v4.07 Released for Windows
log in

Advanced search

Message boards : News : v4.07 Released for Windows

1 · 2 · Next
Author Message
Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16497 - Posted: 16 May 2013, 18:00:40 UTC
Last modified: 16 May 2013, 18:00:53 UTC

Windows Version 4.07 has been released for Collatz, Mini_Collatz, and Solo_Collatz. This version fixes a bug in checkpointing as well as allows Catalyst 13.4 drivers to use OpenCL optimization which can increase the speed by up to 6 times versus v4.06. (Note: It still isn't as fast as the ATI 2.09 app on older GPUs though.) As usual, please report any issues.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16505 - Posted: 16 May 2013, 20:47:31 UTC

If these work OK, please let me know and I'll also add 4.07 Intel GPU versions.

Profile Peciak
Avatar
Send message
Joined: 4 Aug 09
Posts: 13
Credit: 669,955,177
RAC: 931
Message 16507 - Posted: 16 May 2013, 22:21:06 UTC

solo_collatz
ATI 7970 amd_catalyst_13.5_beta2
1xWU GPU 70%
2xWU GPU 99% -> ~700k pkt/day
BRAVO

Hans-Ulrich Hugi
Send message
Joined: 14 Sep 09
Posts: 6
Credit: 1,433,843,661
RAC: 124,166
Message 16508 - Posted: 17 May 2013, 4:47:15 UTC

All OpenCL Tasks fail immediately running an AMD69x0 With Catalyst 13.1 / Win7-X64.
Error message in "solo_collatz v4.07 (opencl_ati_100)" and "collatz v4.07 (opencl_ati_100)" is allways exaclty the same:

- Unhandled Exception Record -
Reason: Access Violation (0xc0000005) at address 0x00000000776CE4B4 write attempt to address 0x00000024

Examples:
Task 140478053 (solo_collatz_2379515330667410598248_824633720832_0) or
Task 140471613 (collatz_2379514024138359155048_824633720832_1)

Kwartet!
Send message
Joined: 21 Nov 12
Posts: 6
Credit: 322,984,444
RAC: 400,136
Message 16510 - Posted: 17 May 2013, 11:44:47 UTC
Last modified: 17 May 2013, 11:49:40 UTC

Windows 7 64bit
ATI 7770 OC to 1175 MHz, Catalyst 13.4
solo_collatz
no app_config.xml
collatz.config file as per your example
[EDIT where did I find the config file?
verbose=1
items_per_kernel=20
kernels_per_reduction=9
threads=8
sleep=1
build_options=-Werror]

GPU 97%
average runtime 2264 secs GPU + 47 secs CPU
~300,000 credits/day
(from http://boinc.thesonntags.com/collatz/results.php?userid=44569)

Thanks, I could finally upgrade Catalyst from 12.10!

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16512 - Posted: 17 May 2013, 13:38:50 UTC - in response to Message 16508.

All OpenCL Tasks fail immediately running an AMD69x0 With Catalyst 13.1 / Win7-X64.
Error message in "solo_collatz v4.07 (opencl_ati_100)" and "collatz v4.07 (opencl_ati_100)" is allways exaclty the same:

- Unhandled Exception Record -
Reason: Access Violation (0xc0000005) at address 0x00000000776CE4B4 write attempt to address 0x00000024

Examples:
Task 140478053 (solo_collatz_2379515330667410598248_824633720832_0) or
Task 140471613 (collatz_2379514024138359155048_824633720832_1)


See the thread about optimizing 4.07 applications and adding the line:
build_options=-Werror -cl-opt-disable

Per AMD Support:
I could see your kernel crashing the Kernel Analyzer when run without disabling optimizations. When using -cl-opt-disable, the kernel compiles fine. Forwarding it to relevant engineering team.


What they can't explain is why it happens to work on my 6970 and 7790 but not on all GPUs. Changing the build options as shown above will fix it although it will run slower.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16514 - Posted: 17 May 2013, 13:51:18 UTC - in response to Message 16510.

Windows 7 64bit
ATI 7770 OC to 1175 MHz, Catalyst 13.4
solo_collatz
no app_config.xml
collatz.config file as per your example
[EDIT where did I find the config file?
verbose=1
items_per_kernel=20
kernels_per_reduction=9
threads=8
sleep=1
build_options=-Werror]

GPU 97%
average runtime 2264 secs GPU + 47 secs CPU
~300,000 credits/day
(from http://boinc.thesonntags.com/collatz/results.php?userid=44569)

Thanks, I could finally upgrade Catalyst from 12.10!



The config file layout and info can be fount at:

http://boinc.thesonntags.com/collatz/forum_thread.php?id=1009

Hans-Ulrich Hugi
Send message
Joined: 14 Sep 09
Posts: 6
Credit: 1,433,843,661
RAC: 124,166
Message 16517 - Posted: 17 May 2013, 14:44:24 UTC

Thank you Slicker!

I found the article without your link and modified the config files. Because i run Collatz "out of the box" and unchanged i didn't notice that i must modify anything for a 69x0. And without to read it completely i thought the change is needed only for the 79x0 boards to run Collatz. My fault.

Anyway. Performance is not slower, i would say performance is very (!) poor:
my "Collatz 2.09 (ati13ati)" tasks finish in less than 30 minutes
but a "Collatz 4.07 (opencl_ati_100)" task is at 1,3% after 30 minutes

I'm aware that the 69x0 cards don't have the opencl power of the 79x0 boards. But compared to the old app it's a shame!

Profile Peciak
Avatar
Send message
Joined: 4 Aug 09
Posts: 13
Credit: 669,955,177
RAC: 931
Message 16518 - Posted: 17 May 2013, 18:00:32 UTC

After editing the file

solo_collatz_4.07_windows_x86_64__opencl_ati_100.config

GPU ATI7970 performance has increased to over one 1.000.000 credits/day

BRAVO

Profile Beyond
Avatar
Send message
Joined: 30 Jul 09
Posts: 213
Credit: 939,653,252
RAC: 3,969
Message 16519 - Posted: 17 May 2013, 18:29:40 UTC
Last modified: 17 May 2013, 18:33:04 UTC

Tried v4.07 on 3 different 7xxx GPUs. Times:

HD 7770 @ 1050MHz: 45:21 @ 97% usage
HD 7790 @ 1100MHz: 35:38 @ can't say, on my 7790 GPUs the Afterburner usage graph constantly bounces between 0% & 100% for all apps (WCG, Einstein, Collatz)

HD 5850 at various speeds, even underclocked always returned computation error 5 (via Boinctasks) after a second or two:

<core_client_version>7.0.64</core_client_version>
<![CDATA[
<message>
(unknown error) - exit code -1073741819 (0xc0000005)
</message>
<stderr_txt>
Collatz Conjecture v4.07 x86_64 for OpenCL 1.0 or later
Based on the AMD Brook+ kernels by Gipsel

Unhandled Exception Detected...

- Unhandled Exception Record -
Reason: Access Violation (0xc0000005) at address 0x00000000775DE4B4 write attempt to address 0x00000024

So the 77xx cards are running fine here and pretty good performance too. The 7850 does not run at all with v4.07 as far as I can tell (with my one 7850 GPU).

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16520 - Posted: 17 May 2013, 19:54:12 UTC - in response to Message 16519.

HD 7790 @ 1100MHz: 35:38 @ can't say, on my 7790 GPUs the Afterburner usage graph constantly bounces between 0% & 100% for all apps (WCG, Einstein, Collatz)


GPU-Z isn't any better. It always shows 0%. I have to look at the performance tab from Catalyst in order to see the load.

As far as being very slow. I agree. The compiler optimization really does make a difference. Hopefully the engineers can figure out the problems (more than one since others are having the same issues) and have something that works in an upcoming release.

I'm not sure that will happen though since it really does need to be tested on every single GPU type as it works on some but not on others. Or, the 32-bit integer kernel (the one in 4.07) seems to work on some HD 7xxx GPUs but not all and the 64-bit integer kernel (4.04) doesn't seem to work on any HD 7xxx GPUs (or at least not on ones that are true HD 7xxx and not just re-branded HD 6xxx models).

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16521 - Posted: 17 May 2013, 20:01:51 UTC - in response to Message 16518.

After editing the file
solo_collatz_4.07_windows_x86_64__opencl_ati_100.config

GPU ATI7970 performance has increased to over one 1.000.000 credits/day

BRAVO


Ssssshhhhhhhhh!!!! The kredit kops will raid the place! ;-)

I guess I forgot to mention that I figured it was fair to increase the credit as a motivation to run it since:
a) it uses a little more CPU in order to do the validation internally
b) there have been numerous issues with the 4.xx apps on AMD hardware
c) twice as many numbers get checked with the same number of computers since no wingman is required compared to the other collatz apps
d) it has to make up for the speed difference in OpenCL vs Brook+ apps in the collatz and mini_collatz apps
e) it puts it more in line with the credit of other OpenCL GPU apps (MW, POEM, Donate, etc.)

Profile Beyond
Avatar
Send message
Joined: 30 Jul 09
Posts: 213
Credit: 939,653,252
RAC: 3,969
Message 16522 - Posted: 17 May 2013, 20:11:49 UTC - in response to Message 16520.

HD 7790 @ 1100MHz: 35:38 @ can't say, on my 7790 GPUs the Afterburner usage graph constantly bounces between 0% & 100% for all apps (WCG, Einstein, Collatz)

GPU-Z isn't any better. It always shows 0%. I have to look at the performance tab from Catalyst in order to see the load.

As far as being very slow. I agree. The compiler optimization really does make a difference. Hopefully the engineers can figure out the problems (more than one since others are having the same issues) and have something that works in an upcoming release.

I didn't say slow. I think it's performing pretty well. Almost as fast as a 5850 with the 2.09 app.

I'm not sure that will happen though since it really does need to be tested on every single GPU type as it works on some but not on others. Or, the 32-bit integer kernel (the one in 4.07) seems to work on some HD 7xxx GPUs but not all and the 64-bit integer kernel (4.04) doesn't seem to work on any HD 7xxx GPUs (or at least not on ones that are true HD 7xxx and not just re-branded HD 6xxx models).

As far as it not working on my 7850: the UPS man just arrived with a new 7850 card (different brand) and I slammed it into one of the boxes. It's running a WU now, will post times later. So one of my 7850s works, one doesn't. Maybe a bad card, although it does run Einstein at good speed.

Profile Beyond
Avatar
Send message
Joined: 30 Jul 09
Posts: 213
Credit: 939,653,252
RAC: 3,969
Message 16523 - Posted: 17 May 2013, 22:22:45 UTC - in response to Message 16519.

Tried v4.07 on 3 different 7xxx GPUs. Times:

HD 7770 @ 1050MHz: 45:21 @ 97% usage
HD 7790 @ 1100MHz: 35:38 @ can't say, on my 7790 GPUs the Afterburner usage graph constantly bounces between 0% & 100% for all apps (WCG, Einstein, Collatz)

HD 5850 at various speeds, even underclocked always returned computation error 5 (via Boinctasks)

Have some times now for the new HD 7850, it only ran at 87% with one WU so running 2x and percent GPU usage in the upper 90s:

HD 7850 @ 1000MHz: average of 56:00 for 2 WUs, so 28:00/WU

As I remember, that's about as fast as a 5870 with the v2.09 app (with about 1/2 the power usage).

Profile tpl
Avatar
Send message
Joined: 19 Jan 10
Posts: 113
Credit: 7,453,517,172
RAC: 17,272,022
Message 16540 - Posted: 22 May 2013, 8:12:38 UTC
Last modified: 22 May 2013, 8:16:29 UTC

4.07 running without Problems on my 7990
Running 2 Units on each gpu with98-99%
i need 1900-2250 sec for a solo wu and (i don´t know
why) nearly the same time for the Collatz wu´s?!?
Win 7-64 Ati 12.1 Driver, Boinc 7.0.64
Boinc don´t want to use the Intel HD4000
____________

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16541 - Posted: 22 May 2013, 15:20:37 UTC - in response to Message 16540.

4.07 running without Problems on my 7990
Running 2 Units on each gpu with98-99%
i need 1900-2250 sec for a solo wu and (i don´t know
why) nearly the same time for the Collatz wu´s?!?
Win 7-64 Ati 12.1 Driver, Boinc 7.0.64
Boinc don´t want to use the Intel HD4000


Does BOINC identify the machine as having an Intel GPU but you don't get work, or because of the AMD GPU, does it not acknowledge the Intel GPU? I believe there are known BOINC issues when both an Intel and a non-Intel GPU exist in the same machine.

Whey the known issue? The problem with agile development is that if one truly doesn't know what future design features will be required, if becomes easy to be short sighted on the initial design making all subsequent changes look like hacks.

One of the big problems now is the mixture of OpenCL and non-OpenCL within projects or across projects and there isn't a real good way to handle that regardless of the design. We expect BOINC to track whether a device is in use or not but when the OpenCL names don't match the CUDA or CAL names, it becomes impossible. Add to that the "hack" way in which GPUs were added -- the data for ALL GPUs in a computer is stored in a single field named "SerialNum" in the BOINC database along with the BOINC client version -- and it is a wonder it works at all.

The BOINC developers never expected multiple GPUs in a single computer so the design (the single database field) had to be altered to support them. When multiple GPU support was added, it was assumed that they would be used in crossfire or sli mode and be the exact same brand and type of GPU. Not only do users have different types, they also have different brands. So, some major assumptions made in the design were totally incorrect.

The fact that enough hacks have to been done to get it to work at all is somewhat amazing. But, changing the design now to handle all those additional requirements (multiple GPUs of different brands and types running both CAL, CUDA and OpenCL applications from the same or different projects) is a major re-write. It would probably take a year to do (including testing). It would require changing both the client and the server code which means getting all the users and all the projects to upgrade. That's a whole other challenge.

Profile chip
Avatar
Send message
Joined: 8 May 11
Posts: 30
Credit: 41,295,305
RAC: 0
Message 16543 - Posted: 22 May 2013, 16:13:29 UTC - in response to Message 16540.

Boinc don´t want to use the Intel HD4000

Need to install intel_sdk_for_ocl_applications_2013_runtime_setup

zombie67 [MM]
Volunteer tester
Avatar
Send message
Joined: 3 Jul 09
Posts: 156
Credit: 612,750,117
RAC: 219
Message 16544 - Posted: 22 May 2013, 22:42:20 UTC
Last modified: 22 May 2013, 22:44:47 UTC

Hi Slicker. I know you've been busy with HW gremlins lately, but any word on solo_collatz for OSX (CPU and GPU), or for getting the Intel GPU to download?



263486 Collatz Conjecture 5/22/2013 3:43:28 PM [sched_op] Starting scheduler request
263487 Collatz Conjecture 5/22/2013 3:43:28 PM Sending scheduler request: To fetch work.
263488 Collatz Conjecture 5/22/2013 3:43:28 PM Requesting new tasks for intel_gpu
263489 Collatz Conjecture 5/22/2013 3:43:28 PM [sched_op] CPU work request: 0.00 seconds; 0.00 devices
263490 Collatz Conjecture 5/22/2013 3:43:28 PM [sched_op] intel_gpu work request: 108000.00 seconds; 1.00 devices
263491 Collatz Conjecture 5/22/2013 3:43:30 PM Scheduler request completed: got 0 new tasks
263492 Collatz Conjecture 5/22/2013 3:43:30 PM [sched_op] Server version 701
263493 Collatz Conjecture 5/22/2013 3:43:30 PM Project requested delay of 182 seconds
263494 Collatz Conjecture 5/22/2013 3:43:30 PM [sched_op] Deferring communication for 3 min 1 sec
263495 Collatz Conjecture 5/22/2013 3:43:30 PM [sched_op] Reason: requested by project
____________
Dublin, California
Team: SETI.USA

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16545 - Posted: 23 May 2013, 3:01:53 UTC - in response to Message 16544.

Hi Slicker. I know you've been busy with HW gremlins lately, but any word on solo_collatz for OSX (CPU and GPU), or for getting the Intel GPU to download?



263486 Collatz Conjecture 5/22/2013 3:43:28 PM [sched_op] Starting scheduler request
263487 Collatz Conjecture 5/22/2013 3:43:28 PM Sending scheduler request: To fetch work.
263488 Collatz Conjecture 5/22/2013 3:43:28 PM Requesting new tasks for intel_gpu
263489 Collatz Conjecture 5/22/2013 3:43:28 PM [sched_op] CPU work request: 0.00 seconds; 0.00 devices
263490 Collatz Conjecture 5/22/2013 3:43:28 PM [sched_op] intel_gpu work request: 108000.00 seconds; 1.00 devices
263491 Collatz Conjecture 5/22/2013 3:43:30 PM Scheduler request completed: got 0 new tasks
263492 Collatz Conjecture 5/22/2013 3:43:30 PM [sched_op] Server version 701
263493 Collatz Conjecture 5/22/2013 3:43:30 PM Project requested delay of 182 seconds
263494 Collatz Conjecture 5/22/2013 3:43:30 PM [sched_op] Deferring communication for 3 min 1 sec
263495 Collatz Conjecture 5/22/2013 3:43:30 PM [sched_op] Reason: requested by project



Can you PM the sched_request_boinc.thesonntags.com_collatz.xml from the BOINC Data folder right after the Intel GPU request?

zombie67 [MM]
Volunteer tester
Avatar
Send message
Joined: 3 Jul 09
Posts: 156
Credit: 612,750,117
RAC: 219
Message 16548 - Posted: 23 May 2013, 4:07:01 UTC - in response to Message 16545.

Done! Check your PM inbox.
____________
Dublin, California
Team: SETI.USA

1 · 2 · Next
Post to thread

Message boards : News : v4.07 Released for Windows


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.