11/10-16/10 The Moon*Shine Challenge-Einstein

[Alez]

sometimes the best option when it all goes Pete Tong

[robmacagain]

Ayeeeee ken ya big man .

[scole of TSBT]

Only other thing I could think of was maybe the current running WUs were set to use the full GPU and won't use the new settings until the current WUs finishes.

[robmacagain]

Nope they finished i deleted the app_config re did it again re read the config then re booted and it's still doing 1 per gpu core, so i have decided to just leave it to run 1 per core and that's that, at least its doing some better than none eh.

[scole of TSBT]

One thing I did determine today. Einstein AMD GPU apps need to run on in x8 or x16 speed PCIe slot. PCIe x4 slots are a huge bottleneck and will at least double the run time. At least that's the case for R9 270 and HD 7970 GPUs.

Also, are any of the Einstein apps double precision?

On another issue, Rob, what does your cc_config.xml look like?

[robmacagain]

This, it's huge >

<cc_config>
<log_flags>
<file_xfer>1</file_xfer>
<sched_ops>1</sched_ops>
<task>1</task>
<android_debug>0</android_debug>
<app_msg_receive>0</app_msg_receive>
<app_msg_send>0</app_msg_send>
<async_file_debug>0</async_file_debug>
<benchmark_debug>0</benchmark_debug>
<checkpoint_debug>0</checkpoint_debug>
<coproc_debug>0</coproc_debug>
<cpu_sched>0</cpu_sched>
<cpu_sched_debug>0</cpu_sched_debug>
<cpu_sched_status>0</cpu_sched_status>
<dcf_debug>0</dcf_debug>
<disk_usage_debug>0</disk_usage_debug>
<file_xfer_debug>0</file_xfer_debug>
<gui_rpc_debug>0</gui_rpc_debug>
<heartbeat_debug>0</heartbeat_debug>
<http_debug>0</http_debug>
<http_xfer_debug>0</http_xfer_debug>
<mem_usage_debug>0</mem_usage_debug>
<network_status_debug>0</network_status_debug>
<notice_debug>0</notice_debug>
<poll_debug>0</poll_debug>
<priority_debug>0</priority_debug>
<proxy_debug>0</proxy_debug>
<rr_simulation>0</rr_simulation>
<rrsim_detail>0</rrsim_detail>
<sched_op_debug>0</sched_op_debug>
<scrsave_debug>0</scrsave_debug>
<slot_debug>0</slot_debug>
<state_debug>0</state_debug>
<statefile_debug>0</statefile_debug>
<suspend_debug>0</suspend_debug>
<task_debug>0</task_debug>
<time_debug>0</time_debug>
<trickle_debug>0</trickle_debug>
<unparsed_xml>0</unparsed_xml>
<work_fetch_debug>0</work_fetch_debug>
</log_flags>
<options>
<allow_multiple_clients>0</allow_multiple_clients>
<allow_remote_gui_rpc>0</allow_remote_gui_rpc>
<client_version_check_url>http://boinc.berkeley.edu/download.php? ... _check_url>
<client_new_version_text></client_new_version_text>
<client_download_url>http://boinc.berkeley.edu/download.php< ... wnload_url>
<coproc>
<type>miner_asic</type>
<count>1</count>
<peak_flops>0.000000</peak_flops>
<device_nums>0</device_nums>
</coproc>
<disallow_attach>0</disallow_attach>
<dont_check_file_sizes>0</dont_check_file_sizes>
<dont_contact_ref_site>0</dont_contact_ref_site>
<exit_after_finish>0</exit_after_finish>
<exit_before_start>0</exit_before_start>
<exit_when_idle>0</exit_when_idle>
<fetch_minimal_work>0</fetch_minimal_work>
<fetch_on_update>0</fetch_on_update>
<force_auth>default</force_auth>
<http_1_0>0</http_1_0>
<http_transfer_timeout>300</http_transfer_timeout>
<http_transfer_timeout_bps>10</http_transfer_timeout_bps>
<max_event_log_lines>2000</max_event_log_lines>
<max_file_xfers>8</max_file_xfers>
<max_file_xfers_per_project>2</max_file_xfers_per_project>
<max_stderr_file_size>0</max_stderr_file_size>
<max_stdout_file_size>0</max_stdout_file_size>
<max_tasks_reported>0</max_tasks_reported>
<ncpus>-1</ncpus>
<network_test_url>http://www.google.com/</network_test_url>
<no_alt_platform>0</no_alt_platform>
<no_gpus>0</no_gpus>
<no_info_fetch>0</no_info_fetch>
<no_priority_change>0</no_priority_change>
<os_random_only>0</os_random_only>
<proxy_info>
<socks_server_name></socks_server_name>
<socks_server_port>80</socks_server_port>
<http_server_name></http_server_name>
<http_server_port>80</http_server_port>
<socks5_user_name></socks5_user_name>
<socks5_user_passwd></socks5_user_passwd>
<http_user_name></http_user_name>
<http_user_passwd></http_user_passwd>
<no_proxy></no_proxy>
</proxy_info>
<rec_half_life_days>10.000000</rec_half_life_days>
<report_results_immediately>0</report_results_immediately>
<run_apps_manually>0</run_apps_manually>
<save_stats_days>30</save_stats_days>
<skip_cpu_benchmarks>0</skip_cpu_benchmarks>
<simple_gui_only>0</simple_gui_only>
<start_delay>0.000000</start_delay>
<stderr_head>0</stderr_head>
<suppress_net_info>0</suppress_net_info>
<unsigned_apps_ok>0</unsigned_apps_ok>
<use_all_gpus>1</use_all_gpus>
<use_certs>0</use_certs>
<use_certs_only>0</use_certs_only>
<vbox_window>0</vbox_window>
</options>
</cc_config>

[robmacagain]

Afaic a pcie gen 4 slot should be enough on paper to handle a 7990, mine is in a x 4 slot so that should do, i do have a x 16 slot but the reservoir is in the way so it can't go in it, but as i said its in a x 4 slot so it should be fine.

The tech spec on the HD7990 says it's PCI Express 3.0.

[scole of TSBT]

Afaic a pcie gen 4 slot should be enough on paper to handle a 7990, mine is in a x 4 slot so that should do, i do have a x 16 slot but the reservoir is in the way so it can't go in it, but as i said its in a x 4 slot so it should be fine.

The tech spec on the HD7990 says it's PCI Express 3.0.

Each of your GPUs is taking ~5900 sec to finish a single BRP4G WU. My 7970s in x8 slots are each finishing 2 of those WUs in ~2200 secs. If your card was in the x16 slot, I bet each GPU would finish a WU in 1500 sec (running 1 WU per GPU). Mine did.

[robmacagain]

I have just tested it now and cpuz it says its running at x 16 so who knows whats up?

Going to try changing the driver, what one are you running with?

[scole of TSBT]

CCC 14.4

[robmacagain]

Ok thanks, have just wiped with AMDCLEANUP utility, just installed 14.4 and am re booting now.....

[robmacagain]

well as the Chinese say F*** me Sideways, GRRRRRRRR well it looks like ccc14.10 won't play, This is now 14.4 and L@@K >


:bounce: :mrgreen: :bounce: :mrgreen: :bounce: :wav:.

[robmacagain]

Thanks for the help btw, much appreciated :).

Anyone want a mug of Columbian ?

[scole of TSBT]

I have just tested it now and cpuz it says its running at x 16 so who knows whats up?

Going to try changing the driver, what one are you running with?

Where does cpuz show you the PCIe speed of a card? I only saw my board supported a max of x16. One slot is x16, but the other is only x4.

[robmacagain]

Ok, at the side of where it says bus interface there is a ? click it and it will verify your slot configuration, click start render test then on the bottom right hand side, sorry LEFT hand side it will tell you what the slot is running at.

[robmacagain]

It wont let me take a screen shot so i have taken a picture and uploaded to pb.


image ru

[Alez]

Well at least it's working now. The drivers seem to be really screwed up lately. Think it's all the power saving features for the newer cards. I'm using 14.4 as I had problems with my R270 playing well with the 7770 it shares a home with. Rolling my nVidias back to 337.88 made the 4 card setup work correctly for the first time in ages without crashing every few days and weird performance. ( thanks for the suggestion scole )
I think some boards are x16 and x8 but it doesn't matter which slot you use. The first slot used will be x16 as long as the others are not occupied.

TSBT is still 9th but with a far more respectable total. Scole sitting pretty in 15th and I have dragged my sorry butt up to 21st. On the whole I have Einstein running better but still not correct yet. Of course having 466 pending now doesn't help ...
I also incresed the page file amount in windows to help as I was getting a high page file usage. I also plugged in a few USB sticks and used them in readyboast mode as an attempt to help as well. Desperate times

As far as I am aware Einstein is single precision.

[scole of TSBT]

I'm disappointed with the GTX 780 performance. Maybe I have a bad configuration but it's taking 8000 sec to finish a BRP5 WU while running 2 at a time. I was expecting it to do better than 7970, not 33% slower.

[Alez]

I've stopped running them just now. My GTX 660ti's are taking 25000 secs to do 2. My 7970's are taking @16000 secs but that is running 4. Maybe they are far more memory intensive ? Anyhoo, stopped them just now and see what difference that makes to my RAC.

[robmacagain]

My Gpu is taking 10,400 for 2, these are ridiculously sloth like, i don't ken how , and most of them are just sitting in the validation queue ffs.

[scole of TSBT]

I'm telling ya. It's a PCIe bottleneck. I recently put a new mobo in a system with a R9 270 and 7970. It wouldn't boot with the 7970 in 1st slot, so the 270 went in the 1st one and 7970 in the 2nd one. I assumed with both slots used, they were both running at x8, but the 7970 was taking 16000 secs to finish 2 WUs. I checked the mobo spec and be darned if Gigbyte didn't make it so that the 1st slot was always x16 and the 2nd was always x4. I guess because of a PCIe version compatibility issue, the 7970 wouldn't work in the 1st slot. So I did a quick mobo switch with one the 7970 would work in the 1st x16 slot (but 270 still had to go in a x4). Now the 7970 is finishing 2 WU in the expected 6k-7k sec range, while the 270 is flailing in the x4 slot finishing 2 WU in 17000 sec.

With 4 GPUs, I'd bet the PCIe bus as a whole is over saturated.

I just saw Rob's times. I think gpuz is fibbing to him and it's really a x4 slot. What is mobo model #?

[robmacagain]

GigaByte GA-MA790X-UD3P

[scole of TSBT]

GigaByte GA-MA790X-UD3P

Spec says 2nd slot is x8, which should be fine for a single 7970, but perhaps since there's really 2 7970 GPUs on the card, each GPU is running like a 7970 in a x4 slot. It's running better than single 7970s in x4 slots, but not as well as it could.

[robmacagain]

May get a x16 riser and use that, but its going to be too late for this.

[Alez]

Big thread on Einstein about the BRP5 units. http://einstein.phys.uwm.edu/forum_thre ... 028#123640

trawling through it a few points came to light

"BRP5 workload compared to BRP4 tasks has increased by a factor of ~5x, but the processing times are ~10x and CPU loading is ~15x." I think the processing times are closer to 5 - 6 times but the credit is only 3 1/2.

Also BRP5 has a lot more data transfer between GPU and memory, hence which slot a card is in will have a big effect. Also the speed of your memory modules will have a big effect. I would also guess that the memory speed and more importantly the bus width will have a big effect. The 7970's have large memory bus probably better than 780.

Also explains why my 6990 is poor here as it is in an old server board with dd2 memory.

Finally for the nVidia's the apps are cuda 3.3 which was optimised for 5 series cards, for best effect 7 series cards need cuda 5.5 or 6

seems basically board/slot/memory/GPU combinations will have a big effect on performance for BRP5. Means we can't compare our performance amongst ourselves or even between our own systems unless they are identical.

[scole of TSBT]

Rats! BOINC@AUSTRALIA is slowly increasing their lead. Somebody go on a recon mission, find Megacruncher and git him home!

[scole of TSBT]

Rob, try changing your <cpu_usage> from .5 to 1 and see how it does.
...rule out any CPU bottleneck...
<cpu_usage>1</cpu_usage>

[scole of TSBT]

Here in North Carolina, they would say, the race for 7th, 8th and 9th is tighter than a tick!

Scotties, we need more GPU power! Janos, how about a hand here in the final stretch?

[Alez]

4 hours to go and we have just nipped ahead of the Aussies for 7th.

I'm now in a respectable 17th only 2 places behind the pesky colonial pirate in 15th Pity about the 500 odd tasks I have pending

[scole of TSBT]

Pesky!? We're on the same team! :)

[Alez]

I believe internal competition is just as healthy as beating the teams around us.

We are now sitting 85th in the world for Einstein but the team RAC is 39th. Going to stay on this for a while since it seems I have finally got most of it sorted. Let's see if we can grab another place or 2 in the world

For the competition TSBT 7th from 21. A very good effort I think

Individually
scole of TSBT 15th
alez [ TSBT's Pirate ] 17th
Paul of TSBT 108th A clean sweep of the podium for the pirates
UrbanFlux 207th
Robert Maclennan (TSBT) 214th
Janos of TSBT 216th
Peter Teago 305th
B McT 378th
Mark Mair 396th
Kostoglotov 447th
Dominic Young 494th
DonTex 598th
Giles 689th
Wolfsschanze 809th
DJPercival 983rd
Davie Firth 984th
Robert Hendry 1032rd
tiggsy 1183rd
Mad McMax 1188rd
Michael Richards 1231st
outcast 1270th

well done all.

[scole of TSBT]

There was less than 20,000 separated 7th, 8th and 9th. Less than 8,000 between 7th and 8th. Another close one.

Great job everyone. :bounce:

[Silver]

Another good effort by TSBT
Although I promise I'll do better next time. Missing the first day owing to my boinc update and resultant missing config files didn't get me off to a good start. My pc deciding to take a holiday yesterday didn't really help me at the end either!
Till the next time.....

[Alez]

and it took me about 3 days to get it working properly ... yet the pirates all rose to the top of the pile

[Alez]

just been doing a bit of maintenance and in a stunning bout of stupidity, I have just realised that I ran the entire competition using only 1/2 my 6990 and the last 2 days using only 2 of the 4 nVidias in Caro. What a muppet

[Silver]

just been doing a bit of maintenance and in a stunning bout of stupidity, I have just realised that I ran the entire competition using only 1/2 my 6990 and the last 2 days using only 2 of the 4 nVidias in Caro. What a muppet

Oh dear!
And you still finished 17th

[Alez]

well at least now I know why the 6990 was such a poor performer .... and here's me blaming the slow DDR2 memory on the server board instead of seeing what was right in front of me

[Silver]

Its always the obvious stuff that's hardest to see.
We've all been there!