Posts by HAL9000


log in
21) Message boards : Number crunching : combo 4650 and 6570? (Message 1565104)
Posted 2 days ago by Profile HAL9000
Can I combine a 4650 and a 6570 in the same machine with these conditions:


The 4650 will be the only one plugged into the monitor. It will be on a riser.

The 6570 will be used only for lunatics.

I imagine I may be asking a lot with this question and if I don't get an answer, it's fine.

The reason I would be doing this is so that I don't interrupt the lunatics work if I am doing other work on the machine. Now, I only have the 6570 on the machine. As it works now, other work is degraded dramatically by lunatics.

If you want to use the 4650 for the display and the 6570 to crunch with. Then that can be done easily. You would just to add one line to your cc_config.xml in fact. <ignore_ati_dev>N</ignore_ati_dev> With the ID # of the 4650. Then BOINC will only use the 6570.
22) Message boards : Number crunching : Panic Mode On (89) Server Problems? (Message 1564503)
Posted 4 days ago by Profile HAL9000
total channels to do: 420 0

It's gonna be a long time before we see fresh APs.

It will be 4 days at least. If they follow their recent patterns of loading the new data sets after Tuesday maintenance.
23) Message boards : Number crunching : SETI@Home Wow!-Event 2014 (Message 1564483)
Posted 4 days ago by Profile HAL9000
Looks like it is now complete.
Thanks to SETI.Germany, & people involved, for creating the event & allowing others to participate.
24) Message boards : Number crunching : combo 4650 and 6570? (Message 1564441)
Posted 4 days ago by Profile HAL9000
So long as the Catalyst version you are using still has OpenCL support for HD4000 cards. Then they can probably both be made to work. I think Cat 12.4 or 12.6 was as high as I could go with my HD4870. I will have to double check to be sure.

It looks like since change from r1832 to r2399 your 4650 is generating only errors.

If you could try the card over on Beta with the v7.03 AP app I am sure the devs would appreciate it.
http://setiweb.ssl.berkeley.edu/beta/

Using just the 6570 in the system might be the easiest route to go.
25) Message boards : Number crunching : RAC flatlined... no clue why (Message 1564369)
Posted 4 days ago by Profile HAL9000
The stat sites show you are getting credit. When you say flat lined are you looking at the graph in BOINC? Sometimes it doesn't update correctly. So you may just want to restart BOINC. If it is not updating after that there may be an issue with your statistics_setiathome.berkeley.edu.xml not getting updated.

Edit: You can see the credit granted on your machines by looking at their valid tasks. Your machine 7356542 for example.
26) Message boards : Number crunching : GPU task limits (Message 1564358)
Posted 4 days ago by Profile HAL9000
@Falconfly

There are a essy way to solve your problem, run 2 instances of Boinc on the host, on each one you allow only one of your GPU work then you will be abble to mantain the cache filled separately with WU for each GPU´s.

Not required with the per vendor limits being fixed.
27) Message boards : Number crunching : HP printer cartridges (Message 1564356)
Posted 4 days ago by Profile HAL9000
Actually working for one of the large printer manufactures I can tell you there are a lot of shenanigans going on with ink/toner.
The IC they put in the cartridge is used to limit how much of the stuff in side is actually used. In some cases people are throwing away cartridges that still contain 50% of their contents.
Sometimes the IC will also limit which printer the cartridge can go in or give you reduced usage of the cartridge.

It is much like gaming consoles. The devices are sold at near cost, or sometimes a loss, and it is expected to make up for it in consumables.

Also something many don't realize is that the consumables that ship with the unit are generally much smaller than the replacements you buy. The toner cartridges that ship with our units, starters, are normally 1500-2000 pages. Then the replacements come in values from 5000 to 18000 depending on the printer.
28) Message boards : Number crunching : GPU task limits (Message 1564143)
Posted 4 days ago by Profile HAL9000
And hopefully remains so, until they double the numbers. Then double them again.

Probably won't happen until something major changes in the data base.

Or they score some serious hardware. Or both.

I think the db limit might be a software issue rather than hardware related. When hit a limit of out in the field the db would just barf all over itself. I don't recall the number but something in the range of 4-4.5 million seems familiar.
Maybe a 40 core 3GHz server with 1TB+ of ram would solve that if it is hardware related. I have not seem any major things pop up on the GPU equipment fundraiser site in a while. So I would guess their wishlist doesn't include hardware at the moment.

"Results out in the field" reached 10,725,146 on Sunday 4 November 2012, and was rising by a million a day. That was the point when I asked for an emergency shutdown of the splitters - it probably peaked a little higher before the email got through.

From memory, it took us about a week and a half to recover from that event.....

Details are buried somewhere in http://setiathome.berkeley.edu/forum_thread.php?id=69890 (about three-quarters of the way back to the beginning of the thread)

With people connecting data centers to SETI@home we should probably be very glad for the limits.
29) Message boards : Number crunching : GPU task limits (Message 1564131)
Posted 4 days ago by Profile HAL9000
And hopefully remains so, until they double the numbers. Then double them again.

Probably won't happen until something major changes in the data base.

Or they score some serious hardware. Or both.

I think the db limit might be a software issue rather than hardware related. When hit a limit of out in the field the db would just barf all over itself. I don't recall the number but something in the range of 4-4.5 million seems familiar.
Maybe a 40 core 3GHz server with 1TB+ of ram would solve that if it is hardware related. I have not seem any major things pop up on the GPU Users group equipment fundraiser page in a while. So I would guess their wishlist doesn't include hardware at the moment.
30) Message boards : Number crunching : GPU task limits (Message 1564062)
Posted 4 days ago by Profile HAL9000
It looks like they are working on getting the GPU limits to prevent starvation working correctly. My systems with ATI & Intel GPUs are now at the limit with 200 tasks instead of 300 as it was meant to be.

Looking at other machines it looks like it went from (limit * total # GPUs) to (limit * # GPUs from single vendor) at the moment. So those with multiple GPUs from the same vendor are still getting a higher limit as before. They might still be working on it. So it could go to (limit * # vendor GPUs). Which is what I think the original purpose was in changing the BOINC code.


It's been 100 per cpu/gpu for at least a year, from my limited knowledge. I did enable my Intel Gpu recently and picked up another 100 wu's. So I guess the limit it what it is ;-)

P.

Yes, the project set of limit of 100 CPU & 100 GPU quite a long while ago.

However somewhat recently, few months ago?, Dr. Anderson made a change to the BOINC server code. The intended effect wast so that systems with mixed GPUs would be able to fetch work for each vendor type. So that a machine with say ATI & Nvidia would not download 100 tasks for ATI & leave the Nvidia GPU idle. The change was implemented but in an unintended way which would still lead to GPU starvation. You could download 200 task for the ATI GPU and the Nvidia would still sit idle.
It has been changed so that the starvation issue looks to have been corrected. Also you will not pick up another 100 tasks for your ATI card by enabling your Intel GPU, but if you are using your iGPU it should be able to download up to 100 tasks for it to process.
If you were to have 2 ATI GPUs it looks like you would still download 200 GPU tasks... for now.
31) Message boards : Number crunching : blanked AP tasks (Message 1563949)
Posted 5 days ago by Profile HAL9000
HAL9000,

Yes, thank you, I will put my 6570 over to machine2 with my 4650 by using the riser idea that Zalster gave me. Over at 'Help - GPUApps', I asked for ideas on what my upgrade path would be and the response was the r7 265 so that is what I bought. If the price was about the same I would have gone with the r7 270 had I known about it. Also,I will try harder to go in the right direction from now on. :-) Yours truly,
merle van osdol


Interesting choice. I would have gone with the R9 270 myself. As it is the same power usage & about 25% faster.

If you add a card like that I would remove the 6570. Having a card that much slower than two faster ones may throw off the estimates. Which will could cause your system to start throwing out errors instead of getting any work done.

The general idea is to get as much done as possible, but you seem hell bent on going the other way. :P

I really meant that in the sense that before you had worked out the issues with the current setup you were looking to add a 3rd GPU to cause even more issues for yourself.
The R9 270 probably does cost more so the R7 265 might be a better card for the money.
32) Message boards : Number crunching : HP laptop power cords (Message 1563945)
Posted 5 days ago by Profile HAL9000
I saw this yesterday on engadget & found my notebook is one of the effected. I haven't been using the cord that came with it, but filled submitted to get a new one anyway.
33) Message boards : Number crunching : GPU task limits (Message 1563882)
Posted 5 days ago by Profile HAL9000
It looks like they are working on getting the GPU limits to prevent starvation working correctly. My systems with ATI & Intel GPUs are now at the limit with 200 tasks instead of 300 as it was meant to be.

Looking at other machines it looks like it went from (limit * total # GPUs) to (limit * # GPUs from single vendor) at the moment. So those with multiple GPUs from the same vendor are still getting a higher limit as before. They might still be working on it. So it could go to (limit * # vendor GPUs). Which is what I think the original purpose was in changing the BOINC code.
34) Message boards : Number crunching : blanked AP tasks (Message 1563865)
Posted 5 days ago by Profile HAL9000
Thanks TBar,
I just did that. Only one question thereto: If I have a 4 core 8 thread system, does that mean that I have to let one thread held as idle for each gpu that I have? In my 2 gpu setup that means I have to put 75% in "use at most" computer preferences. If that is the case it seems as though I would be putting quite a bit of power in an idle state. Right?

For now you can just alter the app_config file to run one at a time with 1 CPU core reserved. Hal's example will work for that;
<app_config> <app> <name>astropulse_v6</name> <gpu_versions> <gpu_usage>1.0</gpu_usage> <cpu_usage>1.0</cpu_usage> </gpu_versions> </app> </app_config>

Then add the Command Line setting to increase the GPU usage while just running one task at a time;
3) Find the ap_cmdline_win_x86_SSE2_OpenCL_ATI.txt file in your setiathome.berkeley.edu folder and paste the following line in the file;
-unroll 10 -ffa_block 4096 -ffa_block_fetch 2048 -sbs 256

See how that works for now.

With the app_config.xml setting it will automatically reserve the CPU cores/threads for the GPU. So you do not have to adjust your computing preferences. If the GPUs run out of work a CPU core will revert to running a CPU tasks if there are more to do.
If you were to have
<gpu_usage>1.0</gpu_usage>
<cpu_usage>2.0</cpu_usage>
Then 4 CPU cores/threads would be reserved while 2 GPU tasks were running.
35) Message boards : Number crunching : Show and tell your machine. Here's mine. (Message 1563777)
Posted 5 days ago by Profile HAL9000
Similar spec to my new server -- I got one half of it running today. Here's a view from the back showing the two redundant power supplies in the middle and one of the server slides. Hasn't got any GPUs (yet... :-) but there's 128 GB of RAM in each server.

Second half is now up and running. I turned hyperthreading off to see what the difference might amount to.

I will be interested to see how it compares to my 12c/24t machine. I limit it to 20 tasks at a time.
36) Message boards : Number crunching : Panic Mode On (89) Server Problems? (Message 1563756)
Posted 5 days ago by Profile HAL9000
These tapes:
29no13aa
30au13ac
30my13ac
31au13aa
31au13ac
31mr13ae

They are from 2013, and they will not be split until all older tapes from earlier years have been split and in re-crunched. Maybe you all missed it when they said that older already crunched work, may be issued again, to be re-crunched with the newer MB V7 and AP V6.

The data sets are split in alphanumerical order. Those sets just have the disadvantage of being from the last few days of the month. So and data set with a date prior to the 29th, 30th, or 31st is done before.
Unless we have several sets of data with dates near these they may not be split for some time.
These were loaded around the same time as several other sets from 2013, but with dates near the beginning of their month. So they were split and were gone very quickly.


When it comes to the 2013 tapes, you are wrong IMO. They are deliberately being stopped from being split for the very reason I stated. Just look at 29no13aa (7 completed channels), and 30au13ac (2 completed channels) , those taped were started, but stopped from splitting more.

As I recall those were being processed before more data was added. Then after maintenance splitting resumed on the lower numbered data sets. Deliberately stopping the processing on a data set doesn't seem logical when it was already done for AP.
We both have ideas about what is going on. It is all speculation unless we have actual word from the inside.
37) Message boards : Number crunching : Panic Mode On (89) Server Problems? (Message 1563742)
Posted 5 days ago by Profile HAL9000
These tapes:
29no13aa
30au13ac
30my13ac
31au13aa
31au13ac
31mr13ae

They are from 2013, and they will not be split until all older tapes from earlier years have been split and in re-crunched. Maybe you all missed it when they said that older already crunched work, may be issued again, to be re-crunched with the newer MB V7 and AP V6.

The data sets are split in alphanumerical order. Those sets just have the disadvantage of being from the last few days of the month. So and data set with a date prior to the 29th, 30th, or 31st is done before.
Unless we have several sets of data with dates near these they may not be split for some time.
These were loaded around the same time as several other sets from 2013, but with dates near the beginning of their month. So they were split and were gone very quickly.
38) Message boards : Number crunching : blanked AP tasks (Message 1563734)
Posted 5 days ago by Profile HAL9000
But if the data is changed as explained by Zalters to blank data then there are a way to know that, or change the data will be allmost impossible too.

Your machine is where this will happen. There is not a new server process to do this. Again blanking is only known after you get the data and your machine looks at it.

If that is the case i could ask why the fix was not made before, but try to do that when V7 is to close is sure a waste of time/resources.

So let´s wait for V7

I would guess that the change was only recently devised.
39) Message boards : Number crunching : blanked AP tasks (Message 1563728)
Posted 5 days ago by Profile HAL9000
But if the data is changed as explained by Zalters to blank data then there are a way to know that, or change the data will be allmost impossible too.

Your machine is where this will happen. There is not a new server process to do this. Again blanking is only known after you get the data and your machine looks at it.
40) Message boards : Number crunching : blanked AP tasks (Message 1563716)
Posted 5 days ago by Profile HAL9000
I still don´t understand why they don´t do like the Vlars on NV GPU´s, simply not sending High % Braked WU to GPU´s hosts who can´t realy handle them, sending then instead to CPU hosts who crunch this WU a lot faster. Simple better use of the avaiable resources.

My 2 cents

As it has been said several times now. Blanking is not known previous to processing. Unlike AR which is known before.


Previous 20 · Next 20

Copyright © 2014 University of California