Questions and Answers :
Windows :
C-50 and AMD C-60 APU with Radeon(tm) HD Graphics [Family 20 Model 2 Stepping 0] owners
Message board moderation
Author | Message |
---|---|
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
To my amazement I discovered 15 C-50's listed in the "cpus" list. And 16 AMD C-60 APU with Radeon(tm) HD Graphics [Family 20 Model 2 Stepping 0]'s. Now up to yesterday I had no idea that there were standalone motherboards of "AMD C-60 APU with Radeon(tm) HD Graphics [Family 20 Model 2 Stepping 0]" being sold. And I also learned earlier today that there were both C-50 and AMD C-60 APU with Radeon(tm) HD Graphics [Family 20 Model 2 Stepping 0] happily munching along as part of the Seti project. I also learned that this might help: app_config.xml ;; ;; This supposed to deal with AP being starved for CPU time ;; <app_config> <app> <name>setiathome_v7</name> <gpu_versions> <gpu_usage>1.00</gpu_usage> <cpu_usage>0.19</cpu_usage> </gpu_versions> </app> <app> <name>astropulse_v6</name> <gpu_versions> <gpu_usage>1.00</gpu_usage> <cpu_usage>1.1</cpu_usage> </gpu_versions> </app> </app_config> What this will do is limit your system to a single task running on the GPU. If you decide you want two setiathome_v7 tasks running change the "gpu_usage" to "0.5" without the quotes. It will then try to run two. I lack experience so far on how fast the GPU runs things. So I am starting with 1 task at a time to establish a baseline. I am running using the "high power" plan. I still allow the cpu to go as low as 5% because I have had bad experiences with overheating it if I don't. Does this thing have a cpu fan or not? Anything else you want to talk about? Tom A proud member of the OFA (Old Farts Association). |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
I also learned that this might help: app_config.xml You can't put comments like this inside app_config.xml ; is not the XML way <!-- This is XML comment --> <!-- These are many XML comments --> (and I'm not sure that even this proper XML syntax will be accepted by all BOINC versions) Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
I have gotten the comments out. I think I got that syntax from a cuda .cfg file :) As you know this particular system is pretty "dinky" in comparison to almost any other system. I would like to "tune" it a little. So I have tried to create a MB_CMDLINE.TXT file. I am assuming it goes in the project directory for Seti? mb_cmdline.txt -v 0 -sbs 223 -hp abo The -sbs is set that low because that is all the free space that is reported in the event log. I am also trying to bump up the priority but the text file(s) where this is mentioned don't give anything but "-hp" which so far, at least hasn't bumped the priority up on the process. There does seem to be 2 empty files with specific versions of the mb_cmline.txt for each of the ati processes. Should I be using them instead? And last but not least here is my current app_config.xml file in response to the note that said I had to have at least 1 cpu assigned to the gpu. <app_config> <app> <name>astropulse_v6</name> <gpu_versions> <gpu_usage>0.50</gpu_usage> <cpu_usage>1.0</cpu_usage> </gpu_versions> </app> <app> <name>setiathome_v7</name> <gpu_versions> <gpu_usage>1.0</gpu_usage> <cpu_usage>1.0</cpu_usage> </gpu_versions> </app> </app_config> I would also like to note that I have been "blessed" with 3 astropulses, 2 in progress "right now". 1 was a gpu version for which the above astropulse was setup because I read in another thread that the gpu version must have a cpu assigned or it gets "starved" for cpu cycles. So at the moment I have 2 astropulses running as cpu only versions. Those are alternating or when I suspend the gpu because I am on the system they both execute. My goal is to try to get the gpu based processes to run "faster". I know the text file I am reading calls for 2 cpus on this gpu. I might try that for a day. But I need to get those astropule work units processed before I make the 2 cpus for one gpu permanent. Thanks, Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
Ok, I found the "trick" apparently "they" are ignoring the mb_cmdline.txt file. But they don't ignore the: mb_cmdline-7.03-opencl_ati5_sah.txt or mb_cmdline-7.03-opencl_ati_sah.txt Suddenly I have "high priority" and equally suddenly I have a flashy screen and then a video driver crash. I have turned off the gpu always executing and amended the two files to this: -v 0 -sbs 223 -hp -period_iterations_num 18 The documentation file says period_iterations_num default is 20. And if a driver quits it should be "experimented with". I am guessing the slightly lower might help. It warns of performance impact. The driver crashed while I have the cpu for the gpu lowered to .24 which usually will let all cpu cores to run while servicing the gpu. I put it back to 1.0 cpu for the time being. Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
For those of you who are wondering "what do I do when it runs so hot that the turbo starts throttling from 120% down to 70%"? One solution costs about $15 at your local Walmart. Get a laptop cooler table/platform thingy. The one I bought doesn't blow that hard but it seems to get the job done and the price was right. btw, it plugs into a netbook usb port so its completely portable. For those of you who haven't stumbled over it you can find the turbo frequency by starting the "resource monitor" button from the task manager. The overview mode includes a nice display of both the cpu process % and the cpu frequency %. Now I am waiting with baited breath to see if my "through put" and my average processing times for gpu work units goes down... (it is averaging around 12 hours right now, I am shooting for maybe 8 hours) [remember just how slow this thing is]. Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
To keep it from crashing hanging/etc if you are running with gpu->1.0 and cpu->1.0 in the app_config.xml file, do this: -v 0 -hp -period_interations_num 25 This is in both the "mb_cmdline-7.03-opencl_ati5_sah.txt" and the "mb_cmdline-7.03-opencl_ati_sah.txt" files. The default for "-period_interations_num" is: 20 The documentation says if the GUI gets laggy or the video driver starts crashing change this. Later on I learned you are supposed to set it "higher". Which is why I am set there. This allows me to both run it with a priority of high "-hp" and to have it not crash as still have a useable screen. The mouse is still a little laggy but not like it was before. I have been reminded that it was even worse with cpu->0.24 and gpu->1.0 by re-reading the thread. I just had to bump the -period... upto 40. Now to see how much that slows down the gpu processing. I was actually getting at least some 4 hour processing done on the gpu tasks. I have since removed the -sbs command since it really is only useful if you are going to process more than one task at a time on the gpu. Right now I have no reason to believe that doing 2 will increase my thruput. What confuses things right now is it is still processing two of the long running astropulse wu on the cpu's. Thanks, Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
I ran across a thread (in the gpu message area) about trying to overclock a C-60 with Radeon Graphics. He really worked at it. He eventually was able to slow the cpu(s) a little bit while bumping up the GPU more than that. To get "the most" out of this machine I went through the high power, power plan and besides telling it to never go to sleep while plugged in then went through the advanced settings. I did allow the HD to shut off but turned off all the "sleep" choices eg. "never". I did allow the cpu to slow as low as 5% in both the battery and plugged in mode. I did this because recently I let another system get too hot and partially scrambled the HD by not allowing it to slow down if it got to hot. I also went through and set everything so when it is on battery it doesn't shutdown for 3-4 hours. That way I have a better chance of noticing that I need to plug it in before it shuts down. So when you are done turning on anything that looks like it will improve performance then save that all off. Also get a $15 laptop cool table/mat from Walmart. Now you should have a C-60 with Radeon that will run in "Turbo-boast" mode (about 120% of stock) nearly all the time. Thanks, Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
While I am running these 2 occurrences of Astropulse on the cpu(s) I have been getting about 8 hours per (regular) wu on the gpu. So now I am trying 2 wu's on the gpu. It maybe I can get about 8 hours for 2 wu's. If so, my thru put goes up significantly. So far I have not been able to use the -hp (high priority) command line command for the gpu. It slows my gui to the point of near un-usability even when I try the suggested compensation settings. The documentation on the memory sizing per instance command assumes that you don't need less than 156 mb per instance. Since that is all I have and less if you include "usable" I am uncertain if I should part ion the useable memory into 2 or just let the application fight it out. Hmmm, now that I am trying out 2 instances. I suppose it can't hurt to partition the memory down to 1/2 of the available memory. A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
Here is where I ended up. Set it to run one GPU task at a time with suspension of the GPU anytime I am active. I think there just isn't enough horsepower/ram space in the GPU to do it any other way. I have now installed the Lunatic package setting up the Astropulse to be processed by a hybrid of the cpu with "some" gpu. I also installed it to only use the SSE3 (I think that is the descriptor) on the presumption that is the most efficient instruction set since it builds on the lower version. Since the hardware is so skimpy I am looking for all the efficiency I can get. Will see what happens. Right now it is not getting any GPU-based processing. Thanks, Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
Ok, here is my current update (where am I now on my C-60). 1) I have segmented the Amd C-60 (aka: Aspire 0722) by putting it in a different group. (School). The school group has Seti at Home apps ticked on but NOT the Astropulse apps as well as "don't send me any other work if there is nothing for me" turned off. This basically means I will not get hit with those "calculate" forever Astropulse analysis. I don't really like watching the netbook spend nearly a week on just one task that takes between 25 hours and 48 hours on my other machines. 2) Here is my current app_config.xml file. <app_config> <app> <name>astropulse_v6</name> <gpu_versions> <gpu_usage>1.00</gpu_usage> <cpu_usage>1.00</cpu_usage> </gpu_versions> </app> <app> <name>setiathome_v7</name> <gpu_versions> <gpu_usage>1.00</gpu_usage> <cpu_usage>1.00</cpu_usage> </gpu_versions> </app> </app_config> This significantly speeds up the processing on my Seti at home Gpu tasks because a CPU is devoted fulltime to "feeding" the GPU. Apparently this advice on other Radeon-based GPU's apply here too. 3) I have the gpu preference set to suspend whenever someone active on the netbook. 4) The poor thing was still feeling overwhelmed so in the local preference tabs I decided that one cpu was sufficient to devote to Seti. I may switch that to 1.5 cpus out of 2 later on. It all depends on the responsiveness of the netbook. The consequences are that a base Seti wu is processed only if I am on the netbook, or if the netbook runs out of Gpu tasks. The Gpu tasks are taking roughly half as long as they were previously. But the netbook is now usable without completely suspending Bonic (and then forgetting to let it run again) There are text files that you may find: one starts with ap_etc.txt This one shows up if you are allowing Astropulse to be processed. It applies to GPU tasks. Mine is empty. Couldn't find anything I could put in there that was useful. Mostly made the graphics driver crash. The other two start with mb_etc.txt Those are the "multi-beam" Seti at home tasks that process in the GPU. Mine are empty. I couldn't find anything I could put in there that didn't make the video driver crash. This advice probably does NOT apply to netbooks that use any other hardware combination (including Amd cpus that use discrete external GPU's). HTH, Tom If you have any questions please let me know. Preferably on this thread. Here is the Netbook in question: http://setiathome.berkeley.edu/show_host_detail.php?hostid=7207673 A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
I am now using the preference setting to only run GPU tasks on this system. The baseline seti calculations were taking upwards to 18 hours. I am using one cpu for the gpu. As some reviewers of the Amd cpus have suggested, it takes a little time before an Amd becomes responsive when it is busy. This has made the Acer Aspire more responsive. Since I have 2-3 other much larger systems doing much more crunching than this one it is more of a "well, why not" issue. If you have any questions post it here or send me an e-mail. Tom A proud member of the OFA (Old Farts Association). |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
C-60 is the father and mother device of current OpenCL ATi apps :D Cause it's my dev host (and profiling too). To maximize its throughput try to use BrazosTweaker utility. Though it can't tweak GPU speed (on Acer netbook I use) it can slowdown and undervolt CPU part that in turn allows BIOS to rise GPU speed to 400MHz instead of usual 276MHz. Running AP in such config would be most productive (though user response can degrade due to lower CPU freqs). |
©2024 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.