Vega Frontier Edition

Message boards : Number crunching : Vega Frontier Edition
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Profile RueiKeProject Donor
Volunteer tester
Avatar

Send message
Joined: 14 Feb 16
Posts: 270
Credit: 104,714,987
RAC: 234,215
Taiwan
Message 1886964 - Posted: 30 Aug 2017, 22:27:18 UTC

I have just installed a new Vega Frontier Edition GPU in my workstation and just started processing GPU tasks. It is still air-cooled and will water block it when in goes into a new Threadripper build. I have not had much time with it, but first observation is that it doesn't like SBS set over 1024. Since it has 16G of HBM, I thought I would try 4096, but didn't work. Here is a link to the host: Nemesis
YouTube Channel: Rick's Performance Computing
ID: 1886964 · Report as offensive     Reply Quote
PappaLitto

Send message
Joined: 27 Jul 15
Posts: 6
Credit: 1,282,258
RAC: 1,571
United States
Message 1886965 - Posted: 30 Aug 2017, 22:46:32 UTC

First of all, love your videos, and hope you bring your nano build back to science where it belongs. I saw Vega topping the charts for WU time in einstein@home, so I would assume anything memory bandwidth limited it would excel in. Let me know how it does on this project.
ID: 1886965 · Report as offensive     Reply Quote
Profile RueiKeProject Donor
Volunteer tester
Avatar

Send message
Joined: 14 Feb 16
Posts: 270
Credit: 104,714,987
RAC: 234,215
Taiwan
Message 1886969 - Posted: 30 Aug 2017, 22:54:28 UTC - in response to Message 1886964.  

Here is the error I get when running with sbs at 4096:
ERROR: OpenCL kernel/call 's_clCreateBuffer(gpu_WorkData)' call failed (-61) in file ..\analyzeFuncs.cpp near line 1037.
Waiting 30 sec before restart...

Is 1024 a software limitation?
YouTube Channel: Rick's Performance Computing
ID: 1886969 · Report as offensive     Reply Quote
Profile RueiKeProject Donor
Volunteer tester
Avatar

Send message
Joined: 14 Feb 16
Posts: 270
Credit: 104,714,987
RAC: 234,215
Taiwan
Message 1886971 - Posted: 30 Aug 2017, 22:58:05 UTC - in response to Message 1886965.  

First of all, love your videos, and hope you bring your nano build back to science where it belongs. I saw Vega topping the charts for WU time in einstein@home, so I would assume anything memory bandwidth limited it would excel in. Let me know how it does on this project.

The Nano's will be back, but for now, they are working to fund some upgrades. Still some work to do with the new card. Up too early with jetlag after returning from the US trying to get some work done with it, but really need a weekend to dig into it.
YouTube Channel: Rick's Performance Computing
ID: 1886971 · Report as offensive     Reply Quote
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6467
Credit: 175,988,749
RAC: 56,579
United States
Message 1886978 - Posted: 30 Aug 2017, 23:23:08 UTC

In testing with my 8GB R9 390X using sbs values over 2048 showed no improvement. Until there is a 64-bit version of the app I wouldn't expect values of 4096 or greater to work.

It looks like the Vega 64 runs SETI@home tasks in about the same amount of time as a Fury X. My 390X runs normal AR tasks about the same as a Fury X taking ~6min.
I was hoping the Vega 56 might be a worthy replacement for my 390X, but the reviews I have seen for it are not all that great.
Since I also run Milkyway the reduced DP GFLOPs of Vega are not really desirable either.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the BP6/VP6 User Group today!
ID: 1886978 · Report as offensive     Reply Quote
Profile RueiKeProject Donor
Volunteer tester
Avatar

Send message
Joined: 14 Feb 16
Posts: 270
Credit: 104,714,987
RAC: 234,215
Taiwan
Message 1886979 - Posted: 30 Aug 2017, 23:29:58 UTC - in response to Message 1886978.  

In testing with my 8GB R9 390X using sbs values over 2048 showed no improvement. Until there is a 64-bit version of the app I wouldn't expect values of 4096 or greater to work.

It looks like the Vega 64 runs SETI@home tasks in about the same amount of time as a Fury X. My 390X runs normal AR tasks about the same as a Fury X taking ~6min.
I was hoping the Vega 56 might be a worthy replacement for my 390X, but the reviews I have seen for it are not all that great.
Since I also run Milkyway the reduced DP GFLOPs of Vega are not really desirable either.


I am having trouble seeing updated results with command line changes. Seems like in the past, the Valid task list was in a logical order. Now , I have trouble finding data for latest tasks. For tasks with sbs=4096, there is an error with a 30s wait time. I have not seen results since I have changed back to sbs=1024. Now, I have also changed to 2 tasks per GPU to see how it works out.
YouTube Channel: Rick's Performance Computing
ID: 1886979 · Report as offensive     Reply Quote
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6467
Credit: 175,988,749
RAC: 56,579
United States
Message 1887006 - Posted: 31 Aug 2017, 1:14:16 UTC - in response to Message 1886979.  

In testing with my 8GB R9 390X using sbs values over 2048 showed no improvement. Until there is a 64-bit version of the app I wouldn't expect values of 4096 or greater to work.

It looks like the Vega 64 runs SETI@home tasks in about the same amount of time as a Fury X. My 390X runs normal AR tasks about the same as a Fury X taking ~6min.
I was hoping the Vega 56 might be a worthy replacement for my 390X, but the reviews I have seen for it are not all that great.
Since I also run Milkyway the reduced DP GFLOPs of Vega are not really desirable either.


I am having trouble seeing updated results with command line changes. Seems like in the past, the Valid task list was in a logical order. Now , I have trouble finding data for latest tasks. For tasks with sbs=4096, there is an error with a 30s wait time. I have not seen results since I have changed back to sbs=1024. Now, I have also changed to 2 tasks per GPU to see how it works out.

A believe tasks are sorted by the Task ID or the Task Name. Depending on which view you are using. I think being able to sort by the reported/deadline time would be a lot more helpful.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the BP6/VP6 User Group today!
ID: 1887006 · Report as offensive     Reply Quote
Profile Keith Myers
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 2445
Credit: 185,234,304
RAC: 368,828
United States
Message 1887020 - Posted: 31 Aug 2017, 2:23:30 UTC
Last modified: 31 Aug 2017, 2:24:05 UTC

I wish for better sorting too. Tedious to find a just finished task to look at its stderr.txt output. What I have done is to use BoincTasks History function to get the proper task ID name and then go to Setiathome and input that into the search box on the Tasks tab to find that task. The website layout COULD be a lot more helpful IMHO.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1887020 · Report as offensive     Reply Quote
Profile DarrellProject Donor
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,374,299
RAC: 506
United States
Message 1887035 - Posted: 31 Aug 2017, 4:09:06 UTC - in response to Message 1886969.  

Here is the error I get when running with sbs at 4096:
ERROR: OpenCL kernel/call 's_clCreateBuffer(gpu_WorkData)' call failed (-61) in file ..\analyzeFuncs.cpp near line 1037.
Waiting 30 sec before restart...

Is 1024 a software limitation?


Yes, I get the same error if I take -sbs higher that 1504 on my RX 480. It is a limitation in the way the app was built. Kartsen on his RX 480 can take -sbs higher because he uses a Lunatics app, I run the Seti stock app and an -sbs setting of 1536 or higher causes the same error. So I use 1504 when running one task at a time on the GPU and 1280 when running two tasks at a time.

As for making tasks easier to find on the website, if you drop your cache size to zero anytime you are going to make a major change to Boinc, or your system (driver upgrade), or in your case a hardware upgrade, or experimenting with command line options. Getting tasks one at a time, until you are done making changes and the system is stable finding the result of the last task completed is very easy. Once you have the system running the way you want, return cache size back to your normal setting.
... and still I fear, and still I dare not laugh at the Mad Man!

Queen - The Prophet's Song
ID: 1887035 · Report as offensive     Reply Quote
Profile MikeProject Donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 30615
Credit: 57,680,075
RAC: 30,710
Germany
Message 1887047 - Posted: 31 Aug 2017, 7:04:38 UTC

Both Lunatics and stock app are the same source.
Stock is just a bit newer.

I will check -sbs after work but i doubt ist a Software Limitation.
With each crime and every kindness we birth our future.
ID: 1887047 · Report as offensive     Reply Quote
Profile Ageless
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 14244
Credit: 3,545,193
RAC: 1,998
Netherlands
Message 1887050 - Posted: 31 Aug 2017, 7:50:17 UTC - in response to Message 1887020.  

The website layout COULD be a lot more helpful IMHO.
Which isn't going to be fixed if you post about it in 'obscure threads' in a project forum, because then only other plain users will read it, not the possible user-developers that develop BOINC these days. So if you want something like this changed, it's got to be done on the BOINC back-end part, and that needs to be requested via Github issues: https://github.com/BOINC/boinc/issues?q=is%3Aopen+is%3Aissue.

Then best explain in detail what you now see and what you want to see. A possible developer should be able to visualize it as you describe it.
Jord

Ancient Astronaut Theorists suggest that in many ways, you can be considered an alien conspiracy!
ID: 1887050 · Report as offensive     Reply Quote
Profile Brent Norman
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 1824
Credit: 107,655,254
RAC: 460,934
Canada
Message 1887054 - Posted: 31 Aug 2017, 8:41:50 UTC - in response to Message 1887020.  

I wish for better sorting too. Tedious to find a just finished task to look at its stderr.txt output.
Look at BoincMonitor that comes with BoincTasks. It keeps the stderr files for a period of time locally.
ID: 1887054 · Report as offensive     Reply Quote
Profile MikeProject Donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 30615
Credit: 57,680,075
RAC: 30,710
Germany
Message 1887080 - Posted: 31 Aug 2017, 13:21:29 UTC - in response to Message 1886969.  

Here is the error I get when running with sbs at 4096:
ERROR: OpenCL kernel/call 's_clCreateBuffer(gpu_WorkData)' call failed (-61) in file ..\analyzeFuncs.cpp near line 1037.
Waiting 30 sec before restart...

Is 1024 a software limitation?


Its driver limitation.
The driver can only adress 3.2GB RAM and using -sbs 768 1460 MB are already allocated.

You can check in stderr.txt.

OpenCL Platform Name: AMD Accelerated Parallel Processing
Number of devices: 1
Max compute units: 64
Max work group size: 256
Max clock frequency: 1600Mhz
Max memory allocation: 3221225472
Cache type: Read/Write
Cache line size: 64
Cache size: 16384
Global memory size: 3221225472
Constant buffer size: 3221225472
Max number of constant args: 8
Local memory type: Scratchpad
Local memory size: 32768
Queue properties:
Out-of-Order: No
Name: gfx900
Vendor: Advanced Micro Devices, Inc.
Driver version: 2442.8 (PAL,HSAIL)
Version: OpenCL 1.2 AMD-APP (2442.8)
Extensions: cl_khr_fp64 cl_amd_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_fp16 cl_khr_gl_sharing cl_amd_device_attribute_query cl_amd_vec3 cl_amd_printf cl_amd_media_ops cl_amd_media_ops2 cl_amd_popcnt cl_khr_d3d10_sharing cl_khr_d3d11_sharing cl_khr_dx9_media_sharing cl_khr_image2d_from_buffer cl_khr_spir cl_khr_gl_event cl_amd_liquid_flash
With each crime and every kindness we birth our future.
ID: 1887080 · Report as offensive     Reply Quote
Profile RueiKeProject Donor
Volunteer tester
Avatar

Send message
Joined: 14 Feb 16
Posts: 270
Credit: 104,714,987
RAC: 234,215
Taiwan
Message 1887090 - Posted: 31 Aug 2017, 13:56:20 UTC - in response to Message 1887020.  

I wish for better sorting too. Tedious to find a just finished task to look at its stderr.txt output. What I have done is to use BoincTasks History function to get the proper task ID name and then go to Setiathome and input that into the search box on the Tasks tab to find that task. The website layout COULD be a lot more helpful IMHO.


I remember that in the past it was sorted by time reported, so it was very easy to monitor your systems. I think it changed when the website was redesigned. I feel like I am running nearly blind now. Not sure if anyone ever has a need to sort by task/WU number. Certainly not an ordering choice I would ever use. Maybe the best solution is to have user select which column to sort by. Easiest solution would be sort by time reported. I think this is an important enhancement, since real-time feedback would keep contributors engaged/obsessed.
YouTube Channel: Rick's Performance Computing
ID: 1887090 · Report as offensive     Reply Quote
Profile DarrellProject Donor
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,374,299
RAC: 506
United States
Message 1887101 - Posted: 31 Aug 2017, 14:44:34 UTC - in response to Message 1887080.  
Last modified: 31 Aug 2017, 14:53:49 UTC

Karsten has an XFX RX 480 8gb GPU, I have an MSI RX 480 8gb GPU. He runs a Lunatics app on Windows 10, I run Seti stock app on Windows 7. The OpenCL capabilities as reported in the task output is the same as mine, yet he is able to run with an -sbs setting of 2048 and higher with no problems, whereas if I raise the -sbs setting to 1536 or higher, I get the same error message as the OP reported on his Vega. This shows that the error is a software limitation based upon how the app was compiled for a given operating system. The 3221225472 is 3gb of OpenCl memory, so allocating the 1460 MB or 1.5gb is no problem. When I use an -sbs setting of 1504, 1577MB of OpenCL memory is allocated for buffers.

The next test for the OP is to see how the Vega reacts to an Astropulse task. I suspect it will be like what happens on our GPUs and he won't be happy.
... and still I fear, and still I dare not laugh at the Mad Man!

Queen - The Prophet's Song
ID: 1887101 · Report as offensive     Reply Quote
Profile MikeProject Donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 30615
Credit: 57,680,075
RAC: 30,710
Germany
Message 1887115 - Posted: 31 Aug 2017, 16:23:12 UTC - in response to Message 1887101.  

Karsten has an XFX RX 480 8gb GPU, I have an MSI RX 480 8gb GPU. He runs a Lunatics app on Windows 10, I run Seti stock app on Windows 7. The OpenCL capabilities as reported in the task output is the same as mine, yet he is able to run with an -sbs setting of 2048 and higher with no problems, whereas if I raise the -sbs setting to 1536 or higher, I get the same error message as the OP reported on his Vega. This shows that the error is a software limitation based upon how the app was compiled for a given operating system. The 3221225472 is 3gb of OpenCl memory, so allocating the 1460 MB or 1.5gb is no problem. When I use an -sbs setting of 1504, 1577MB of OpenCL memory is allocated for buffers.

The next test for the OP is to see how the Vega reacts to an Astropulse task. I suspect it will be like what happens on our GPUs and he won't be happy.


There is no difference between r3584 you are using and r3557 which is the version from the Lunatics installer.
Also keep in mind other settings are using GPU memory also not only the buffers.
With each crime and every kindness we birth our future.
ID: 1887115 · Report as offensive     Reply Quote
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6467
Credit: 175,988,749
RAC: 56,579
United States
Message 1887133 - Posted: 31 Aug 2017, 19:12:53 UTC - in response to Message 1887101.  
Last modified: 31 Aug 2017, 19:15:31 UTC

Karsten has an XFX RX 480 8gb GPU, I have an MSI RX 480 8gb GPU. He runs a Lunatics app on Windows 10, I run Seti stock app on Windows 7. The OpenCL capabilities as reported in the task output is the same as mine, yet he is able to run with an -sbs setting of 2048 and higher with no problems, whereas if I raise the -sbs setting to 1536 or higher, I get the same error message as the OP reported on his Vega. This shows that the error is a software limitation based upon how the app was compiled for a given operating system. The 3221225472 is 3gb of OpenCl memory, so allocating the 1460 MB or 1.5gb is no problem. When I use an -sbs setting of 1504, 1577MB of OpenCL memory is allocated for buffers.

The next test for the OP is to see how the Vega reacts to an Astropulse task. I suspect it will be like what happens on our GPUs and he won't be happy.

I would say it is more likely that it is a difference in how the drivers operate between Windows 10 & Windows 7.
I am able to use sbs vales of 2048 and higher in Windows 7 with r3557 or r3584.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the BP6/VP6 User Group today!
ID: 1887133 · Report as offensive     Reply Quote
Profile Keith Myers
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 2445
Credit: 185,234,304
RAC: 368,828
United States
Message 1887140 - Posted: 31 Aug 2017, 19:54:54 UTC - in response to Message 1887050.  

The website layout COULD be a lot more helpful IMHO.
Which isn't going to be fixed if you post about it in 'obscure threads' in a project forum, because then only other plain users will read it, not the possible user-developers that develop BOINC these days. So if you want something like this changed, it's got to be done on the BOINC back-end part, and that needs to be requested via Github issues: https://github.com/BOINC/boinc/issues?q=is%3Aopen+is%3Aissue.

Then best explain in detail what you now see and what you want to see. A possible developer should be able to visualize it as you describe it.

I never thought any lowly end-user like myself with zero programming knowledge would/should be allowed to post to a developers forum. And based on the last years major changes in the website. I am not impressed with the current design philosophy, if anything, things have regressed in usability.

So if I joined the forum, what weight could I expect from any of my proposals since I won't be able to offer a single line of code for their perusal?
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1887140 · Report as offensive     Reply Quote
Profile Keith Myers
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 2445
Credit: 185,234,304
RAC: 368,828
United States
Message 1887141 - Posted: 31 Aug 2017, 19:57:20 UTC - in response to Message 1887054.  

I wish for better sorting too. Tedious to find a just finished task to look at its stderr.txt output.
Look at BoincMonitor that comes with BoincTasks. It keeps the stderr files for a period of time locally.

Wow, never knew that was there before. I don't remember seeing it mentioned in any of the documentation. I just installed BoincTasks and let it run and have been very happy with it. Thanks for the tip, Brent. Something new to play with.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1887141 · Report as offensive     Reply Quote
Profile Ageless
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 14244
Credit: 3,545,193
RAC: 1,998
Netherlands
Message 1887265 - Posted: 1 Sep 2017, 12:55:46 UTC - in response to Message 1887140.  

So if I joined the forum, what weight could I expect from any of my proposals since I won't be able to offer a single line of code for their perusal?
Most people adding requests to BOINC Issues on Github don't know how to program, but they're able to ask in precise words what they would like to see (changed). Look, I could make an issue for you, but then when one of the developers asks specific questions, I wouldn't know how to answer them as I'm only the MITM.

Github isn't a forum. It's a bug tracking and code development platform in one.
How David added the Bootstrap cascading style sheet content management system (CSS CMS) last year isn't possible anymore under the present rules and regulations - which will be revealed to the BOINC community tonight and that David signed and reluctantly agreed with as well. But he's now also bound by those rules when it comes to making changes to Bootstrap, when they're for BOINC wide.

And while what you're requesting may be for this project only, it's very possible other projects already have done so and have the code at the ready, or that they're interested in such a change as well. So therefore I ask that you put it where it belongs.

Remember, nothing ventured, nothing gained. By just asking for it here, it'll never happen, while if you post it over there, one of the volunteer developers might find it challenging enough to try it.
Jord

Ancient Astronaut Theorists suggest that in many ways, you can be considered an alien conspiracy!
ID: 1887265 · Report as offensive     Reply Quote
1 · 2 · Next

Message boards : Number crunching : Vega Frontier Edition


 
©2017 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.