Index  | Recent Threads  | Unanswered Threads  | Who's Active  | Guidelines  | Search
 

Quick Go »
No member browsing this thread
Thread Status: Active
Total posts in this thread: 363
Posts: 363   Pages: 37   [ Previous Page | 12 13 14 15 16 17 18 19 20 21 | Next Page ]
[ Jump to Last Post ]
Post new Thread
Author
Previous Thread This topic has been viewed 575419 times and has 362 replies Next Thread
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: HCC GPU APP HAS LAUNCHED.

I'm running 4 on sapphire 7950 OC, cpu 2500k at 4,4Ghz

I started each task manually, so I have almost constant gpu load around 90%, one task starting, one finishing and two in progress 40-50% apart.

auto fan control changing rpm all the time produced some vibration noise in lower speeds so I switched it to fixed 1870rpm (45%), gpu temp is 52-54°C

edit: btw with this setting I see each wu finishing somewhere in 90-120 secs that means one result under 30 secs avereage :)
----------------------------------------
[Edit 1 times, last edit by Former Member at Oct 14, 2012 9:36:40 AM]
[Oct 14, 2012 9:31:53 AM]   Link   Report threatening or abusive post: please login first  Go to top 
Crystal Pellet
Veteran Cruncher
Joined: May 21, 2008
Post Count: 1313
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: HCC GPU APP HAS LAUNCHED.

Guys maybe I should have been more specific. This file is best suited for high end cards like the GTX680 and Radeon7970. Do not attempt to use it if you GPU is running near 95%. If it's lower you can use the file to add more tasks. To do that change the <count>N</count parameter. 1 will run 1 task. 0.50 will run 2. 0.33 wil run 3. 0.25 will run 4. 4 gets my GPU tp 95%. That's enough for me.

This 'little Einstein' wanted to try 2 tasks on a non 'high end' card like a Nvidia GT240 with an already 99% busy GPU and 99% busy CPU-core.
To give the GPU some extra time between chunks, I added this cmdline into the co_processor part of the app_info file:
<cmdline>--gpu-target-frequency 16 --gpu-polling-mode 4 --gpu-disable-checkpointing</cmdline>
My screen is still responding quite well. To benefit the most of running two at a time, I started the tasks with 5 minutes cascade,
so when 1 task is only using the CPU at the beginning and the end of a task, the other task has full access to the GPU.
The average time when running a single task without app_info was 13'36" and when running two concurrently 21'56".
Two single tasks sequential would be over 27 minutes. Perhaps adjusting the cmdline parameters can speed this up a bit.

Edit: The cmdline wasn't used at all (ignored).
Edit again: I thought the cmdline wasn't used, because I found nothing in the client_state,
but after I removed that line and restarted BOINC running 2 GPU-tasks the screenrefresh was more sluggish and the runtime went down to about 20 minutes.
----------------------------------------

----------------------------------------
[Edit 3 times, last edit by Crystal Pellet at Oct 14, 2012 11:32:43 AM]
[Oct 14, 2012 9:35:28 AM]   Link   Report threatening or abusive post: please login first  Go to top 
mmstick
Senior Cruncher
Joined: Aug 19, 2010
Post Count: 151
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: HCC GPU APP HAS LAUNCHED.

Earlier Watts was mentioned so once again just for an idea of the power being used I give you:

MVG, 3770K at 4.8, 8G samsung green @ 2000, Samsung spinpoint hj103 plus 1 fan running together with an asus hd5870 Idles in windows at 100W

Running 8 x GFAM takes this to 210 watts

Running 1 x GPU plus 7 x GFAM takes the power use to 270 watts

Running 2 x GPU plus 6 x GFAM takes power use to 360 watts so a single asus hd5870 card would seem to use 90 watts in running

Each card should produce around 16K ppd in the above setup. Typically seeing card loaded to around 93%

I am in UK so I use 230V psu. These measurements are with Kill-a-Watt clone and are power from the wall figures.


Whether you use 230V or 120V makes no difference in how much watts the power supply consumes that you'd measure from a Kill-a-watt meter. There are measures of efficiency, is it 60%, 70%, 80%, 90%. The older the power supply the less efficient it is and the more power it will draw from the socket.


For the record this is running on a seasonic x 650. Loaded to 360 watts this puts it at around 55% load on psu which is in the sweetspot. I believe that I am right in saying that using a 230v supply gives a small improvement in the efficiency of a power supply. Those graphs you might find when looking for any psu on 80Plus site are usually produced using 115v. There are a couple of server psu's using 230v that now meet 80 PLUS Titanium: 90% 94% 96% 91% at 10, 20, 50 and 100% load respectively


No, it has nothing to do with efficiency, it only has to do with where you live. In the United States, outlets output at 115V at up to around 15 amps, in Europe you will find that your power outlets are running at 230v with a much lower amperage, but it equals the same maximum watt output. If you plug a 115V power supply into a 230V outlet it will likely explode, likewise if you plug a 230V power supply into a 115V outlet it will likely die. Some power supplies have switches to switch between the standards.
[Oct 14, 2012 12:17:48 PM]   Link   Report threatening or abusive post: please login first  Go to top 
LUFTY
Cruncher
Joined: Apr 27, 2007
Post Count: 25
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: HCC GPU APP HAS LAUNCHED.

Just a word of warning I am getting lots of invalid work units when running eight simultaneous GPU work units on a ATI 7970 so it is worth checking your own results.

I am doing some experimenting at the moment and am currently running with four simultaneous GPU work units to see if this makes a difference.

I will post an update later.

Kind regards


Lufty
[Oct 14, 2012 12:22:50 PM]   Link   Report threatening or abusive post: please login first  Go to top 
nanoprobe
Master Cruncher
Classified
Joined: Aug 29, 2008
Post Count: 2998
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: HCC GPU APP HAS LAUNCHED.

Just a word of warning I am getting lots of invalid work units when running eight simultaneous GPU work units on a ATI 7970 so it is worth checking your own results.

I am doing some experimenting at the moment and am currently running with four simultaneous GPU work units to see if this makes a difference.

I will post an update later.

Kind regards


Lufty

Running any more than the amount of tasks it takes to get your GPU load to 95% +/- is the problem. 4 maybe 5 on a 7970 would be the max. I would stick with 4.
----------------------------------------
In 1969 I took an oath to defend and protect the U S Constitution against all enemies, both foreign and Domestic. There was no expiration date.


[Oct 14, 2012 12:54:41 PM]   Link   Report threatening or abusive post: please login first  Go to top 
nanoprobe
Master Cruncher
Classified
Joined: Aug 29, 2008
Post Count: 2998
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: HCC GPU APP HAS LAUNCHED.


Guys maybe I should have been more specific. This file is best suited for high end cards like the GTX680 and Radeon7970. Do not attempt to use it if you GPU is running near 95%. If it's lower you can use the file to add more tasks. To do that change the <count>N</count parameter. 1 will run 1 task. 0.50 will run 2. 0.33 wil run 3. 0.25 will run 4. 4 gets my GPU tp 95%. That's enough for me.

When you were running one, what was the percentage being used on your card? My 6950 hits around 85 to 90% so I know I'm out trying this. With you running 4, what are your temps at along with your fan setting, if you don't mind posting it. Am looking at the 7950 for around 300 bucks at the egg. Would think I should be able to run 2 comfortably. Your thoughts on the 7950.

1 task on the 7970 gave an 80% load, 70-75 seconds runtime per task. Running 4 gave a 95% load and 130-140 seconds runtime per 4 task or about 35 seconds per task. About a 100% increase in efficiency. My temps are below 60c with fan @ 50% but this card runs caseless so the temps are skewed.
Shop around and look for specials before you buy a 7950 for $300. I bought a 7970 on an Amazon sale for $311 AMR. Newegg had the same deal but that went OOS quickly.
----------------------------------------
In 1969 I took an oath to defend and protect the U S Constitution against all enemies, both foreign and Domestic. There was no expiration date.


[Oct 14, 2012 1:10:43 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Jim1348
Veteran Cruncher
USA
Joined: Jul 13, 2009
Post Count: 1066
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: HCC GPU APP HAS LAUNCHED.

I started each task manually, so I have almost constant gpu load around 90%, one task starting, one finishing and two in progress 40-50% apart.

The problem I have had with that is that they don't stay apart. The completion times are somewhat random, and so they will drift in and out of sync. WCG needs an automated solution to ensure that the CPU-intensive portions of the work units are not in sync. That would probably mean that one is paused for a short period of time, but that is better than not being able to run two at once.
[Oct 14, 2012 1:53:38 PM]   Link   Report threatening or abusive post: please login first  Go to top 
mmstick
Senior Cruncher
Joined: Aug 19, 2010
Post Count: 151
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: HCC GPU APP HAS LAUNCHED.

Just a word of warning I am getting lots of invalid work units when running eight simultaneous GPU work units on a ATI 7970 so it is worth checking your own results.

I am doing some experimenting at the moment and am currently running with four simultaneous GPU work units to see if this makes a difference.

I will post an update later.

Kind regards


Lufty

Running any more than the amount of tasks it takes to get your GPU load to 95% +/- is the problem. 4 maybe 5 on a 7970 would be the max. I would stick with 4.


I'm certain there is much more to why his tasks are invalid than whether there is a specific load on the graphics. GPUs don't have schedulers so it is up to the CPU and program feeding the opencl driver to do the scheduling for it. Therefore, each work unit being fed to it is attempting to use the entire graphics card, but for whatever reason the program isn't optimized to run well by itself. Introducing additional tasks increases the queue to use the graphics card, but there must be a limit in the project as to how fast the task needs to respond before an error will occur. I'd assume such limit is to reduce the CPU required to run the app.
----------------------------------------
[Edit 1 times, last edit by mmstick at Oct 14, 2012 2:44:11 PM]
[Oct 14, 2012 2:42:43 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Ingleside
Veteran Cruncher
Norway
Joined: Nov 19, 2005
Post Count: 974
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: HCC GPU APP HAS LAUNCHED.

No, it has nothing to do with efficiency, it only has to do with where you live. In the United States, outlets output at 115V at up to around 15 amps, in Europe you will find that your power outlets are running at 230v with a much lower amperage, but it equals the same maximum watt output. If you plug a 115V power supply into a 230V outlet it will likely explode, likewise if you plug a 230V power supply into a 115V outlet it will likely die. Some power supplies have switches to switch between the standards.

If you've only got access to 115 V it's obvious you'll need to use this, regardless of efficiency. But, atleast in some non-residential installations in USA like example server-rooms you've also got the option of 230 V. If it's possible to cheaply get 230 V also in residential USA-homes I've no idea.

If both 115 V and 230 V is available, always buy a 230 V-capable PSU and run off 230 V, since PSU is more efficient running off 230 V than 115 V. Due to this efficiency-difference, some test-sites does their PSU-testing at both voltages, and report these. As an example, according to http://www.anandtech.com/show/6168/rosewill-fortress-platinum-450w/5 you've got:
load - 115V  -  230V
5% - 71.25% - 73.29%
10% - 84.97% - 86.85%
20% - 91.03% - 91.84%
50% - 91.64% - 92.42%
80% - 91.22% - 91.85%
100%- 90.79% - 91.08%
110%- 90.38% - 90.72%
As can be seen from the table, regardless of load, if you've got the option to run off 230 V it's always higher efficiency. As an example, if assumes your computer with GPU-crunching uses 200 W and this is with the 50%-load-numbers, the difference is 1.84 W and this means 16.14 kWh per year.


BTW, atleast around here it's 230 V at 10 A, meaning can use upto 2300 W while 115 V at 15 A is only 1725 W so it's not just the efficiency being higher at 230 V. wink
----------------------------------------


"I make so many mistakes. But then just think of all the mistakes I don't make, although I might."
[Oct 14, 2012 2:50:26 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: HCC GPU APP HAS LAUNCHED.

Rather get shocked by a 230V wire at 50 hertz than a 110-115V at 60 hertz. Latter kills and the frequency has much to do with that.
[Oct 14, 2012 3:18:45 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Posts: 363   Pages: 37   [ Previous Page | 12 13 14 15 16 17 18 19 20 21 | Next Page ]
[ Jump to Last Post ]
Post new Thread