APs provide larger credits on both GPUs and CPUs than MBs. You can check the status of the servers to see if APs are available from the home page under server status. When they load new tapes, the APs are split first. After that they split the MBs from the tapes. Usually they load them around Tuesdays after the weekly update, but that is not true all the time. Sometimes with long weekends, they will wait till just before the weekend starts to load the tapes. Best advice is just to see what the tapes are doing. Currently there is a large supply of APs as they load an extra amount of tapes, probably as they were expecting this outage they had planned to last longer than it actually did. I've post the link to the server status below

http://setiathome.berkeley.edu/sah_status.html


Zalster

Thanks again.

Because astropulse is a much larger wu than MB which I was used to, I think it is just the slow pace of the % changes, that make it seem like they are not "turning over".

It's not "Because astropulse is a much larger wu".
The % progress pace is the decision of the GPU app maker for performance reasons.

Every % progress update needs communication between GPU and CPU parts of the app and then to BOINC
For performance reasons GPU app only updates % progress (and checkpoints) at the end of some big loop.
This, depending on the speed of GPU, takes 2-4 minutes (or even more on slowest GPUs)

(CPU AstroPulse apps update % progress every few seconds)
 

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Thanks BilBg.

Darrell Wilcox wrote:

[quote]<coproc>

specify a GPU. Used in combination with the Anonymous platform mechanism. The element has the form

<coproc>
<type>some_name</type>
<count>1</count>
<device_nums>0 2</device_nums>
[ <peak_flops>1e10</peak_flops> ]
</coproc>

_______________________________________________________________

1) the <coproc> statement goes in cc_config.xml, correct?
2) re. peak flops: My 7770 is rated at 1280GFLOPS, what would that representation be in <peak_flops>?
3) I can't find any representation to flops of any kind in my app_info.xml file.

Also, I would need two <coproc> items, for each type of gpu (ie. 7770 and the 6570), correct?

I found my coflops which is 1.28e12.

I tried this but it continues to be overridden by the config_app of 2wu's for each of 3 gpu's.

My coflops for the older 6570's is 6.24e11.

 
I may try to give some info but mind I can't test it and never done it before!

The idea is to make HD 7770 a 'strange' kind of GPU ;) (for the 'new type GPU' I use bellow the name ATI7)
NOTE: I don't see a way for good use of app_config.xml in this case, we'll use equivalent edits in app_info.xml instead.


1) I think you may just omit the <peak_flops> (but maybe estimates will be very wrong)

If you use it - start with value from here:
http://setiathome.berkeley.edu/host_app_versions.php?hostid=7304674
AstroPulse v6 (anonymous platform, ATI GPU)
Average processing rate 219.84 GFLOPS - this is 'mix' performance of 3 GPUs / 2 tasks each

For HD 7770 you may use 300e9 for start
If you use 1280e9 tasks will be estimated to run too short


2) In cc_config.xml (see BOINC Event Log for <device_nums>)

<coproc>
  <type>ATI7</type>
  <count>1</count>
  <device_nums>2</device_nums>
</coproc>

3) For app_info.xml

in SETI@home directory (<BOINC_Data>\projects\setiathome.berkeley.edu\)
- find app_info.xml , app_config.xml and all .aistub files and make backup copy (to return to old config if needed)

- find file AP6_win_x86_SSE2_OpenCL_ATI.aistub and Edit it
- Change (in all places):
<avg_ncpus>0.04</avg_ncpus>
<max_ncpus>0.2</max_ncpus>

to:
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>1.0</max_ncpus>

and:
<count>1</count>

to:
<count>0.5</count>


- Copy the edited file AP6_win_x86_SSE2_OpenCL_ATI.aistub
- Rename the copy to AP6_win_x86_SSE2_OpenCL_ATI7.aistub
- Edit the copy and change (in all places):

        <coproc>
            <type>ATI</type>
            <count>0.5</count>
        </coproc>

to:

        <coproc>
            <type>ATI7</type>
            <count>0.33</count>
        </coproc>

Now you have to have 3 .aistub files in SETI@home directory
AKv8c_SSE2.aistub (or similar - for CPU setiathome_v7)
AP6_win_x86_SSE2_OpenCL_ATI.aistub
AP6_win_x86_SSE2_OpenCL_ATI7.aistub


Exit BOINC
Run aimerge.cmd
- this will make new app_info.xml from the 3 .aistub files

Rename app_config.xml or move it away (to not be used)

Start BOINC and pray ;) (not necessarily in this order)

 

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Thanks BilBg.
I am going to take my time to work on this. It may turn out
that the benefit I can get is not worth too much lost time
and frustration. I will let you know if any interesting issues
come up or questions I may have. Thanks again.

I keep getting these notices from the server: Your app_info file does not have a usable version of Seti@home v7.

Since I use lunatics and am not running Seti@home v7, I assume that I should just ignore these red letter postings from the server, correct?

It seems one of your Turks GPUs (AMD Radeon HD 6570 - second device = device #1) is Overclocked to 1000Mhz
http://setiathome.berkeley.edu/result.php?resultid=3646888399

I'm not sure if this is possible (seems too high for Turks to work and not hang)
What is shown in SIV about GPU-1 MHz?

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

siv shows gpu1 at 650mhz.
what is this term "turks"?

SIV and boinc agree on the numbering of the gpu devices but MSIafterburner uses a different numbering scheme. When I was attempting to use the <coproc> option in cc_config, I was mistakenly using the MSI numbering scheme. Maybe the <coproc> option in cc_config works using the correct device number for my 7770?

I am going to start building/acquiring components for a pc that will replace my old dell system. I expect it will take over a year for me to complete this. Is this venue at all useable if I were to ask questions from time to time here, of course under a separate thread title? If not, any suggestions where I might do this. I would prefer to stay here since I already have a little experience with this group. Thanks.

If you want help with ideas of building a system then I'd suggest going over to Number Crunching in Message Boards. That area tends to get more users who can give you some ideas. They also debate which systems are better. If you give them an idea of how much you are willing to spend, GPU vs CPU crunchers,Intel vs AMD chips, Motherboards, PCIe vs Risers, Cases, Coolers, AMD vs NV GPUs. There are even some thread where people post pics of their computers.

http://setiathome.berkeley.edu/forum_forum.php?id=10


Either way the choice is yours and I'm sure you will find many willing to help.

Happy Crunching...

Zalster

Thanks, Zally? Appreciate your continuing help.

what is this term "turks"?

Every GPU and CPU have Codename used internally in the developing company (Intel, AMD, NVIDIA, ...)

'Radeon HD 6000 Series' as a whole is Codenamed 'Northern Islands'
and 'Radeon HD 6500/6600' models are Codenamed 'Turks'
http://en.wikipedia.org/wiki/Radeon_HD_6000_Series#Radeon_HD_6500.2F6600

In SIV if you click [Windows] button you'll see:
Video Adapter         AMD Radeon HD 6570 (NI TURKS)


'Radeon HD 7000 Series' = 'Southern Islands'
'Radeon HD 7700' = 'Cape Verde'
http://en.wikipedia.org/wiki/Radeon_HD_7000_Series#Radeon_HD_7700

 

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Thanks BilBg. All these tips help me understand this forum better.

Is it inherent in the raw data from our radio sources that astropulse is always short of work units while MB always has more than enough work units? If not, why is it that way in Seti@home?

Hey Merle,


When the tapes are first loaded, the APs are the first things that are cut from the tapes. Since the APs provide a larger amount of credit than the Multibeams, many users prefer them. When they do become available it is a feeding frenzy to grab as many as you can before they run out. After all the tapes have the Aps cut from them, they go back and cut the Multibeams from them. MBs are not as popular so there is less demand for them. Some users don't even crunch the MBs preferring to wait for the new round of APs to come out. As such, the cache tends to build up and doesn't run out as quick. At least that is how it was explained to me, lol

Happy Crunching..


Zalster