Welcome to MilkyWay@home

Posts by mmstick

1) Message boards : Number crunching : Invalid Results (Message 62939)
Posted 3 Jan 2015 by Profile mmstick
Post:
I wouldn't worry about it at all. I've got 7550 valids, 26 invalids and 230 errors -- it's just the usual. I still make over 400,000 PPD with a 7950 regardless. If your invalid/error to valid ratio is too high, try lowering frequencies and check your temps.
2) Message boards : News : New Modfit Runs (Message 62931)
Posted 2 Jan 2015 by Profile mmstick
Post:
....the runs might not fully utilize your GPU. ...try configuring your GPU to run 2 of these at the same time...

Jake


Uhm, is there a tutorial for that?

Copy this in Notepad (not Word):


<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.05</cpu_usage>
</gpu_versions>
</app>
<app>
<name>milkyway_separation__modified_fit</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.05</cpu_usage>
</gpu_versions>
</app>
</app_config>


Save the file as "app-config.xml" and place the file here:
(Unhide your folders if you haven't already)
Program data\Boinc\Projects\MilkyWay
Restart Boinc and you'll be running two at a time.


Is this app not available on Linux? I simply get an error that it cannot find an app by that name, so it only discovers the normal milkyway app.
3) Message boards : Number crunching : Lowering required cpus (Message 62905)
Posted 29 Dec 2014 by Profile mmstick
Post:
Hi Astromancer,

App_info is ye olde waye; these days it's possible to use an app_config.xml file in the project's data directory. Mine looks like this (under Windows, save as ANSI, not UTF):

<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.05</cpu_usage>
</gpu_versions>
</app>
<app>
<name>milkyway_separation__modified_fit</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.05</cpu_usage>
</gpu_versions>
</app>
</app_config>

On a HD7950, i5-3570, W7-64 bit, other BOINC tasks claiming the cpu, my run times are between 45 and 150 seconds, depending on application (modified fit or not).


Is that 45 seconds the time it takes to process two work units, or individually (completionTime / concurrentTasks)? My Radeon HD 7950 on Arch Linux completes one every 20-22 seconds.
4) Message boards : Number crunching : want a new GPU (Message 60190)
Posted 20 Oct 2013 by Profile mmstick
Post:
You'll notice that Radeon HD 7000 series is going out of stock quickly. They are being replaced by the new R7/R9 series which are pre-overclocked and slightly improved Radeon HD 7000s. I would buy them as fast as you can as the R9 270X (1280 cores) which costs the same as a HD 7950 (1792 cores) is actually a HD 7870 (1280 cores) clocked to ~1100Mhz (or in my case 1200Mhz in Linux). You can easily clock a HD 7950 to 1100Mhz for this and other projects with minimal voltage increase.

AMD GPUs have the best DP performance, especially if you opt into the later tiers (aka HD 7900s) which don't cut DP performance.
5) Message boards : Number crunching : Linux Has Massively Superior AMD OpenCL Performance to Windows (Message 60157)
Posted 17 Oct 2013 by Profile mmstick
Post:
Also, for future reference, if you have issues even though you have OpenCL properly installed, as indicated by running 'clinfo' from a terminal, boinc needs to have permissions to use X in order to use the GPU for OpenCL. To do that, you just run 'xhost local:boinc' in a terminal. To automate the restart of BOINC and to do the xhost thing on boot, add this to /etc/rc.local:

xhost local:boinc
service boinc-client restart
6) Message boards : Number crunching : Linux Has Massively Superior AMD OpenCL Performance to Windows (Message 58507)
Posted 5 Jun 2013 by Profile mmstick
Post:
Linux + 1200Mhz HD 7950 = Estimated AMD GPU GFLOP/s: 4301 SP GFLOP/s, 1075 DP FLOP/s; or so BOINC tasks say. In Windows, I was only getting 4050 SP GFLOP/s at 1200MHz. Factoring in the significantly faster environment that Linux is over Windows, there is much to gain.
7) Message boards : Application Code Discussion : MW@home on next gen Playstation/Xbox? (Message 58506)
Posted 5 Jun 2013 by Profile mmstick
Post:
There is a huge difference between the last generation of consoles and the upcoming generation. Since the new consoles are utilizing PC architectures, you could theoretically run BOINC on the new consoles and it would be able to run in the background while gaming quite easily if such a program is made. I wouldn't rule out the possibility of getting these consoles running BOINC projects; I am more than certain that this is going to be done. BOINC devs just need to release a version of BOINC for the new consoles.
8) Message boards : Number crunching : Linux Has Massively Superior AMD OpenCL Performance to Windows (Message 58466)
Posted 1 Jun 2013 by Profile mmstick
Post:
Running your 7950 @ 975 MHz - what's the voltage you're running in Linux compared to Windows? What are your GFLOPS now in comparison to Windows? I prefer a Windows OS as I remote into all my headless machines while emulating a basic VGA monitor to ensure the GPU's remain active.


I'm running at 1200Mhz with 1.200 voltage. I'm thinking about bypassing the 1200Mhz barrier and aiming for 1300Mhz. I haven't tested to see what the lowest stable voltage is. GFLOPS will naturally be ~12% higher than in Windows as suggested from my first post here. We are talking about 300,000+ PPD in this project with 16-18 seconds to complete a single work unit.

When you also consider the fact that Linux has a significantly better optimized kernel and software, and Linux-based apps are more than likely equally optimized since they are compiled with GCC, there's also a good bonus in CPU runtime. For example, WCG's GFAM project is running 4x faster on every CPU core in Linux than on Windows. So my FX-8120 gets 4x the returned work units per day. Sadly they don't award 4x the runtime equivalent, which I think they should make their work units worth a specific amount of runtime instead of going by realtime runtime. Then again, basing effort on runtime alone is silly.

You can easily run BOINC OpenCL GPU projects in Linux without a GUI or monitor. No need for emulating VGA monitors. However, I don't run headless. I use Ubuntu on all my systems, and this is my desktop. My headless machines don't have graphics cards.
9) Message boards : Number crunching : Linux Has Massively Superior AMD OpenCL Performance to Windows (Message 58450)
Posted 31 May 2013 by Profile mmstick
Post:
It has come to my attention after quite a lot of testing that the OpenCL performance on my Radeon HD 7950, Linux is significantly higher than Windows, and somehow also allows me to overclock higher and use less electricity, thus less heat.

I'm seeing on average a 12% higher degree of computations per second on Linux (comparing Catalyst 13.4 on Linux to Catalyst 13.4 on Windows) than Windows on my 4GHz FX-8120 machine with 16GB DDR3. That, and it's consuming much less energy since the graphics card is ~10C cooler. Furthermore, desktop responsiveness while running OpenCL applications to 100% GPU usage is a lot better on Linux than Windows. I'm not sure why OpenCL is better on Linux when almost all of AMD's engineering effort is in the Windows drivers, but that is simply how it is.

The only downside of course is that Radeon HD 7xxx is unable to play any kind of games in Linux because the performance in 2D/3D is 10-300x slower than Windows, or completely broken (TF2 for example).
10) Message boards : Number crunching : Run Multiple WU's on Your GPU (Message 58391)
Posted 24 May 2013 by Profile mmstick
Post:
You should clock that GPU to 1200MHz core frequency and you'll be completing a work unit every 9 seconds (4 every 36 seconds). I just earned 350,000+ yesterday with just one 7950.
11) Message boards : Number crunching : Milky Way -- no credits in Linux? (Message 58381)
Posted 24 May 2013 by Profile mmstick
Post:
As a Linux cruncher getting a work unit uploaded every ~18 seconds, shouldn't be much problem with validation.
12) Message boards : Number crunching : Why so restrivtive small stock of tasks for me ??? (Message 58380)
Posted 24 May 2013 by Profile mmstick
Post:
It is perfectly relevant, and the date of posts does not matter at all. Hardware is no different today than it was then. It isn't okay to call them ATI and it never will be okay. Stating that a high end AMD card takes a ridiculous amount of time compared to the reality of ~18 seconds per work unit spreads misinformation.
13) Message boards : Number crunching : Why so restrivtive small stock of tasks for me ??? (Message 58371)
Posted 22 May 2013 by Profile mmstick
Post:
According to Wikipedia:
Entry-level (low-end):    73xx - 75xx
Mid-range cards:          76xx - 77xx
High-end cards:           78xx - 7970 GHz Edition
Enthusiast cards:         7990


Anyway, the exact runtimes are completely unintresting in this thread hence I didn't even bother to find it exactly out for that posting above, just written something, that was going to be about right (and probably was for HD6900 series cards and the WUs we had at that time).

And I think everybody knows what ATI cards are.


And everybody knows that ATI cards no longer exist and haven't existed for 7 years. They are AMD cards, not ATI cards. Radeon HD 6900 series is over 2 years old at this point. I've owned this 7950 for more than a year.

Wikipedia is not a reliable source. If you are familiar with the naming scheme AMD uses:

x1xx - x6xx = OEM-only
x7xx = low-end
x8xx = mid-end
x9xx = high-end
x99x = enthusiast

End of story.
14) Message boards : Number crunching : Why so restrivtive small stock of tasks for me ??? (Message 58351)
Posted 20 May 2013 by Profile mmstick
Post:
My high end 'AMD' card, Radeon HD 7950, completes 40 tasks in ~13 minutes. 20 seconds per task. A low end AMD card, HD 7770, completes 40 tasks in just ~40 minutes. Therefore, you are REALLY off on how fast AMD cards are.

My post was from 30 Jan when we had about 2x longer WUs than now, so at that time it was not that much off. Also "high-end ATI cards" include not just the fastest GPU. I don't have such GPU myself, so of course I don't know the exact times, but that was also not the point of my post.


> HD 7990 is enthusiast
> HD 7900 is high end
> HD 7800 is mid end
> HD 7700 is low end
>> HD 7950 is the slowest high end AMD card

I'm not sure where you are coming from. Also, don't call them ATI, they are AMD cards. ATI hasn't existed in years.
15) Message boards : Number crunching : Computational Error? (Message 58306)
Posted 17 May 2013 by Profile mmstick
Post:
I've been running Catalyst 13.4 via xorg edgers PPA with my Radeon HD 7950 without a single problem.
16) Message boards : Number crunching : Computational Error? (Message 58294)
Posted 16 May 2013 by Profile mmstick
Post:
I'm not sure why you'd be getting that error. It must pertain to how you installed BOINC. Using Ubuntu here, I just added a PPA that keeps up to date with the latest BOINC clients, so it will automatically upgrade to the latest BOINC over time. It worked perfectly fine after installation without needing to do anything special.

sudo add-apt-repository ppa:costamagnagianfranco/locutusofborg-ppa; sudo apt-get install boinc -y
17) Message boards : Number crunching : POEM and MW on one machine (Message 58293)
Posted 16 May 2013 by Profile mmstick
Post:
When I was at POEM, my monstrous array of GPUs would devour all work units I'd be given in under 30 minutes and be left for hours without units, so I ditched POEM at the time. I spent the last year bitcoin mining with my 7950, which earned me A good 60 coins ($6,600+ on current exchange rate). Now I'm back to running them on BOINC.
18) Message boards : Number crunching : Request help updating app_info.xml for Linux (Message 58292)
Posted 16 May 2013 by Profile mmstick
Post:
You should be using gedit to make the file, and you want to make sure it is properly aligned/formatted. You also need to make sure it has proper permissions. I'm not sure why, but when I copied an identical set of parameters from the forum which removes all the formatting, it didn't work. However, after formatting/aligning it properly it worked. The moral of the story is to write it yourself rather than copy it from here and paste it.
19) Message boards : Number crunching : Milky Way -- no credits in Linux? (Message 58291)
Posted 16 May 2013 by Profile mmstick
Post:
It might be wise to only opt into the CPU only work units and let GPUs handle the GPU work units.
20) Message boards : Number crunching : POEM and MW on one machine (Message 58272)
Posted 15 May 2013 by Profile mmstick
Post:
You don't use an app_info.xml for this project, but an app_config.xml. First I would disable n-body as it will use every CPU core to run a work unit and it does not use the graphics card to compute yet.

app_config.xml goes in the same position as app_info.xml would. Make sure you have the latest BOINC (7.065).

<app_config>
<app>
<name>milkyway</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.25</gpu_usage>
<cpu_usage>.1</cpu_usage>
</gpu_versions>
</app>
</app_config>

Make sure it's properly formatted in your text editor or else boinc will misread it.

I'm doing this from Linux with my 7950.


Next 20

©2024 Astroinformatics Group