Message boards :
Number crunching :
Questions about iGPU and GPU card use
Message board moderation
Author | Message |
---|---|
Ron Kosinski Send message Joined: 11 Nov 00 Posts: 42 Credit: 35,074,534 RAC: 22 |
I have some questions on 2 of my BOINC boxes. I built this box about a year ago: Asus Z87 A motherboard Intel i5-4670 CPU 3.4 GHz, o/c'd to 4.1 GHz (1) Gigabyte GTX 760 GPU, o/c'd, Nvidia 335.23 drivers 650 watt power supply 2 sticks of 4 gig ram Windows 7 Pro SP1, 64 bit I just finished building this box: Asus Z87 A motherboard Intel i7-4770 CPU 3.5 GHz, o/c'd to 4.1 GHz, hyper threading disabled, could clock higher but had heat problems (>65 deg C) (1) Gigabyte GTX 760 GPU, o/c'd, Nvidia 335.23 drivers 550 watt power supply 1 stick of 8 gig ram Windows 7 Ultimate SP1, 64 bit The only projects I run are Optimized SETI (Lunatics v0.41, Cuda 50) and Einstein. BOINC Manager version is 7.2.42. Open CL installed for both CPU (v 1.2) and GPU (v 1.1). 4 boxes going through a KVM switch to one keyboard, mouse and monitor. Monitor is a HP w1907 LCD with VGA and DVI-D inputs (using VGA now due to KVM switch). I will occasionally game on one of the rigs (i5-4670) All units run 24/7, CPU and GPU temps average between 50 to 60 deg C. My first (and possibly only) question is; Is it worthwhile to move the monitor to the integrated GPU and just use the GPU card for BOINC processing? How much processing power does a monitor really steal? Using TThrottle (v7.22) I notice my GPU load never gets to 100%. I understand that I may need to install dummy monitor plugs into the GPU cards and write a Client Config file. I may also need to configure Windows look for 2 monitors. I do not have a CPU core set aside for GPU use. Would it help to set a core aside? Now, I believe, the GPU just grabs what it needs from a CPU core and the w/u processing speed slows down on that core. I hope I have supplied enough information. My apologies if this has been answered before. Any advice will be appreciated. Thanks for your help! Ron |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
My first (and possibly only) question is; Is it worthwhile to move the monitor to the integrated GPU and just use the GPU card for BOINC processing? That greatly depends on what's being done on the graphics card (the monitor simply displays an image and has no effect. It is the GPU's processing power here that we're talking about). Every app takes a little bit of processing power. Even modern web browsers utilize GPU acceleration, which can take a bit of GPU power to display graphics in web pages. The more intense the graphics, the more processing power it takes. How many workunits are you running on the GPU? Typically 2 or 3 will max out the GPU at 100%. I do not have a CPU core set aside for GPU use. Would it help to set a core aside? Now, I believe, the GPU just grabs what it needs from a CPU core and the w/u processing speed slows down on that core. Absolutely, yes. If a GPU must wait while a CPU finishes what it's doing before it can respond to a GPU's request, the overall performance of the GPU will be affected. Since the GPU is far more powerful at crunching than the CPU, you are hindering yourself by not giving the GPU it's own dedicated resource to feed it. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
On both of my i5-4670K machines I am using the Intel graphics for crunching. Initially I didn't plan on even enabling the built in GPU but then saw how efficient they were. On my gaming machine I run 4 CPU + the Intel GPU. Then on my HTPC I run 3 CPU + Intel GPU. I free up a CPU on that system because of the various thing I do slow down SETI if I run 4+1. It is also only doing MB instead of MB+AP like my gaming box. For both system I use the PCIe card for the display. No dummy plugs are needed on the Intel connectors. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Ron Kosinski Send message Joined: 11 Nov 00 Posts: 42 Credit: 35,074,534 RAC: 22 |
Thanks for the reply OzzFan, How many workunits are you running on the GPU? Typically 2 or 3 will max out the GPU at 100%. I am currently running only one workunit on the GPU. I am unsure how to enable more. Absolutely, yes. If a GPU must wait while a CPU finishes what it's doing before it can respond to a GPU's request, the overall performance of the GPU will be affected. Since the GPU is far more powerful at crunching than the CPU, you are hindering yourself by not giving the GPU it's own dedicated resource to feed it. Where would I put the code to disable one of the cores? That greatly depends on what's being done on the graphics card (the monitor simply displays an image and has no effect. It is the GPU's processing power here that we're talking about). Usually the boxes are sitting idle. I can do most of my work on the two boxes that do not have the graphics cards installed if needed, except for the occasional gaming. Thanks, Ron |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
Look this msg: http://setiathome.berkeley.edu/forum_thread.php?id=74386&postid=1495559 there are some instructions to do that. The 760 could easely handle 2, for 3 you must test. About the use of the IGPU or the GPU, remember, if you do some graphics work, the IGPU could slow down the CPU itseft due the thermal protection, on other hand the GPU could easely handle most of the graphics with allmost no loose in the performance. So unless you realy don´t do any graphic work (becareful even Browing actualy uses graphics) leave the GPU handle the task, your CPU will run a lot coolder. |
Ron Kosinski Send message Joined: 11 Nov 00 Posts: 42 Credit: 35,074,534 RAC: 22 |
Thanks for the reply, On both of my i5-4670K machines I am using the Intel graphics for crunching. Initially I didn't plan on even enabling the built in GPU but then saw how efficient they were. Are the Intel GPU's you mentioned the integrated ones, or plug-in cards? What plug-in GPU cards are in your boxes? No dummy plugs are needed on the Intel connectors. What OS are you using? Did you need to do anything special for the system to recognize both GPU's? It is also only doing MB instead of MB+AP like my gaming box. I am doing both MP and AP workunits on all 4 of my boxes. Thanks, Ron |
Ron Kosinski Send message Joined: 11 Nov 00 Posts: 42 Credit: 35,074,534 RAC: 22 |
I believe the rated temperature for my GPU cards is 90 deg C. I know running this high of a temperature would shorten the life of it. Any idea what a good long term temperature should be? Under my current conditions, my GPU temps range from about 45 deg C to 60 deg C. When I am crunching SETI, my temps (both CPU & GPU) are about 10 to 15 deg C higher than when I am crunching Einstein. I know that some of the Einstein GPU apps don't use much processing power from the GPU. Thanks, Ron |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
To avoid the temperature problem you must install a program to rise the fan speed, i use EVGA Precision (free to DL from the EVGA site and works with most NV, there are others like MSI Afterburner, ASUS Fan control, etc.) and set the fan to work faster than the defoult curve. I set my temperature target to be bellow 70C but i have the 780 not the 760. I set my max temperature on the 670 for example at 75C. In this host with a 780 for example i set the fan speed at 100% at 75C and 80% at 70C, 50% at 50C and leave the program cares about the rest. I don´t like to crunch on the IGPU because the heat produced who slow down the CPU itself. Yes E@H uses a lot less GPU than SETI, but thats because it uses a lot more memory transfers than actualy GPU crunching, but you could increse the GPU usage there by wunning more than 1 WU at a time there too. |
Ron Kosinski Send message Joined: 11 Nov 00 Posts: 42 Credit: 35,074,534 RAC: 22 |
Look this msg: http://setiathome.berkeley.edu/forum_thread.php?id=74386&postid=1495559 there are some instructions to do that. Here is a copy of the app_config.xml from the above mentioned post: <app_config> Why the different CPU usage between MB and AP? GPU usage setting at 0.5 will run 2 WU, 0.33 will run 3 WU at the same time, correct? What does the MB CPU usage setting at 0.04 do? What does the AP CPU usage setting at 1 do? I apologize for my ignorance on the subject. Is there more information in the BOINC WIKI? Thanks, Ron |
Ron Kosinski Send message Joined: 11 Nov 00 Posts: 42 Credit: 35,074,534 RAC: 22 |
Juan, To avoid the temperature problem you must install a program to rise the fan speed, i use EVGA Precision (free to DL from the EVGA site and works with most NV, there are others like MSI Afterburner, ASUS Fan control, etc.) and set the fan to work faster than the defoult curve. I am using the Gigabyte OC Guru II v1.46. I have my fan set for 80% at 40C, 90% at 50C, 100% at 60C. If I run into heat problems running 2 WU on the GPU at the same time, would reducing the o/c on the card help? Yes E@H uses a lot less GPU than SETI, but thats because it uses a lot more memory transfers than actualy GPU crunching, but you could increse the GPU usage there by running more than 1 WU at a time there too. I guess I would need to write a app_config.xml file for Einstein also? Thanks, Ron |
betreger Send message Joined: 29 Jun 99 Posts: 11361 Credit: 29,581,041 RAC: 66 |
I guess I would need to write a app_config.xml file for Einstein also? Not necessary they have made it an option on your preferences, GPU utilization factor of BRP apps |
Ron Kosinski Send message Joined: 11 Nov 00 Posts: 42 Credit: 35,074,534 RAC: 22 |
I guess I would need to write a app_config.xml file for Einstein also? Would the setting here be the same as for SETI, 0.5 would be 2 WU, 0.33 for 3 WU? Thanks, Ron |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
Why the different CPU usage between MB and AP? MB Crunch uses almost no CPU instead of AP who uses a lot of CPU to feed the GPU. GPU usage setting at 0.5 will run 2 WU, 0.33 will run 3 WU at the same time, correct? Yes but don´t push to hard on the 760 i belive 2 is the optimal setting. What does the MB CPU usage setting at 0.04 do? Actualy almost nothing, but the configuration requires a parameter, so we use a very low parameter. MB uses almost no CPU. What does the AP CPU usage setting at 1 do? That´s makes your host separate one entire core to feed each AP GPU task, the AP cruching needs one core to constantly feed the GPU. I apologize for my ignorance on the subject. Is there more information in the BOINC WIKI? Don´t worry, ask anything you need. You could find more information on: http://boinc.berkeley.edu/wiki/Client_configuration About your question on E@H, actualy E@H has the function to run more than 1 WU at a time in the project preferences: GPU utilization factor of BRP apps who works allmost in the same way (0.5 for 2, 0.33 for 3), but there is an important diference, the BRP setting works only when you DL the WU, and the app_config works instantly after you read the configuration file, so i prefear the app_config way, but any works. If I run into heat problems running 2 WU on the GPU at the same time, would reducing the o/c on the card help? Yes less OC will give you less temperature but It´s hard to belive you will be have any temperature problem with a 760 and your fan settings, even if you run 2 WU at a time. But each host is unique so you need to test. If you experience any trouble please post here so we could try to give some path to follow to fix it. You need to test but I belive you could slow down the GPU fan a little, try (not allways possible) not to use above 80% so they could last longer. Kepplers run better at <75C and even better bellow 70C. Why? I don´t know but that information comes from highly experienced sources. Sorry for the delay i was out. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Thanks for the reply, They are the integrated GPU's. Intel has not made an add-in card in quite a long time IIRC. For the add-in cards I have a Radeon HD6870 & HD5750 at the moment. Both are running Windows 7 x64 like your machines. You can get the full machine specs from the links I put in the previous post. Also if you need help on adding the Intel displays for BOINC to detect them. Try this quick write up I did with images. http://www.hal6000.com/seti/manual_display_detect.htm SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
betreger Send message Joined: 29 Jun 99 Posts: 11361 Credit: 29,581,041 RAC: 66 |
I guess I would need to write a app_config.xml file for Einstein also? Seti does not have that option on their preferences, unfortunatly, so for Seti you will have to write an app config file which I know nothing about, but the princpals are the same, so in short, yes |
Ron Kosinski Send message Joined: 11 Nov 00 Posts: 42 Credit: 35,074,534 RAC: 22 |
What does the MB CPU usage setting at 0.04 do? Is it possible to reserve a partial CPU core for GPU use? For example set the CPU usage for MB at 0.1 and for AP use 0.75. Sorry for the delay i was out. Not a problem, we all have a life outside of these forums! Thanks, Ron |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
Is it possible to reserve a partial CPU core for GPU use? For example set the CPU usage for MB at 0.1 and for AP use 0.75. Yes. But maybe we should amplify that. By 'reserve a core', we use boinc's tools to either start, or not start, a CPU task. If a task is running, it uses as core as intensively as it can: if it not running, it uses nothing. So, using your numbers: 3 x MB tasks (0.3 cores) + 1 x AP (0.75 cores) = 1.05 cores reserved in total. That would stop one CPU task from running - a fractional over-committment is allowed. And so on. |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
Not a problem, we all have a life outside of these forums! LOL - I was out again, Richard answer your question. Thanks Richard. FYI You could "play" with the settings, for example I use AP = 0.51 and MB = 0.33 on some of my hosts, so it start 1 AP + 1 MB when AP is avaiable or 3 MB when no AP works is avaiable. In other i use AP=0.5 and MB = 0.25, etc. What is the best? You need to test, that´s is the key, each host is unique. There are so many variables who interfear in the host performance, OS, OC, Driver´s, other apps runnig at the host, etc. the list is very extensive. Something who works perfect on one simply slow down other, so we could only give you a path to follow, the rest is with you. |
Ron Kosinski Send message Joined: 11 Nov 00 Posts: 42 Credit: 35,074,534 RAC: 22 |
Ok, I think I have it now. I will set the GPU usage to 0.5 for both MB and AP, that will let 2 WU run on the GPU. I will set the CPU usage for MB at 0.04, that will let all 4 CPU cores crunch with no reserve core for GPU use. I will set the CPU usage for AP at 1, that will reserve one CPU core to help the GPU with crunching. Will this cause any problems when SETI and Einstein WU are running together at the same time? I am planning on writing a app_config.xml file for Einstein as well to allow multiple WU on the GPU. How can I benchmark my processing speed when using, and not using, the app_config.xml file? Thanks, Ron |
Ron Kosinski Send message Joined: 11 Nov 00 Posts: 42 Credit: 35,074,534 RAC: 22 |
FYI You could "play" with the settings, for example I use AP = 0.51 and MB = 0.33 on some of my hosts, so it start 1 AP + 1 MB when AP is avaiable or 3 MB when no AP works is avaiable. In other i use AP=0.5 and MB = 0.25, etc. These settings you are mentioning, are they usage settings for the GPU or the CPU? Thanks, Ron |
©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.