Advanced search

Message boards : Graphics cards (GPUs) : GT 1030 — yes, it does work.

Author Message
AndyCivil
Send message
Joined: 5 Feb 14
Posts: 6
Credit: 25,848,270
RAC: 0
Level
Val
Scientific publications
watwatwatwatwat
Message 52909 - Posted: 29 Oct 2019 | 2:02:26 UTC

The recommended GPUs thread http://www.gpugrid.net/forum_thread.php?id=2507 says:

The GT1030 should also work but is below entry level & might not finish some work fast enough.
so I thought I'd give it a try, and I was successful. The result is here http://www.gpugrid.net/result.php?resultid=21466370. It took 36.23 hours, but hey, it was a cheap card, OK? Anyway, I just thought I'd confirm that for anyone else wondering.

kingcarcas
Avatar
Send message
Joined: 27 Oct 09
Posts: 18
Credit: 378,626,631
RAC: 0
Level
Asp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 52948 - Posted: 8 Nov 2019 | 19:08:34 UTC - in response to Message 52909.

That thread needs an update .
____________

Piri1974
Send message
Joined: 2 Oct 18
Posts: 38
Credit: 6,896,627
RAC: 0
Level
Ser
Scientific publications
wat
Message 52950 - Posted: 8 Nov 2019 | 20:52:45 UTC
Last modified: 8 Nov 2019 | 21:09:28 UTC

I also finished a long run work unit with a GT1030 last week (see below in n° 3).
And other long run work units with other old GPUs like a GT710, a GT730 and a GTX745:

I must point to these 2 factors though which could have badly affected the performance of these GPUs:
- The computers below were doing 'Mapping Cancer Markers' CPU units at the same time.
- The computers all have different CPUs from different generations.

1) I even had a GT710 (with slow DDR3 RAM) finishing a long work unit in time. :-)
See
http://www.gpugrid.net/result.php?resultid=21484300
CPU = AMD A8-9600 (4-core Bristol Ridge generation)

It finished it in about 116 hours,
which is less than the 5-day deadline.
But it means you need to let it run almost 24/7 to meet the deadline.

That is why I continue to be in favor of GPUGRID offering short run work units for people with entry level graphics cards.

2) This one was done with a lightly more powerful GT730 (also with slow DDR3 RAM):
http://www.gpugrid.net/result.php?resultid=21483644
This one finished in about 103 hours.
CPU = Core i5-8400 (8-core Coffee Lake generation)

3) And this one was done by another GT1030: (with faster DDR5 RAM)
http://www.gpugrid.net/result.php?resultid=21482039
It finished the job in about 34 hours.
CPU = Celeron E3300 (2-core Wolfdale generation)

4) And this one was done with a Geforce GTX745 (with slow DDR3 but on a 128-bit bus):
http://www.gpugrid.net/result.php?resultid=21483567
It finished the job in about 44 hours.
CPU = AMD FX-4320 (4-core Piledriver generation)

As you can see, all these old gpus can do long run work units and meet the deadline.
But as I said: It would be wiser of GPUGRID to continue to provide some kind of small work units so that people don't have to let their PC run 24/7, which most people are NOT prepared to do.
Milkyway (my backup project) has work units for example which even an old GT710 can finish in about 45 minutes. It even runs well on an integrated Intel GPU (OpenCl).
I am not saying you should make such TINY work units as well, but the short runs from the past were a very good idea to keep everybody on board. :-)

Hope you like these stats.
Enjoy;
Carl

Piri1974
Send message
Joined: 2 Oct 18
Posts: 38
Credit: 6,896,627
RAC: 0
Level
Ser
Scientific publications
wat
Message 52953 - Posted: 9 Nov 2019 | 10:48:11 UTC - in response to Message 52950.

A little correction:

I stated that my Core i5-8400 is an 8-core cpu.
That is not correct.
It is a 6-core cpu.
6-core, 6-thread.
My mistake. :-)

Cheers;
Carl

Profile ServicEnginIC
Avatar
Send message
Joined: 24 Sep 10
Posts: 566
Credit: 5,943,177,024
RAC: 10,879,591
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 52954 - Posted: 9 Nov 2019 | 21:16:06 UTC

Hope you like these stats.

I do! Thank you, Carl.

For those of us who are modestly contributing to the project with slow GPUs.
This may be of interest:
GPUGrid project has currently the following credit granting rules:
- For WUs returned to server before 24H from sending time: standard credit amount +50% bonus
- For WUs returned to server after 24H but before 48H from sending time: standard credit amount +25% bonus
- For WUs returned to server after 48H but before 120H from sending time: standard credit amount, no bonus
- For WUs returned to server after 120H from sending time or not returned: no credit granted - Timed out

This total time includes: WU download time + time waiting to start processing + WU processing time + result upload time + BOINC Manager waiting time to report WU.
All time you save in every mentioned stages may result in accomplishing WU deadline/bonus or not.
Standard BOINC Manager configuration may take some waiting time from WU is finished to it be reported to server.
To bypass this waiting time, it is possible to edit cc_config.xml configuration file to state that WUs be reported immediately after finished.
Common paths to this configuration file are:
- Windows: C:\ProgramData\BOINC\cc_config.xml
- Linux: /etc/boinc-client/cc_config.xml
(To be able to see ProgramData folder in Windows you first must enable in folder options to show hidden files and folders)

You can edit <options> section in existing cc_config.xml file or create it if it doesn't exist. It should look as follows:
<cc_config>
<options>
<report_results_immediately>1</report_results_immediately>
</options>
</cc_config>


When you restart BOINC Manager, in Event Log should be shown near beginning:
09/11/2019 20:39:57 | | Config: report completed tasks immediately

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 6,169
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 52956 - Posted: 9 Nov 2019 | 23:57:27 UTC

You have to keep in mind that the present long workunits are short according to their definition: "8-12 hours on fastest card". The fastest card which is working at the moment with the present GPUGrid app is the GTX 1080 Ti (and the Titan Xp), which can finish the longest of the present long workunits in 4 hours and 30 minutes. That means the long workunits (according to their definition) could be twice to almost three times longer. If the new app would be released, the fastest card would be the RTX 2080 Ti, which is about 40% faster than the GTX 1080 Ti, that would make definition of the long workunits that much longer.
In short: don't expect that these low-end cards will finish future long workunits within the 5 days deadline.

Piri1974
Send message
Joined: 2 Oct 18
Posts: 38
Credit: 6,896,627
RAC: 0
Level
Ser
Scientific publications
wat
Message 52959 - Posted: 10 Nov 2019 | 14:38:13 UTC - in response to Message 52956.
Last modified: 10 Nov 2019 | 14:42:11 UTC


In short: don't expect that these low-end cards will finish future long workunits within the 5 days deadline.


Indeed.
Thanks for pointing out that the size of work units is a RELATIVE and not an absolute number.

That is why I keep saying that the project should try to keep sending out short work units as well to keep the maximum number of volunteers on board.
Even if they have a slow GPU today, someday they will buy a new GPU.
So it is important to try to keep them aboard in the meantime.
Or by the day that they DO buy a new GPU, maybe they will have moved onto other projects which DO provide short work units for their slow GPUs of today (like Einstein, Milkyway and others).

Another way to keep people with older GPUs on board could be to give them CPU work units,.
But I have not seen any of those in a long time @GPUGRID.
Just wondering: Does anyone know if CPU work is dead at this project? For the time being?

I don't want to criticise anyone by saying these things, I just try to help. :-)

Thanks for all replies!

Have a nice day;
Carl

Aurum
Avatar
Send message
Joined: 12 Jul 17
Posts: 399
Credit: 13,024,100,382
RAC: 766,238
Level
Trp
Scientific publications
watwatwat
Message 52961 - Posted: 10 Nov 2019 | 15:29:24 UTC

I've never seen MCM1 interfere with with GPUGRID WUs. Now MIP1 has a major flaw in managing its L3 Cache and requires 5 MB of L3 Cache per WU. If you exceed that your entire CPU work flow slows by over 60%. ARP1 has a serious checkpoint reset bug.
Asteroids is nice since they produce CUDA55 WUs and only need 0.01 CPU per WU.
____________

Erich56
Send message
Joined: 1 Jan 15
Posts: 1090
Credit: 6,603,906,926
RAC: 18,783,925
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwat
Message 52962 - Posted: 10 Nov 2019 | 17:31:37 UTC - in response to Message 52959.

That is why I keep saying that the project should try to keep sending out short work units as well to keep the maximum number of volunteers on board.
Even if they have a slow GPU today, someday they will buy a new GPU.
So it is important to try to keep them aboard in the meantime.
Or by the day that they DO buy a new GPU, maybe they will have moved onto other projects which DO provide short work units for their slow GPUs of today (like Einstein, Milkyway and others).

I am afraid this is not the point from GPUGRIDs side. Most of the time, they have less tasks available than are the crunchers' capacities.
This is too bad in a way, but it's the way it is :-(


Just wondering: Does anyone know if CPU work is dead at this project? For the time being?

At least for the time being, CPU tasks seem to be dead. And I would be much surprised if there were such tasks within the near future.

Profile ServicEnginIC
Avatar
Send message
Joined: 24 Sep 10
Posts: 566
Credit: 5,943,177,024
RAC: 10,879,591
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 52964 - Posted: 10 Nov 2019 | 19:56:06 UTC

Sincerely, I miss past GPUGrid milestones.
It is evident by taking a look to following thread:

http://www.gpugrid.net/forum_thread.php?id=4880&nowrap=true#52130

Trying to summarize: Today are only working ACEMD long WUs, for Operating systems from Windows 7 to Windows 10, and only for Maxwell, Pascal, and some Kepler series GPUs. Every other combination are currently failing or not getting work.

I wish old good times (or better) may return soon.
In the meantime, all of us are trying to help the best way, are we?...

Piri1974
Send message
Joined: 2 Oct 18
Posts: 38
Credit: 6,896,627
RAC: 0
Level
Ser
Scientific publications
wat
Message 52967 - Posted: 10 Nov 2019 | 23:08:55 UTC - in response to Message 52962.

Thx for the information! :-)

Profile ServicEnginIC
Avatar
Send message
Joined: 24 Sep 10
Posts: 566
Credit: 5,943,177,024
RAC: 10,879,591
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 52972 - Posted: 11 Nov 2019 | 20:10:30 UTC

The GT1030 should also work but is below entry level & might not finish some work fast enough.

so I thought I'd give it a try, and I was successful. The result is here http://www.gpugrid.net/result.php?resultid=21466370. It took 36.23 hours, but hey, it was a cheap card, OK? Anyway, I just thought I'd confirm that for anyone else wondering.

Oh, excuse me, perhaps I've separated a bit from the original topic of this thread. I'm returning...
In the power efficiency point of view:

I was successful... It took 36.23 hours

...
The fastest card which is working at the moment with the present GPUGrid app is the GTX 1080 Ti (and the Titan Xp), which can finish the longest of the present long workunits in 4 hours and 30 minutes.

Let's calculate energy consumed by both GPUs for processing a present long WU.
For GTX 1080 Ti, rated 250W and 4,5 hours working time: 1,125 kWh
For GT 1030, rated 30W and 36,23 hours working time: 1,087 kWh

Energies consumed by both GPUs are very similar!
And GT 1030 working 24/7 is able to achieve +25% bonus.
Therefore, if you also process other project's CPU WUs 24/7 like I do, Processing GUPGrid's GPU WUs could be an excelent way to maintain your GT 1030 GPU productively occupied.

GT 1030 — yes, it does work.

Yes, it does... For the moment being.

Post to thread

Message boards : Graphics cards (GPUs) : GT 1030 — yes, it does work.

//