 13, 2016 9:11 am
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby goatman » Mon Aug 29, 2016 2:06 am

It's not understandable why there is such upset from the "sudden revelation" that the S905 can't operate continuiously at 2GHz. It states in the SECOND PARAGRAPH of the datasheet that the SOC can be *overdriven* (does the Intel trademark Turbo Boost sound familiar?) beyond the nominal operating speed. The problem here is that the CPU stepping algorithm was never implemented properly by developers. HK needs to give the end user total access to CPU control and now. I'm sure the community can come up with something sane.

As far as misrepresentation, it is a decision usually made using lawyers in court in a class action lawsuit, although in some states the government can be petitioned to act on the behalf of the consumer. It seems pretty clear though that Amalogic made an effort to disclose the actual limits of the hardware by using the word overdriven.


Goatman
User avatar
goatman
 
Posts: 17
Joined: Wed Jun 29, 2016 11:12 am
languages_spoken: German, English
ODROIDs: ODroid C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby willmore » Mon Aug 29, 2016 2:43 am

graysky wrote:Here are the data for you to play with... each is an average of 3 runs.

ODROID-C2:
Code: Select all
Time (s)   Freq (Mhz)
17.08   2016
17.1   1752
16.56   1536
19.30   1296
24.47   1000
43.35   500
82.785   250
201.985   100


E5620:
Code: Select all
Time (s)   Freq (Mhz)
2394   24.032
2261   27.408
2128   29.00
1995   30.38
1862   32.03
1729   34.23
1596   36.71


I'm pretty sure you swapped colums for the E5620 data. Try: calculate the reciprical of the time. That gives you throughput. One would expect throughput to be linear with frequency. Then you'll see the issue very quickly. I did a quick plot here on OO and it stuck out very clearly.
willmore
 
Posts: 10
Joined: Mon Aug 29, 2016 2:37 am
languages_spoken: english
ODROIDs: C1 x1
C2 x2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby graysky » Mon Aug 29, 2016 3:01 am

Good suggestion plotting the Freq vs 1/sec which does look much better and picks off the top 3 freqs are highly similar (you were also right about the column swap which I corrected in the post):

Intel:
Image

ODROID-C2:
Image
User avatar
graysky
 
Posts: 122
Joined: Wed Jul 13, 2016 9:11 am
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby willmore » Mon Aug 29, 2016 3:44 am

graysky wrote:Good suggestion plotting the Freq vs 1/sec which does look much better and picks off the top 3 freqs are highly similar (you were also right about the column swap which I corrected in the post):


Much better. You can see the 'brick wall' the C2 runs into at 1.5GHz.

Am I reading this right? Did the C2 (even at 1GHz) outperform the Xeon at 2.4GHz?
willmore
 
Posts: 10
Joined: Mon Aug 29, 2016 2:37 am
languages_spoken: english
ODROIDs: C1 x1
C2 x2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby graysky » Mon Aug 29, 2016 3:46 am

willmore wrote:Am I reading this right? Did the C2 (even at 1GHz) outperform the Xeon at 2.4GHz?


No, completely different data for the zip test.
User avatar
graysky
 
Posts: 122
Joined: Wed Jul 13, 2016 9:11 am
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby willmore » Mon Aug 29, 2016 3:51 am

graysky wrote:
willmore wrote:Am I reading this right? Did the C2 (even at 1GHz) outperform the Xeon at 2.4GHz?


No, completely different data for the zip test.


Okay, I was going to be completely freaked out if that had been the case. ;)
willmore
 
Posts: 10
Joined: Mon Aug 29, 2016 2:37 am
languages_spoken: english
ODROIDs: C1 x1
C2 x2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby gdachs » Mon Aug 29, 2016 7:17 am

endecotp wrote:
Raybuntu wrote:complaining it can't do 2ghz is just absurd..


I really don't understand this attitude.

Looking at all the comparable devices (e.g. Hackerboards' list of 81 SBCs), the stand-out feature of the C2 is that it claimed to have a quad-core, 64-bit, 2 GHz processor. Nothing else came close to that; it was head-and-shoulders above the others. A 1.6 GHz device is much less impressive; it's still good, but it's much closer to the competitors.

I don't understand why you don't understand raybuntu. He clearly stated:
Raybuntu wrote:I'm still gonna buy those devices because they are the best for low powered Kodi Mediacenters.

For this kind of usage you will not notice that it can't do 2 GHz.

I totally disagree that the stand-out feature of the C2 is the 2 GHz processor. The stand-out features are the kernel hackers paid by Hardkernel and the contributors that are fixing the bugs of the kernel provided by Amlogic. Without them you would have no joy with the other SBCs as they all just copy the Hardkernel patches.

Okay, if you are doing nothing else with your C2 than just running benchmarks all the day, then it is a shock that there is no 2 GHz. But I am very sure that all the people currently complaining will never have a setup that would max out the CPU.

Gerald
gdachs
 
Posts: 46
Joined: Fri Dec 26, 2014 7:58 am
languages_spoken: english, german
ODROIDs: Odroid-C1 hopefully soon

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby koschi » Mon Aug 29, 2016 7:37 am

While some users might not max out the CPU or would benefit from another 4x500MHz, there are users who do. Different people, different use cases.

As I wrote in my initial post, I am using my 8 C2s for distributed computing. In our team we have some members with dozens of C2s, XU4 and older boards. Of course we want to fully utilize each board and care mostly for the CPU performance before other features.
Other people might use their fleet of C2s for distributed rendering, compiling, building hash tables, etc... They also care for CPU performance.
When only little or no benchmarks between two A53 based boards available, the outstanding 2GHz CPU becomes a strong selling point for those primarily looking for CPU power.

While I do appreciate the HK software support, our use case usually doesn't require sophisticated functions, HW, GPU or codec support, etc. We install the board, install BOINC and then remote control the BOINC agent, without messing around much with the SBC. We do care for CPU performance though.
koschi
 
Posts: 15
Joined: Tue Mar 22, 2016 4:32 am
languages_spoken: english
ODROIDs: 8 x C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Mon Aug 29, 2016 7:54 am

For me HK provides a service for their boards that is impressive enough that I keep coming back to get more
This service includes fixing the manufacturer's kernel, responding to issues, openly discussing problems, etc...

I have purchased many other vendors boards, typically complaints about defects or not living up to specifications fall on deaf ears. HK listens and
responds, often fixing the issue and where they can't discussing why (thinking of the USB cpu usage issue among others).

I don't like the ancient kernels (not their fault). However, I would like more transparency into the binary blobs, as for me anyway, the fun with these
boards is making them do things they weren't designed to do (like the old PC hacking days of the 80's). I want to overclock a beta board because it
exposes weaknesses in the design or implementation, not because I want a box faster than anyone elses.

I have an issue with the C2 beyond the artificially capped clock rate. I'm not convinced the ARM L2 cache is setup properly or perhaps working at all. I know there were
L2 cache patches put into their u-boot to set it up, but take a look at the linux kernel starting at 3.18 where all the ARM L2 cache patches go in.... Then imagine a benchmark on the C2 that gets to actually use a properly working L2 cache. If I was at ARM I would hate to see all the bad press or poor benchmarks engendered by bad implementations or out of date kernels.

Most of the Linux ARM dev boards available are side-effects of the Android first market. For years those vendors could get away with an ancient kernel because Android used it and that's what Linux gets to inherit. Perhaps that will change now with the newer Android kernels. Meanwhile we get to work with ancient kernels using hidden binary blobs and inflated specs that no one can easily double-check or fix.

ARM should take a page from the Google Nexus device approach
Every year a Android and a Linux dev board should get released that features a new ARM CPU (A72, A53, etc..) which implements basic ARM IP blocks and has full upstream kernel support. These become the benchmark devices of what works right and provides a base from which to implement custom boards.
In some sense this is what Linaro should do for the OS, as would 96boards for hardware, but so far I have not been impressed with the 96boards offerings (give me Sata or usb3.0 and 1gb networking under the $100 price point and I might get interested).
Last edited by mlinuxguy on Mon Aug 29, 2016 2:40 pm, edited 3 times in total.
mlinuxguy
 
Posts: 793
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, N1, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby afremont » Mon Aug 29, 2016 9:00 am

After reading some of these posts, I can't help but think about Edwin Armstrong and something he said about courtroom battles that he was forced into. He said that court was a place where people substituted words for facts and then argued about the words. There seems to be some of that going on here.

There's no question that the C2 is a great board. It absolutely blows my Pi 2 away running libreelec. Even when running Kodi on top of full Ubuntu, the program guide loads almost a full order of magnitude faster.

That said, it's a shame that it's being limited to 75% of it's rated clock speed for some, as yet, unknown reason. It doesn't seem that it's for thermal reasons, so maybe (and this is pure speculation) it's for reliability or stability reasons. It could be nothing more than a code error.

I get that there is no apparent "decrease in performance", since we don't seem to have any proof that it ever ran faster, the fact is that performance would be better with a faster clock.

Another fact is that the chipset is marketed as running at 2GHz. Turbo boost semantics aside, there appears to be no evidence of any type of boosting occurring at any time. To me, that rules out that argument.

Personally (and this is my opinion) hardkernel is not to blame, especially since they openly said they were going to change their marketing info right away now that they have confirmed that there seems to be a problem.

Is this going to ruin my day, no. Would I buy another one to use for a media front-end, yes because they still rock at the job. Am I happy, not exactly.

If I had the time and better skills, I'd reverse engineer the binary and see if the chip could be forced into real 2GHz or faster operation. Sooner or later, I believe that someone will do it anyway and that makes me wonder why amlogic only gives us a binary in the first place.

This is not all that unusual. Practically every 8 thru 32 bit microcontroller that I've tinkered with has issues of one kind or another when it comes to some aspect of the datasheet and marketing claims.




Sent from my Moto G (4) using Tapatalk
afremont
 
Posts: 128
Joined: Sun Mar 08, 2015 12:58 am
languages_spoken: english

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby endecotp » Mon Aug 29, 2016 11:38 am

gdachs wrote:I am very sure that all the people currently complaining will never have a setup that would max out the CPU.


What on earth do you mean? Anyone doing even something like kernel compilation will regularly "max out" the CPU.
endecotp
 
Posts: 230
Joined: Sun May 08, 2016 12:30 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby Ameridroid » Mon Aug 29, 2016 2:00 pm

HK informed all their distributors, including ameriDroid.com, to change their pages to read 1.5GHz for the C2. If they find a way to increase the speed rating, I'm sure we'll be able to increase the specs on our sites.

On another note, there has been some discussion in this thread about active coolers for the C2. in the event that active cooling is needed for the C2 (or C1/C1+) for any reason now or in the future, you may want to consider this cooling case made exclusively for ameriDroid.com: Image
http://ameridroid.com/products/odroid-c-series-case-with-fan
Ameridroid
 
Posts: 157
Joined: Thu Mar 17, 2016 4:19 am
languages_spoken: english

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby Raybuntu » Mon Aug 29, 2016 3:27 pm

endecotp wrote:
gdachs wrote:I am very sure that all the people currently complaining will never have a setup that would max out the CPU.


What on earth do you mean? Anyone doing even something like kernel compilation will regularly "max out" the CPU.


I'd never compile anything on ARM devices it's just too slow regardless of 2Ghz or whatever. Regarding my attitude: I find it absurd because 2 weeks ago most of the people complaining here didn't even notice or care.
It's just playing with Numbers and people love numbers. They compare this issue with VW's Dieselgate which is really annoying. Comparing such an irrelevant issue with years/decades of environmental pollution is just sick.
BTC: 12zLUYC7JzwM7a8cQKekAvZr9kWxjzTfxm
Raybuntu
 
Posts: 1245
Joined: Mon Nov 30, 2015 4:23 pm
languages_spoken: english, german
ODROIDs: C1+, C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby gdachs » Mon Aug 29, 2016 5:04 pm

endecotp wrote:
gdachs wrote:I am very sure that all the people currently complaining will never have a setup that would max out the CPU.


What on earth do you mean? Anyone doing even something like kernel compilation will regularly "max out" the CPU.

You have never heard from cross compiling?

Gerald
gdachs
 
Posts: 46
Joined: Fri Dec 26, 2014 7:58 am
languages_spoken: english, german
ODROIDs: Odroid-C1 hopefully soon

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby crashoverride » Mon Aug 29, 2016 5:35 pm

When the C1 came out, it was fast enough for me to start doing development and compilation on the device itself. When I got a C2, I moved entirely to on-device compilation. I compile the kernel A LOT and the bottleneck comes down more to amount of RAM available and disk I/O speed from what I observed. The change from 1GB in C1 to 2GB in C2 had substantially more impact than a change of only 500Mhz (1.5Ghz to 2Ghz) would. So as stated before, this issue really does not impact me.

On a constructive note, has anyone looked at the datasheet to see if the registers required to change clock frequency and voltages are documented? Although the kernel currently calls into service mode to perform the change, an alternate driver could be provided that performs the operation itself if the registers are available. The service call is preferred since it prevents damage to the device and the resulting RMA request.
crashoverride
 
Posts: 4021
Joined: Tue Dec 30, 2014 8:42 pm
languages_spoken: english
ODROIDs: C1

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby odroid » Mon Aug 29, 2016 7:27 pm

Our test results so far.
1.65Ghz on 4-cores burning test has passed on Ubuntu and Android.
1.75Ghz on 3-cores burning test has passed too.

We are making more detail test reports with How-to guide for other people's testing the new boot blobs and kernel source patch for above two configurations.
It will be released around 20 hours later because we want to perform some benchmarking tests now.

Once we collect various feedbacks from some advanced users for a week, we will make a kernel/bootloader update package with new boot.ini file for easier & simpler clock/core options configuration.
User avatar
odroid
Site Admin
 
Posts: 28859
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby Raybuntu » Mon Aug 29, 2016 7:56 pm

@odroid: Can you disclose more details on that issue? You were saying it is not only an thermal issue!? Why did Amlogic block higher frequencies secretly in bl3 blob?
BTC: 12zLUYC7JzwM7a8cQKekAvZr9kWxjzTfxm
Raybuntu
 
Posts: 1245
Joined: Mon Nov 30, 2015 4:23 pm
languages_spoken: english, german
ODROIDs: C1+, C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby afremont » Mon Aug 29, 2016 8:13 pm

If they can clock 3 cores to 1.75, it seems possible that this is an internal power problem if it's not thermal. If that's the case, only a die revision is going to fix that.

Sent from my Moto G (4) using Tapatalk
afremont
 
Posts: 128
Joined: Sun Mar 08, 2015 12:58 am
languages_spoken: english

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby memeka » Mon Aug 29, 2016 8:23 pm

I agree with crashiverride
Compiling with 8 cores (xu4) burns through ram like hell, sometimes I have to use 4 cores cause the little ones don't add a lot of value but use much ram.
User avatar
memeka
 
Posts: 4055
Joined: Mon May 20, 2013 10:22 am
languages_spoken: english
ODROIDs: XU rev2 + eMMC + UART
U3 + eMMC + IO Shield + UART

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby brad » Mon Aug 29, 2016 10:16 pm

odroid wrote:Our test results so far.
1.65Ghz on 4-cores burning test has passed on Ubuntu and Android.
1.75Ghz on 3-cores burning test has passed too.


Thank you odroid for the update, any better results on 1 & 2 cpu's?
brad
 
Posts: 718
Joined: Tue Mar 29, 2016 1:22 pm
Location: Australia
languages_spoken: english
ODROIDs: C2 N1

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby brad » Mon Aug 29, 2016 10:30 pm

memeka wrote:@brad
of course, what this is however is false advertising from amlogic. but my point is - it's not a DROP in performance, <technically speaking> :)


Technically speaking or speaking out of an arse its the same thing, it dont do 2GHz and never has. I hope it will still!

Drop in performance or performing under what it is specified. Same thing dont give me an English lesson on that one.
brad
 
Posts: 718
Joined: Tue Mar 29, 2016 1:22 pm
Location: Australia
languages_spoken: english
ODROIDs: C2 N1

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby endecotp » Tue Aug 30, 2016 12:12 am

Raybuntu wrote:I'd never compile anything on ARM devices it's just too slow regardless of 2Ghz or whatever.


OK, that starts to explain the different attitudes.

I've replaced my old x86 desktop box with my ODROID-C2. I'm using it as my main computer. I don't find it "too slow" to compile most things, but perhaps that's because I remember how long it took to build gcc 1.x on a SPARCstation IPC, with its 25 MHz processor and 16 MB of RAM.
endecotp
 
Posts: 230
Joined: Sun May 08, 2016 12:30 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Tue Aug 30, 2016 12:34 am

crashoverride wrote:On a constructive note, has anyone looked at the datasheet to see if the registers required to change clock frequency and voltages are documented? Although the kernel currently calls into service mode to perform the change, an alternate driver could be provided that performs the operation itself if the registers are available. The service call is preferred since it prevents damage to the device and the resulting RMA request.

Been there done that, ain't documented and probably not reachable anyway.

Analogy: Imagine the embedded M3 controller as a virtual machine host that controls the clock frequency of the virtualized instance. It would violate their "Trustzone" isolation to allow the virtual instance access to those control knobs.

I attempted to do something like this way back in Feb 2016 in my overclocking thread: viewtopic.php?f=139&t=18738
I discuss using scpi to attempt to program in a new volt/freq combo using this struct:
Code: Select all
struct scpi_opp_entry {
    u32 freq_hz;
    u32 volt_mv;
} __packed;

The M3 just ignored my entry since its not built into its internal table, and "best of all" Linux just reported itself running at my newly programmed in higher clock.
Perhaps there is some service mode switch you can do while inside uboot before transitioning to the last protected mode level. My bet is any docs on that are hidden deep within NDA and need to know basis by ARM licensees. Mere mortals and opensource need not guess those workings.

On a side note
ARM really missed an opportunity to use the M3 controller as a "Integrated remote console" with reporting the status of its main processor and its health along with control of the power state of the host processor. But who needs that for smartphones eh?
Example would be the processor appears hung, connect to the M3 processor and you can see the power draw and whatever else the M3 might know about the current state of the main processor. You could then reset it via the M3.
mlinuxguy
 
Posts: 793
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, N1, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby crashoverride » Tue Aug 30, 2016 12:58 am

For TrustZone there is typically not an additional controller. It runs on the boot CPU just like Intel/AMD System Management Mode transparent to any OS. I thought I recalled seeing PLL and clock register controls for the CPUs in the datasheet, but I may be mistaken. Overclocking and/or running out of specification is not of interest to me so I have not explored in any detail nor do I intend to. Its not a philosophical premise; rather, its simply that as a developer I require system stability over performance when debugging.
crashoverride
 
Posts: 4021
Joined: Tue Dec 30, 2014 8:42 pm
languages_spoken: english
ODROIDs: C1

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Tue Aug 30, 2016 1:08 am

odroid wrote:Our test results so far.
1.65Ghz on 4-cores burning test has passed on Ubuntu and Android.
1.75Ghz on 3-cores burning test has passed too.

We are making more detail test reports with How-to guide for other people's testing the new boot blobs and kernel source patch for above two configurations.
It will be released around 20 hours later because we want to perform some benchmarking tests now.

Once we collect various feedbacks from some advanced users for a week, we will make a kernel/bootloader update package with new boot.ini file for easier & simpler clock/core options configuration.


Better than expected clocks
It would be interesting to verify if you are hitting thermal limits or DVFS power to the cores.
I would look for hot-spots on the CPU die with a thermal imager.
If you recall my post on CPU temperatures the thermal shim you supply on the default heatsink was nearly 10C warmer with it on versus using a high-performance heatsink compound. It was 10C cooler with TIMS removed: viewtopic.php?f=139&t=18739#p130851
If you have a hotspot you need something better than TIMS to disperse the heat away fast.

Not sure how you determine if its DVFS power to the CPU. As far as I can tell that is up to the CPU licensee to implement properly
(though I could be wrong and its a ARM IP block).

However some of us don't use the GPU, we could down-clock it, or for a pure headless operation turn it off for more power to the CPU.

Edit: Can you look into disabling the GPU entirely from boot.ini? If you can disable it see if your max clocks go up higher than what you are reporting now
Last edited by mlinuxguy on Tue Aug 30, 2016 1:34 am, edited 5 times in total.
mlinuxguy
 
Posts: 793
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, N1, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Tue Aug 30, 2016 1:17 am

crashoverride wrote:For TrustZone there is typically not an additional controller.

Perhaps I am misusing the term "TrustZone". I am referring to the embedded "System Control Processor (SCP) based on the Cortex-M3 processor"
mlinuxguy
 
Posts: 793
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, N1, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby joy » Tue Aug 30, 2016 1:37 pm

odroid wrote:Our test results so far.
1.65Ghz on 4-cores burning test has passed on Ubuntu and Android.
1.75Ghz on 3-cores burning test has passed too.

We are making more detail test reports with How-to guide for other people's testing the new boot blobs and kernel source patch for above two configurations.
It will be released around 20 hours later because we want to perform some benchmarking tests now.

Once we collect various feedbacks from some advanced users for a week, we will make a kernel/bootloader update package with new boot.ini file for easier & simpler clock/core options configuration.


Before this report, I should correct the current status for 1.75GHz with 3-cores burning.
For 1.75GHz, we're still checking if it works fully and the test result is for another higher frequency set than 1.5 GHz.
Really sorry to make this confusion.
Please refer to the following report.

And I attached the patches of u-boot and kernel we've adjusted so far.
We will really appreciate you to check them and give us any feedback or mentions.
It will be very helpful for our further investigation.
And also we would like to apologize again for our careless checking and mistakes.


Here is our test result up to now.

1. Current Status and Investigation Points

With the new firmware that has been adjusted CPU frequency sets, we're putting some kind of tests and also keeping investigations for unavailable frequency sets.
Please find two new cpu frequency sets, 1.680GHz and 1.656GHz in below.

And followings are our investigations we're checking so far.
(1) Performance confirmation of the new cpu frequency sets that is higher than 1.536GHz, '1.656GHz', '1.680GHz'
(2) Check the stability for '1.656GHz' and '1.680GHz' and put aging test
: Check the possibility of '1.656GHz' with 4 full cpu-cores
: Check the possibility of '1.680GHz' with 2 or 3 cpu-cores working
(3) Under a diversity of investigations for 1.752GHz and 2.016GHz including the stability check for how many cpus are available for the cases

2. Check List
(1) throughput Check
- sysbench on Ubuntu / Antutu or Geekbench on Android
(2) Stability Check
- Aging Test
- Adjust vcck/vddee voltage
- Adjust thermal-related improvement and throttling management
- Investigation about handling the number of cpu-cores
- Reduce heating
- Check HW-related intension points

3. Benchmark result so far

Code: Select all
# cd /sys/devices/system/cpu/cpu0/cpufreq
# echo <freq> > scaling_max_freq
# cat cpuinfo_cur_freq
# echo performance > scaling_governor
# sysbench --test=cpu --cpu-max-prime=20000 --num-threads=4 run


Image

* Note
As we tested so far, in case of 1680MHz, there is a limitation with full 4 cpu-cores working,
so currently we've just confirmed the frequency set operating with 3 cpu-cores through cpuburn aging test,
and we're examining the case in many ways like reducing heating or controlling voltage and so on.

4. Test for 1.656GHz with 4 full cpu-cores

We put stability and aging test for 1.656 GHz cpu frequency and
based on the result, we adjusted the dvfs configuration and thermal throttling table to support 1.6xx frequency sets.
and still putting more additional investigations.

(1) Performance Test
Please refer to the aforementioned benchmark result table.

(2) Stability and Aging Test
We put aging tests using C2 boards of variant HW revisions and confirmed it's working normally.
- Test Utility
: 'cpuburn-a53' on Ubuntu
: 'stress' utility on Android
- HW
: 9 ea (REV0.1/REV0.2)
- Test Period
: 3 days (about 70 hours)

(3) Updated Thremal Throttling Table
We adjusted thermal throttling trip-points and cooling-maps as following and you can check the update in attached kernel patch.

Image

(4) PATCHES
Please find the attachedment, 'c2_1.6MHz_freq_patch.zip'.
a. uboot patch
- 1656_dvfs_uboot.diff
- uboot_fip_gxb_bl30.bin (replace bl30.bin in uboot/fip/gxb/)
b. kernel patch
- cpu_freq_patch_kernel.diff

5. Test for 1.680GHz and Check how many cpus are available to run the system stably

(1) Performance Test
You can refer to the aforementioned benchmark result table for 4 full cpu-cores.
And we're still check and compare the actual performance for 3 cpu-cores
and I will share the result here asap.

(2) Stability and Aging Test
We put aging test using 2ea C2 boards and confirmed it's working normally so far.
- utility : cpuburn-a53 on Ubuntu
- HW : 2ea
- Test Period : 1 day

(3) PATCHES
Please find the attachedment, 'c2_1.6MHz_freq_patch.zip'.
a. uboot patch
- 1680_dvfs_uboot.diff
- uboot_fip_gxb_bl30.bin (same with 1656 case)
b. kernel patch
- cpu_freq_patch_kernel.diff (same with 1656 case)

6. Investigations for 1.752GHz and 2.016GHz
We think these frequency sets must be beneficial for some services even if it would work just with only 1 or 2 cpu-cores.
So, as I mentioned before, we're checking the possibility for 1.752 and 2.016GHz.
We're investigating and checking now and I will share it here as we have any update.


---------------------------------
Attachments
c2_1.6MHz_freq_patch.zip
(22.5 KiB) Downloaded 132 times
Last edited by joy on Tue Aug 30, 2016 2:15 pm, edited 1 time in total.
joy
 
Posts: 571
Joined: Fri Oct 02, 2015 1:44 pm
languages_spoken: english
ODROIDs: ODROID-C1+, XU4, X

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Tue Aug 30, 2016 1:51 pm

I have this version of u-boot downloaded from github u-boot-odroidc2-v2015.01.zip (latest one I could find)
Can I apply the patch against this one?
mlinuxguy
 
Posts: 793
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, N1, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby joy » Tue Aug 30, 2016 1:56 pm

mlinuxguy wrote:I have this version of u-boot downloaded from github u-boot-odroidc2-v2015.01.zip (latest one I could find)
Can I apply the patch against this one?


If the last commit of u-boot version you have is as below, you can use the uboot with the attached patches.
There is no information of u-boot-odroidc2-v2015.01.zip you mentioned.

Code: Select all
ODROID-C2: Add bl301-related source files
1. trees
- arch/arm/cpu/armv8/gxb/firmware/scp_task
- arch/arm/cpu/armv8/gxb/firmware/acs
- board/hardkernel/odroidc2/firmware

2. version
- the latest one (20160401)
: adjust DVFS table
: set some gpio controls after power off (UPS lines, LED)

Change-Id: I5ec7237cac629cfbc1cb5c0cf27b2dea88bb033a


Can you check it with our github?
https://github.com/hardkernel/u-boot
joy
 
Posts: 571
Joined: Fri Oct 02, 2015 1:44 pm
languages_spoken: english
ODROIDs: ODROID-C1+, XU4, X

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Tue Aug 30, 2016 2:02 pm

That's the version I have, I had to pull it down via a zip file so that is how it got named
mlinuxguy
 
Posts: 793
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, N1, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Tue Aug 30, 2016 2:06 pm

The 2 patches in your zipfile are the same
# diff 1656MHz_dvfs_uboot.diff 1680MHz_dvfs_uboot.diff
#
There is no 1680Mhz patch as that patch file is the same as 1656*
mlinuxguy
 
Posts: 793
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, N1, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby joy » Tue Aug 30, 2016 2:21 pm

mlinuxguy wrote:The 2 patches in your zipfile are the same
# diff 1656MHz_dvfs_uboot.diff 1680MHz_dvfs_uboot.diff
#
There is no 1680Mhz patch as that patch file is the same as 1656*


Could you download the zip file again?
There was a mistake.

There are 3 patches and 1 binary file in c2_1.6MHz_freq_patch.zip

1. update dvfs table on uboot, bl301
Currently, there is no method to fix max cpu frequency among 1656MHz or 1680MHz,
so you can choose one of them for your test.
(1) 1656MHz_dvfs_uboot.diff
(2) 1680MHz_dvfs_uboot.diff

2. update bl30.bin binary
(3) uboot_fip_gxb_bl30.bin

3. cpu_freq_patch_kernel.diff for kernel patch
(4) cpu_freq_patch_kernel.diff
It's just update thermal throttling table
joy
 
Posts: 571
Joined: Fri Oct 02, 2015 1:44 pm
languages_spoken: english
ODROIDs: ODROID-C1+, XU4, X

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby odroid » Tue Aug 30, 2016 2:26 pm

Side notes:
<1>
The old engineering sample PCB Rev 0.1 20150930 seems to be more unstable.
Please use PCB Rev 0.2 20151218 or higher for your test.

<2>
1.68Ghz generates heat a lot. We might need a bigger heatsink or active cooler to avoid the thermal throttling if the ambient temperature is higher than 15C probably.
User avatar
odroid
Site Admin
 
Posts: 28859
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby Snk » Tue Aug 30, 2016 8:17 pm

odroid wrote:Side notes:
<1>
The old engineering sample PCB Rev 0.1 20150930 seems to be more unstable.
Please use PCB Rev 0.2 20151218 or higher for your test.

<2>
1.68Ghz generates heat a lot. We might need a bigger heatsink or active cooler to avoid the thermal throttling if the ambient temperature is higher than 15C probably.


Hello Friend
I may be talking a great bullshit, but since HK think to give some compensation to consumers by this error of Amlogic would be interesting to create an active cooler for cooling the C2, as done in XU4 if get a higher frequency the chip generates more heat.
HK could send its C2 consumers, a unit of this asset sink.
User avatar
Snk
 
Posts: 275
Joined: Sun Jul 31, 2016 6:43 am
languages_spoken: Portuguese
ODROIDs: XU4 + eMMC 32GB + UART

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Wed Aug 31, 2016 3:25 am

Testing 1.656ghz NOTE: all 4-cores are active in my setting however I am using this heatsink not the stock one: viewtopic.php?f=139&t=18739#p130549

Ambient 77F ( 25C )
# ./dumpfreq.sh
scaling_cur_freq: 1656000
Scaling_available_frequencies: 100000 250000 500000 1000000 1296000 1536000 1656000
Scaling_max_freq: 1656000
Scaling_governor: performance
setting performance governor

sysbench --test=cpu --cpu-max-prime=20000 --num-threads=4 run
Freq, total time, total events, time event exec, exec time
1656000, 5.5836, 10000, 22.3096, 5.5774
Slightly slower than your results at 1.656ghz

=== ./cpuburn-a53
Code: Select all
root@odroid64-1:~# ./watchtemp.sh
time,temp,freq
0,36,1656000
5,36,1656000
10,45,1656000
15,51,1656000
.... skip .....
1975,67,1656000
1980,68,1656000
1985,67,1656000
1990,68,1656000

test.png
gnuplot of temperature at 1.65ghz
(27.15 KiB) Downloaded 1724 times

Next up will try higher frequencies. Its not clear from your post if you have any tables in the binary blob for any higher except 1.68ghz
Last edited by mlinuxguy on Wed Aug 31, 2016 12:36 pm, edited 1 time in total.
mlinuxguy
 
Posts: 793
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, N1, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby rooted » Wed Aug 31, 2016 5:15 am

I want to test but only have the beta board, I shall wait patiently for some results.
User avatar
rooted
 
Posts: 5680
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1, C1+, C2
XU3 Lite, XU4
N1
VU7+
HiFi Shield 2
Smart Power (original)

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Wed Aug 31, 2016 8:30 am

There is definitely a power problem at higher frequencies
I was driving the C2 with: 2A USB 5V power supply --> barrel connector --> C2
I got it to boot with 1.65ghz (and at 1.5ghz), however when I tried 1.68ghz I got continual resets as it loaded the kernel

I switched back to the HK power-supply and it booted fine at 1.68ghz
The USB charger OLED dongle claims 5.04v and pulling around 1.21a during boot (bounces around a lot on AMP reading)

Measuring the HK C2 power supply (no load) it shows 5.21v.
Makes me wonder what it gets pulled down to when loaded.
I need a bench power-supply where I can set it to something like 5.5v and check if it boots at the higher frequencies
mlinuxguy
 
Posts: 793
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, N1, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby joy » Wed Aug 31, 2016 11:21 am

mlinuxguy wrote:There is definitely a power problem at higher frequencies
I was driving the C2 with: 2A USB 5V power supply --> barrel connector --> C2
I got it to boot with 1.65ghz (and at 1.5ghz), however when I tried 1.68ghz I got continual resets as it loaded the kernel

Do you mean when you set max 1.65GHz, C2 board repeats reset during booting without any heavy load test like cpuburn-a53?
Last edited by joy on Wed Aug 31, 2016 11:45 am, edited 1 time in total.
joy
 
Posts: 571
Joined: Fri Oct 02, 2015 1:44 pm
languages_spoken: english
ODROIDs: ODROID-C1+, XU4, X

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Wed Aug 31, 2016 11:42 am

Not exactly, I'm still puzzling over what I'm seeing
I can boot 1.5ghz and 1.65ghz off the USB 5v 2a power supply, however when I set it to 1.658ghz it gets to the point midway thru the kernel boot where it does some eMMC frequency settings and then resets
Code: Select all
[    2.630764] aml_sd_emmc_probe: line 2954
[    2.641701] [aml_is_card_insert] card OUT
[    2.650684] [aml_is_card_insert] card OUT
[    2.650684] [    2.672118] aml_sd_emmc_reg_init 677                 <---- sometimes last log is here before reset
[    2.672228] pdata->caps 80000d47
[    2.673256] pdata->caps2 80a1
[    2.676183] get property:                ocr_avail, value:0x00200080
[    2.682487] get property:                    f_min, value:0x00061a80
[    2.688774] get property:                    f_max, value:0x05f5e100
[    2.695078] get property:             max_req_size, value:0x00020000
[    2.700639] usb 1-1: new high-speed USB device number 2 using dwc_otg
[    2.700909] dwc_otg: Indeed it is in host mode hprt0 = 00001101
[    2.713625] get property:                  pinname, str:emmc
[    2.719221] get property:                card_type, value:0x00000001
[    2.725530] get property:                gpio_dat3, str:
[    2.730786] get property:                 hw_reset, str:
[    2.790625] aml_emmc_hw_reset 1356                                   
[    2.790627] [aml_sd_emmc_probe] aml_sd_emmc_probe() success!
[    2.790861] ion_dev has 1 memory regions
[    2.812720] fb: osd_init_module
[    2.812832] fb: viu vsync irq: 35
[    2.813522] meson-fb has 1 memory regions
[    2.818877] meson-fb meson-fb: assigned reserved memory node linux,meson-fb ok
[    2.824662] tv_vout: tv_set_current_vmode[672]fps_target_mode=10
[    2.830596] tv_vout: mode is 10,sync_duration_den=1,sync_duration_num=60
[    2.834390] emmc: BKOPS_EN bit is not set
[    2.841220] [aml_sd_voltage_switch] switch to 1.8V for a non-uhs device.  <--- or resets here per debug UART log
[    2.841220]
[    2.841220] switch_vpu_mem_pd: vencp ON

I have nothing plugged into USB, just 1gb nic. It could be I just have marginal power to boot it, but I'm puzzled over why it was when I set it to 1.68ghz that I
started getting the resets. Once I switched to HK power it booted at 1.68ghz

The board is: Rev 0.2 12151218
It doesn't make sense to be resetting due to the 1.68ghz setting since it shouldn't be at that frequency at that stage of boot.
mlinuxguy
 
Posts: 793
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, N1, USB-IO

PreviousNext

Return to Issues

Who is online

Users browsing this forum: No registered users and 3 guests