| 08:41 | jnettlet[m]> | Ke: fyi my 16GB ECC unbuffered memory showed up today and it works without issues. |
| 08:41 | jnettlet[m]> | I am running at 1600Mhz 2133DDR speeds |
| 08:43 | Ke> | jnettlet[m]: any possibility of getting model number |
| 08:43 | Ke> | you could even start selling some in the accessories |
| 09:10 | jnettlet[m]> | Ke: just some budget reseller from the UK I found on ebay. hold on. |
| 09:12 | jnettlet[m]> | http://www.ebay.co.uk/itm/192205315848 |
| 09:13 | Ke> | picture reads hynix micron samsung o_O |
| 09:13 | Ke> | must be legit |
| 09:18 | jnettlet[m]> | I can assume they are buying whatever DDR4 modules are cheapest |
| 09:19 | Ke> | yup |
| 09:20 | Ke> | apparently you could do 2400 MTps on mbin |
| 09:20 | jnettlet[m]> | no 2400Mtps is not supported, and possibly won't on this revision. Marvell is having some issues |
| 09:21 | Ke> | can you do 2GHz cpu 2100MTps memory?= |
| 09:21 | jnettlet[m]> | yes, with active cooling |
| 09:21 | Ke> | could you possibly note this on the wiki? |
| 09:22 | Ke> | http://wiki.macchiatobin.net/tiki-index.php?page=MACCHIATObin+Interface+list eg. |
| 09:22 | jnettlet[m]> | Marvell is supposed to update it. We have made the request |
| 09:22 | Ke> | right |
| 09:22 | Ke> | https://www.solid-run.com/product/armada-8040-networking-community-board/ also Septmber here annoys me |
| 09:22 | jnettlet[m]> | Google also got upset about this. Marvell moves very slowly |
| 09:23 | Ke> | but I ordered already a unit, so I don't keep refreshing that page anymore |
| 09:23 | jnettlet[m]> | again the holdup is getting the chips from Marvell. |
| 09:24 | Ke> | I actually ordered late on purpose in order to possibly get an updated revision, if applicable, I meant the typo |
| 09:24 | jnettlet[m]> | you most likely will. |
| 09:24 | Ke> | Septmber -> September |
| 09:25 | Ke> | you should just let me have write access =o) |
| 09:25 | jnettlet[m]> | I will note. We have people going through our wiki fixing these currently. |
| 09:37 | jnettlet[m]> | okay, 100 runs of mtest in uboot have succeeded. Hopefully my new heatsinks will arrive today so I can push the board even more |
| 10:16 | topi`_> | more macchiatobin strangeness: why is it that the LED on the 10gbps copper iface #1 has started blinking? |
| 10:16 | topi`_> | it blinks even though there is nothing connected |
| 10:16 | topi`_> | sounds like some kernel bug but there is nothing on dmesg |
| 10:17 | Ke> | NSA spying firmware was accidentally started |
| 10:17 | topi`_> | rabeeh: can you talk about the future products? It seems that the macchiatobin is a solid product (despite of some problems), I wonder if it would be feasible to create lesser variants with the Armada 7040 chip? Is it pin compatible? |
| 10:18 | topi`_> | since the 7040 has only half of the SERDES lanes, I bet it will sell cheaper than the 8040 |
| 10:19 | topi`_> | the ClearFog is a good product, but I'd like to have more/faster cores in my router. Who knows what kind of workloads I want to push there ;) |
| 10:21 | topi`_> | in some cases, the lack of CPU power is very annoying. I have a Synology NAS box with a low-end Marvell Armada (dual-core A9) and the Photo app is sometimes just too slow... |
| 10:21 | topi`_> | especially importing new stuff |
| 10:27 | jnettlet[m]> | topi`_: you missed the PR. We are releasing the 8040 as a COM, and will be releasing a high powered Clearfog that uses it. |
| 10:27 | jnettlet[m]> | As Marvell grows the 8xxx line it will allow us to increase the cores of the COM up to 32 |
| 10:28 | jnettlet[m]> | https://www.solid-run.com/2017/06/28/solidrun-unveils-first-arm-based-com-express-type-7-module/ |
| 10:32 | jnettlet[m]> | one the technologies that is just sorting itself out is NBase-T. This is obviously something that would fit in with the new ClearFog |
| 10:32 | Ke> | what will be the difference between ClearFog and mbin? |
| 10:33 | Ke> | ram soldered to board, no ECC? |
| 10:33 | jnettlet[m]> | size, and the ClearFog will use a COM 7 express, so it can be upgraded. |
| 10:33 | jnettlet[m]> | we will be offering modules with soldered memory, as well as modules that support carrier based memory. |
| 10:34 | jnettlet[m]> | so yes ECC will be supported |
| 10:34 | Ke> | size means clearfog will be smaller? |
| 10:35 | jnettlet[m]> | yes smaller, and different ports. |
| 10:35 | Ke> | and 32 means 32 a72 cores? |
| 10:36 | jnettlet[m]> | note that we will support dual m.2 connectors, so you can in theory have mirrored pcie nvme storage. It is pcie 3.0 x1, so you can have storage that supports 1GB/s throughput...basically full bandwidth of a 10Gbps network link. |
| 10:36 | jnettlet[m]> | yes 32 a72s...but that is far along the roadmap. I would expect 8 cores next. |
| 10:37 | jnettlet[m]> | and 16 end of next year or beyond. |
| 10:37 | Ke> | I guess there will also be a die shrink? |
| 10:37 | jnettlet[m]> | no idea |
| 10:37 | jnettlet[m]> | marvell has put out a PR |
| 10:42 | jnettlet[m]> | basically for us the focus is on the 8040 based COM module and carrier board. |
| 10:44 | Ke> | will you also update mbin for either the module socket or selection of other socs? |
| 10:44 | Ke> | or is it up to Marvell |
| 10:44 | jnettlet[m]> | that is up to Marvell. |
| 10:45 | Ke> | thanks |
| 11:30 | topi`_> | I'm back with Vivante trouble... the client informed that they did nt get the screen on every boot, sigh |
| 11:30 | topi`_> | systemd-udevd[485]: Failed to apply ACL on /dev/dri/card1: Operation not supported |
| 11:30 | topi`_> | this is something suspicious that shows in systemd journal |
| 11:32 | Ke> | CONFIG_TMPFS_POSIX_ACL |
| 11:33 | Ke> | though that should fail deterministicly, not randomly |
| 11:34 | topi`_> | hmm, maybe that comes at every boot. Unfortunately journalctl -b -1 says no persisent journal... :( |
| 11:34 | Ke> | mkdir /var/log/journal |
| 11:34 | topi`_> | I need to learn how to configure persistent journal |
| 11:35 | Ke> | I think you just did |
| 11:35 | topi`_> | ok let me see if that works |
| 12:20 | jnettlet[m]> | topi`_: which distro? basically systemd/journald checks to see if /var/log/journal is writable. If it isn't then it keeps everything in memory cache. |
| 12:21 | Ke> | and whatever happens, do not look at the code that implements this behaviour |
| 12:21 | jnettlet[m]> | there are some config file options that can be tweaked to only use a memory buffer for logging, so if /var/log/journal exists and is writable then check the config files. |
| 12:21 | Ke> | it's bad |
| 12:22 | Ke> | journald space waste has been so awful for me that I have been considering doing a rewrite |
| 12:23 | jnettlet[m]> | but you can limit the size of the journals it keeps |
| 12:23 | jnettlet[m]> | I think the out of the box experience is a bit ridiculous though |
| 12:23 | Ke> | or at least converter that dumps the log and converts it to better format |
| 12:24 | Ke> | even just dumping it in text with all the metadata and compressing will improve the space usage tremendously |
| 12:25 | Ke> | I noticed half of hte log files I am getting are actually zeroed out |
| 12:25 | Ke> | the |
| 12:32 | jnettlet[m]> | I will be able to provide good pcie nvme numbers for the mcbin soon also. |
| 13:17 | topi`_> | jnettlet[m]: we shipped bleeding edge to the customer. That is, debian/stretch |
| 13:18 | topi`_> | I think the message about ACL was not the reason why the Qt-based display service failed. It said that the HDMI1 output was not available and that led to the failure |
| 13:18 | topi`_> | but after every second reboot the HDMI1 is evidently available because it goes on happily... |
| 13:20 | topi`_> | jnettlet[m]: after we go NVMe, is there any reason to buy these M.2 sata cards any longer? :) |
| 13:20 | topi`_> | I guess NVMe is basically just PCIe |
| 13:22 | jnettlet[m]> | topi`_: I don't believe u-boot supports NVMe yet. |
| 13:22 | jnettlet[m]> | I am sure the SOC boot-strap code does not. |
| 13:23 | jnettlet[m]> | and the M.2 sata cards are much cheaper currently. |
| 13:25 | topi`_> | the SOC can boot from eMMC |
| 13:26 | topi`_> | and on the mbin, from SPI NOR |
| 13:26 | topi`_> | maybe the NVMe is too complex to be able to fit a suitable driver on the NOR |
| 13:28 | jnettlet[m]> | I will let you know. That is what this test is about. |
| 13:29 | jnettlet[m]> | but I figured just supporting SATA-3 which maxes out at 500MB/s on a device that can do 10Gbps networking doesn't make much sense. Especially since we can configure the serdes lanes dynamically. |
| 13:30 | jnettlet[m]> | my XPS 13 can take either SATA or NVMe drives which gave me the idea. |
| 13:30 | topi`_> | so you can divert more SERDES lanes to PCIe if you take them away from SATA? |
| 13:30 | jnettlet[m]> | the serdes lanes are physical, and do have some limitations as to the combinations |
| 13:31 | jnettlet[m]> | so basic SERDES can be USB-3, 1x pcie, or SATA 3 |
| 13:32 | jnettlet[m]> | so dual M.2 connectors can be either 2 pcie interfaces, 2 sata, or one of each. |
| 13:32 | topi`_> | qls |
| 13:32 | topi`_> | *sigh* i still need to learn to navigate in a dual-head environment |
| 13:33 | topi`_> | I bought a usb-c dock for my 2016 macbookpro :) |
| 13:33 | jnettlet[m]> | however we are also putting a pcie x4 connector like the mcbin has on the board. which may allow us to use a riser to allow 2 additional sata ports on the Clearfog. |
| 13:33 | jnettlet[m]> | nice which one? |
| 13:33 | topi`_> | the base model. But this also comes with NVMe :) |
| 13:34 | topi`_> | 10165886976 bytes transferred in 7.005993 secs (1451027304 bytes/sec) |
| 13:34 | topi`_> | this is 1.4GB/s, measured with dd if=bigfile ;) |
| 13:34 | topi`_> | oh and my $HOME is also encrypted |
| 13:36 | topi`_> | I think also the 8040 has enough bandwidth to actually realize these kinds of numbers |
| 13:36 | topi`_> | plus the crypto engine is just plain fast |
| 13:36 | jnettlet[m]> | topi`_: what is your mcbin idle temperature at this point? |
| 13:37 | topi`_> | it has a small fan now, so it goes down to 40C |
| 13:37 | topi`_> | now the cores are mostly loaded, and temps at 52C |
| 13:38 | jnettlet[m]> | at 2Ghz? |
| 13:39 | topi`_> | let me check :) |
| 13:39 | topi`_> | I fiddled with those because I had problems with the mv_ddr |
| 13:40 | topi`_> | no cpufreq :( |
| 13:40 | topi`_> | [ 2.795305] cpu cpu0: Failed to get clock rate for CPU 0 |
| 13:40 | topi`_> | by looking at the jumpers, it seems I am currently at 2.0ghz core |
| 13:41 | topi`_> | there is cpuidle with two governors, ladder and menu |
| 13:42 | topi`_> | but I guess that doesn't have any capabilities to actually bring down cores to save power |
| 13:46 | jnettlet[m]> | cpuidle and putting a core offline in the kernel should both put the core in WFI, so save the same amount of power |
| 14:03 | topi`_> | does the WFI drop the voltage from the core? |
| 14:03 | topi`_> | I guess the only way to gain maximum powersavings is to drop the voltage |
| 14:05 | jnettlet[m]> | it depends on how the chip is architected. It is standby so it will never drop the power to zero, but it puts the chip in standby and gates all the clocks. |
| 14:21 | topi`_> | now back to the display problems on the HB... Is there any way to find out what there is on the opposite side of the HDMI cable? |
| 14:21 | topi`_> | to find out whether or not the cable is properly connected and secured. |
| 14:22 | topi`_> | we ship with a HDMI <-> DVI-D cable (the opposite end has a dvi-d connector) |
| 14:24 | jnettlet[m]> | you basically want to check the logs to see if the edid is detected |
| 15:54 | vpeter> | jnettlet[m]: I just noticed that I have only 10MB/s read speed on solidpc with booting from sd card Samsung EVO+. Do i need something special enabled in kernel (4.11)? |
| 16:12 | jnettlet[m]> | vpeter: what does /sys/kernel/debug/mmc0/ios report? |
| 16:12 | jnettlet[m]> | even then 10MB/s is very slow. |
| 16:13 | vpeter> | LibreELEC-SolidPC:~ # cat /sys/kernel/debug/mmc0/ios |
| 16:13 | vpeter> | clock: 50000000 Hz |
| 16:13 | vpeter> | actual clock: 50000000 Hz |
| 16:13 | vpeter> | vdd: 21 (3.3 ~ 3.4 V) |
| 16:13 | vpeter> | bus mode: 2 (push-pull) |
| 16:13 | vpeter> | chip select: 0 (don't care) |
| 16:13 | vpeter> | power mode: 2 (on) |
| 16:13 | vpeter> | bus width: 2 (4 bits) |
| 16:13 | vpeter> | timing spec: 2 (sd high-speed) |
| 16:13 | vpeter> | signal voltage: 0 (3.30 V) |
| 16:13 | vpeter> | driver type: 0 (driver type B) |
| 16:26 | jnettlet[m]> | yeah so just seeing it as a normal everyday sdhc card. |
| 16:26 | jnettlet[m]> | event then it should still do 20MB/s |
| 16:37 | vpeter> | stupid electronic devices |
| 16:38 | Ke> | I have also had problems with absurdly slow bus speeds on my rk3288/C201 |
| 16:38 | Ke> | firefly-rk3399 did seem no better than that |
| 16:38 | Ke> | which was extremely sad |
| 16:42 | vpeter> | usb card reader works fine. |
| 16:43 | vpeter> | To bad emmc is broken too :( |
| 16:46 | jnettlet[m]> | vpeter: do you have the latest BIOS? |
| 16:46 | vpeter> | Yes. |
| 16:47 | jnettlet[m]> | I will test in a bit |
| 16:55 | vpeter> | Seems I have old image running when I was testing emmc. Now I have 20MB/s. But still not enough. |
| 20:41 | jnettlet[m]> | vpeter I am getting 20MB/s |
| 20:42 | jnettlet[m]> | however my card has switched over to 1.8V signalling to support UHS. Unfortunately this is just a crappy card. |
| 20:42 | jnettlet[m]> | the timing spec I am getting is sd uhs DDR40 |
| 20:43 | jnettlet[m]> | the eMMC is HS200 so much faster |
| 20:43 | jnettlet[m]> | blah.. DDR50 for the sdhc |
| 20:44 | jnettlet[m]> | yeah. eMMC is doing reads at ~100MB/s |
| 20:44 | jnettlet[m]> | I will dig up a better card in a bit. |