Installing Lubuntu 18.10

[davidbam]

With reference to https://tsbt.co.uk/forum/viewtopic.php?f=139&t=9319 I thought a new thread was appropriate.

Installation is very easy. Now orienting myself to the different menus.

First problem is that my Preferences and System Tools menus do not have options for Software Updater or Additional Drivers (to replace nouveau ones). I have attached photos of the menus I get

[Alez]

Lubuntu 18.04 comes with Synaptic and Software Updater, both under ‘System Tools’ in the menu or it should. Additional Drivers is a tab on software updater.
Use search for Software Updater. I'll wonder out and check one of my systems to see what I have got.

From screenshot you have installed 18.10 then Software updater is now called Discover Software Center and Synaptic is now muon.

How to Launch

> System Tools ->Muon Package Manager -> Settings-> Configure Software Sources

> System Tools ->Discover Software Center -> Settings -> Ubuntu (Default) -> Settings (Gear icon) -> Software Sources

Lubuntu 18.10 short manual here

[davidbam]

Well it is now working

Crunching a single Collatz sieve on nVidia (which was the main aim), with lots of WU for Primegrid on both CPU and GPU.

I never did get an Additional Drivers tab anywhere but instead searched for nVidia in the Muon package manager & rather took a stab in the dark I'm afraid. Not sure I could repeat the process precisely TBH

Primegrid now reports me as having "NVIDIA GeForce GTX 960 (4095MB) driver: 39087" so I've got a slightly newer driver

lspci reports 02:00.0 VGA compatible controller: NVIDIA Corporation GM206 [GeForce GTX 960] (rev a1)

lsmod reports

Code: Select all

Module                  Size  Used by
bnep                   20480  2
ccm                    20480  6
snd_hda_codec_hdmi     49152  4
bluetooth             548864  9 bnep
ecdh_generic           24576  1 bluetooth
intel_rapl             20480  0
x86_pkg_temp_thermal    16384  0
intel_powerclamp       16384  0
rtl8xxxu              122880  0
arc4                   16384  2
snd_hda_codec_realtek   106496  1
snd_hda_codec_generic    73728  1 snd_hda_codec_realtek
snd_hda_intel          40960  5
coretemp               16384  0
snd_hda_codec         126976  4 snd_hda_codec_generic,snd_hda_codec_hdmi,snd_hda_intel,snd_hda_codec_realtek
mxm_wmi                16384  0
snd_hda_core           81920  5 snd_hda_codec_generic,snd_hda_codec_hdmi,snd_hda_intel,snd_hda_codec,snd_hda_codec_realtek
kvm_intel             208896  0
snd_hwdep              20480  1 snd_hda_codec
kvm                   622592  1 kvm_intel
snd_pcm                98304  4 snd_hda_codec_hdmi,snd_hda_intel,snd_hda_codec,snd_hda_core
rtl8192cu              73728  0
irqbypass              16384  1 kvm
rtl_usb                20480  1 rtl8192cu
rtl8192c_common        57344  1 rtl8192cu
crct10dif_pclmul       16384  0
rtlwifi                86016  3 rtl8192c_common,rtl_usb,rtl8192cu
crc32_pclmul           16384  0
ghash_clmulni_intel    16384  0
pcbc                   16384  0
mac80211              794624  4 rtl_usb,rtl8192cu,rtlwifi,rtl8xxxu
aesni_intel           200704  4
aes_x86_64             20480  1 aesni_intel
crypto_simd            16384  1 aesni_intel
cryptd                 24576  3 crypto_simd,ghash_clmulni_intel,aesni_intel
snd_seq_midi           16384  0
glue_helper            16384  1 aesni_intel
snd_seq_midi_event     16384  1 snd_seq_midi
cfg80211              663552  2 rtlwifi,mac80211
intel_cstate           20480  0
joydev                 20480  0
snd_rawmidi            32768  1 snd_seq_midi
input_leds             16384  0
intel_rapl_perf        16384  0
snd_seq                65536  2 snd_seq_midi,snd_seq_midi_event
snd_seq_device         16384  3 snd_seq,snd_seq_midi,snd_rawmidi
snd_timer              32768  2 snd_seq,snd_pcm
snd                    81920  21 snd_hda_codec_generic,snd_seq,snd_seq_device,snd_hda_codec_hdmi,snd_hwdep,snd_hda_intel,snd_hda_codec,snd_hda_codec_realtek,snd_timer,snd_pcm,snd_rawmidi
mei_me                 40960  0
mei                    98304  1 mei_me
soundcore              16384  1 snd
wmi                    24576  1 mxm_wmi
mac_hid                16384  0
nvidia_uvm            786432  2
binfmt_misc            20480  1
sch_fq_codel           20480  6
parport_pc             36864  0
ppdev                  20480  0
lp                     20480  0
parport                49152  3 parport_pc,lp,ppdev
ip_tables              24576  0
x_tables               40960  1 ip_tables
autofs4                40960  2
btrfs                1163264  0
xor                    24576  1 btrfs
zstd_compress         159744  1 btrfs
raid6_pq              114688  1 btrfs
libcrc32c              16384  1 btrfs
dm_mirror              24576  0
dm_region_hash         20480  1 dm_mirror
dm_log                 20480  2 dm_region_hash,dm_mirror
hid_logitech_hidpp     36864  0
hid_logitech_dj        20480  0
hid_generic            16384  0
usbhid                 49152  0
hid                   126976  4 usbhid,hid_generic,hid_logitech_dj,hid_logitech_hidpp
nvidia_drm             40960  1
nvidia_modeset       1110016  3 nvidia_drm
nvidia              14368768  141 nvidia_uvm,nvidia_modeset
drm_kms_helper        172032  1 nvidia_drm
syscopyarea            16384  1 drm_kms_helper
sysfillrect            16384  1 drm_kms_helper
sysimgblt              16384  1 drm_kms_helper
fb_sys_fops            16384  1 drm_kms_helper
firewire_ohci          40960  0
drm                   458752  4 drm_kms_helper,nvidia_drm
ahci                   40960  1
firewire_core          65536  1 firewire_ohci
tg3                   163840  0
i2c_i801               28672  0
lpc_ich                24576  0
libahci                32768  1 ahci
crc_itu_t              16384  1 firewire_core
ipmi_devintf           20480  0
ipmi_msghandler       102400  2 ipmi_devintf,nvidia

[davidbam]

After the sprint has finished, I'll be trying it out with an AMD GPU too. If it all works interchangeably, then I'll be a very happy bunny

[davidbam]

... presently running a Primegrid PPS Sieve which was the other reason for trying it. All looking good so far

ETA : correction, nVidia always did run PPS Sieve. It was AMD which failed due to missing library on Ubuntu

[Bryan]

When you start running Collatz, don't forget to "optimize" for your card

[davidbam]

Yes thanks, is that the only project where GPU optimisation is possible? I'll be changing out the CPU as well so will go through the whole benchmarking thing again.

I am a bit surprised to see that the GTX980 is doing slightly less well (on PPS Sieve) than it was in the crappy old HP motherboard / G3250 @ 3.20GHz / 4Gb Ram (yes 4) / Ubuntu / USB stick boot ... than it is now in a nice Asrock X79 mobo / i7-4960X CPU @ 3.60GHz / 24Gb RAM / Lubuntu / SSD boot

It definitely has a CPU thread all to itself. Different nVidia driver version I suppose. Will monitor as more results are validated

[Bryan]

That is the only project that comes to mind concerning optimization and I can't think of any that have "optimized" apps available. But I don't do much GPU crunching so that isn't necessarily the definite answer

One thing to watch is the loading on the GPU. On some projects you need to run more than 1 WU at a time. That is usually GPU dependent. Like on PG my 1080Ti runs 60% loading with 1 WU running and it goes to 100% running 2 at a time.

[davidbam]

I don't actually know how to run more than one WU on the GPU If I've been told before, I've forgotten, sorry.

[davidbam]

screendump from psensor - suggests 'graphics' utilisation is pretty high so I assume that is the GPU boincing away

[davidbam]

I have graphed 'graphics utilisation' (seems a bit recursive )

It certainly isn't jammed up there at 99-100% so perhaps 2 WU are what is needed

[Alez]

I don't actually know how to run more than one WU on the GPU If I've been told before, I've forgotten, sorry.

How you do it is project dependent via an app_config.xml the same as on collatz. The GTX 960 is probably better running single units.
To run two you need to add this to the primegrid project directory and then reread config files. Off the top of my head var/lib/boinc/projects/

I prefer to use Gedit

Code: Select all

sudo apt-get install gedit

to edit config files etc.
to run it from terminal

Code: Select all

gksudo gedit

the app_config

Code: Select all

<app_config>
<app>
<name>pps_sr2sieve</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

This will run two apps per card using 0.5 cpu core between them. Change cpu_usage to 1 to run 0.5 core per app. nVidia shouldn't need a core per app but AMD does.
Can you see this this forum ?

[davidbam]

Well I'll try it on the GTX960 for fun & check the effect on run times - presumably any one task will slow but overall credits will be higher. It sure has changed the shape of the GPU utilisation graph (black line)

[davidbam]

In for a penny ... ...
This graph is from Linux Mint with a GTX1080 graphing the GPU utilisation. The step change in the graph (the 1/3rd nearest the right) is using 4 WU on the GPU. The 2/3rds on the left is 3 WU simultaneously. Looks like it could handle 5

Am keeping an eye on temperatures and overall run times

[Alez]

Remember to up the amount of cpu. Whatever you set as cpu is what it uses between all apps on GPU. Go to low and it won't feed the GPU properly.
Also, don't get carried away with loading. The real trick is the times. If you can do one unit in 5 mins then you must be able to do 2 in 10 or less. For 3 , 15 mins or less etc or you are now being inefficient by loading the GPU too much. Also watch the temps on the card as obviously the more you push it the more the temp can rise. It's a balancing act.

[davidbam]

Thanks. Yes, that's what I meant by keeping an eye on temperatures and overall run times

[Bryan]

Your graph is very surprising. My 1080Ti runs about 60% load with 1 WU and 100% loading with 2. As Alez says, be sure you have enough CPU reserved to service the GPU. Try suspending the CPU WU and see what happens to your graph. If you don't have enough CPU reserved -or- you are trying to run too many WU it will cause you to lose credits.

[davidbam]

I am going to pull the card and try it in another machine. I noticed that even a single WU had been taking 3 times longer than it did originally ... Yes, even before I had changed anything.

[Alez]

I am going to pull the card and try it in another machine. I noticed that even a single WU had been taking 3 times longer than it did originally ... Yes, even before I had changed anything.

Check the temperature on the card. It might be down clocking to protect itself. Either that or it is not getting serviced fast enough through either CPU usage or bandwidth. PPS-Sieve tasks tend to be very similar, so each task should take the same time to run within a few seconds. There are no longer or shorter tasks like some apps.