Welcome to MilkyWay@home

General thinking about GPU computation

Questions and Answers : Windows : General thinking about GPU computation
Message board moderation

To post messages, you must log in.

AuthorMessage
Diego

Send message
Joined: 26 Dec 10
Posts: 12
Credit: 1,855,030
RAC: 0
Message 69258 - Posted: 18 Nov 2019, 16:56:15 UTC
Last modified: 18 Nov 2019, 17:05:00 UTC

Recently I upgrade my pc and I upgraded the GPU with a new card of the RX 5700 family (navy)

Reading some tasks done I noticed my new graphic card performs like an old GPU of hd 6900 family (caymann).

I was wondering on how so few was changed in Doble precision Open Cl performance till that period.

The main point is that a navy cost much more than how Cayman did on 2011.

the good point of view on my opinion is I consume at about 115 W for this computation power with a GPU load of 98-99 % per WU with only 64-65 ° C of temperature (separation with ati GPu) - thats good in preserving electronic device

I don't know instead how much does consume the cayman series.

Does someone whant to share his/her thinking?

Have a nice time there to all.

Faithfully.

Diego
ID: 69258 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Joseph Stateson
Avatar

Send message
Joined: 18 Nov 08
Posts: 291
Credit: 2,461,693,501
RAC: 0
Message 69259 - Posted: 18 Nov 2019, 20:51:27 UTC - in response to Message 69258.  


the good point of view on my opinion is I consume at about 115 W for this computation power with a GPU load of 98-99 % per WU with only 64-65 ° C of temperature (separation with ati GPu) - thats good in preserving electronic device

Does someone whant to share his/her thinking?

Diego


Yea, I can share my problem:

I have six of the "firepro"
https://milkyway.cs.rpi.edu/milkyway/top_hosts.php
There are 3 people ahead of me on the stats list. Two have a pair of VII and one has a pair of Titans. My 6 firepro (five s9000, one s9100) draw 1050 watts on a 220v line in the garage.

However,, I bought all 6 of those on ebay for under what a single VII cost. Titans are even more expensive. I have no idea what the power draw of those boards are at the wall.

When I complain to my GF that she failed to turn on the lights at night she complains about my milkyway gridcoin mining system that cannot pay for its electricity.

I may switch to SETI in which case I will slowly disappear off the stats list.
ID: 69259 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,943,471
RAC: 22,405
Message 69264 - Posted: 19 Nov 2019, 12:18:38 UTC - in response to Message 69258.  

Recently I upgrade my pc and I upgraded the GPU with a new card of the RX 5700 family (navy)

Reading some tasks done I noticed my new graphic card performs like an old GPU of hd 6900 family (caymann).

I was wondering on how so few was changed in Doble precision Open Cl performance till that period.

The main point is that a navy cost much more than how Cayman did on 2011.

the good point of view on my opinion is I consume at about 115 W for this computation power with a GPU load of 98-99 % per WU with only 64-65 ° C of temperature (separation with ati GPu) - thats good in preserving electronic device

I don't know instead how much does consume the cayman series.

Does someone whant to share his/her thinking?

Have a nice time there to all.

Faithfully.

Diego


The problem is double precision doesn't help in gaming so the double precision part of the new gpu's is getting worse and worse percentage wise compared to single precision, so yes some of the older gpu's are significantly faster. The last column in this chart is the double precision percentage:
http://www.geeks3d.com/20140305/amd-radeon-and-nvidia-geforce-fp32-fp64-gflops-table-computing/
ID: 69264 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Diego

Send message
Joined: 26 Dec 10
Posts: 12
Credit: 1,855,030
RAC: 0
Message 69266 - Posted: 19 Nov 2019, 19:52:02 UTC - in response to Message 69259.  
Last modified: 19 Nov 2019, 19:52:43 UTC

:)))) For JStateson

Funny story...

A cheer for you :)))

Diego
ID: 69266 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Diego

Send message
Joined: 26 Dec 10
Posts: 12
Credit: 1,855,030
RAC: 0
Message 69267 - Posted: 19 Nov 2019, 19:55:12 UTC - in response to Message 69264.  

Thank you Mikey for the reply :)))

Nice to have seen that list of fp 32 Fp 64 power with GPU models..

I took on mind the next time I upgrade perhaps buying something not new.
ID: 69267 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Diego

Send message
Joined: 26 Dec 10
Posts: 12
Credit: 1,855,030
RAC: 0
Message 69276 - Posted: 20 Nov 2019, 23:44:18 UTC - in response to Message 69259.  
Last modified: 20 Nov 2019, 23:50:14 UTC

Diego[/quote]

Yea, I can share my problem:

I have six of the "firepro"
https://milkyway.cs.rpi.edu/milkyway/top_hosts.php
There are 3 people ahead of me on the stats list. Two have a pair of VII and one has a pair of Titans. My 6 firepro (five s9000, one s9100) draw 1050 watts on a 220v line in the garage.
[/quote]

Hello JStateson

you are more lucky tthan me!! Surely you will climb the table score.... a day or another day...

Me instead,
spent some time in last days on seeing if my Gpu seller replaced a faulty one that emitted high sounds when working with 3d accelaration with a countefeited one.
So that I benchmarked and see if the gpu card work in accordance with its references data SP DP ....
It's told that here in Italy as in other places you have to pay attention to conterfeited devices (that work but that aren't exactly what you buy for)..
Silly isn't it?

Have a nice time there.

Diego
ID: 69276 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Vester
Avatar

Send message
Joined: 30 Dec 14
Posts: 34
Credit: 909,988,687
RAC: 918
Message 69277 - Posted: 21 Nov 2019, 1:43:43 UTC

Diego, here is an old table showing FP32 and FP64 performance for some video cards.

Don't go away, JStateson, I am trying to catch you. I have another HD 7990 on the way. I bought it on eBay for $189.99 and the seller claims that it is new. I will soon have four HD 7990s on my ASUS B250 Mining Expert rig. The four dual-GPU cards cost between $121 and $200 each.
ID: 69277 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Diego

Send message
Joined: 26 Dec 10
Posts: 12
Credit: 1,855,030
RAC: 0
Message 69282 - Posted: 21 Nov 2019, 23:48:11 UTC - in response to Message 69277.  

Hi Vester nice to meet you.

I continue in my thinking about GPU computation with milkiway separation tasks but there is something I missing..

I have a Low estimated Dp in FP 64 but more or less my system takes 100 sec to complete a Wu of at around 2xx,00 credit. (I use my full gpu and a core of my cpu)

I look some statistic of You Vester and also Jstateson. Looking the DP's abilities of yours GPU you should have more and more elaboration power in DP calculation than me.... But why both you two Vester and Jstateson take more time to complete works. What I missing? Do you share GPU time with more projects, dont' you use all the 100% of time of your GPU cards?

I just making these considerations to undetestands data and how things would work but it's sure I'm missing something in my undestanding predictions.

Faithfully

Hoping of heard of you soon.

Diego
ID: 69282 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Vester
Avatar

Send message
Joined: 30 Dec 14
Posts: 34
Credit: 909,988,687
RAC: 918
Message 69288 - Posted: 22 Nov 2019, 20:11:36 UTC - in response to Message 69282.  
Last modified: 22 Nov 2019, 21:01:32 UTC

I look some statistic of You Vester and also Jstateson. Looking the DP's abilities of yours GPU you should have more and more elaboration power in DP calculation than me.... But why both you two Vester and Jstateson take more time to complete works. What I missing? Do you share GPU time with more projects, dont' you use all the 100% of time of your GPU cards?
Many of us run more than one work unit per GPU by putting a file named app_config.xml in the milkyway.cs.rpi.edu_milkyway folder. I have six GPUs running 7 work units each. My information is shown in this image I posted for you on my Facebook page. When you look at the image, you can see that the folder I named is located on my Windows 10 computer at This PC > Local Disk (C) > Program Data > BOINC > Projects > milkyway.cs.rpi.edu_milkyway. If you do not see Program Data in your directory, you must show hidden files in File Folder Options on the View tab. To make app_config.xml you must use a text editor such as Notepad. For a four-core CPU and one GPU running four work units, one can run with:

<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
</app_config> 

On my computer, the gpu_usage value is 1/7 = 0.14 (rounded down from 0.142857143) and with the four-core processor (using six GPUs), I calculate cpu_usage as 4 divided by 42 (six GPUs multiplied by 7 WU each) = 0.095238095. I round down to the value 0.09.

Now that you have your app_config.txt file, you must save it as app_config.xml and be sure to use the drop-down selection "Save as type" "All Files."

The last thing to do is open the BOINC Manager and click on "Read config files" from the Options tab.
ID: 69288 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Vester
Avatar

Send message
Joined: 30 Dec 14
Posts: 34
Credit: 909,988,687
RAC: 918
Message 69290 - Posted: 23 Nov 2019, 2:37:39 UTC - in response to Message 69288.  

I forgot to link to the sticky topic "Run Multiple WUs on Your GPU."
ID: 69290 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Diego

Send message
Joined: 26 Dec 10
Posts: 12
Credit: 1,855,030
RAC: 0
Message 69292 - Posted: 24 Nov 2019, 7:54:06 UTC - in response to Message 69288.  

Sure Vester I've read the post you told on running more wu ontu GPU.

But I thought to do only a workunit because my Gpu is weak on Dp compared to yours and It have a GPU load of 98 %. Perhaps It manages to do more Wu but I'm not sure than I have enought power for normal Pc use or that temperatures growing up.

In any way with your example I perfectly understand how to do the xml file so I will surely give a try.

Woooww So your card take 300 second but computes 42 workunit in that time ... wooowww

Nice to hear about it.

Thank you again and again in having show how things work with this kind of computation.

Have a nice sunday.

Diego
:))
ID: 69292 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Vester
Avatar

Send message
Joined: 30 Dec 14
Posts: 34
Credit: 909,988,687
RAC: 918
Message 69297 - Posted: 24 Nov 2019, 20:12:46 UTC
Last modified: 24 Nov 2019, 20:14:12 UTC

Three hundred seconds is for 7 WUs per GPU that I was running at that time, but remember that I have 6 GPUs, so the 300 should be divided by 7 (not 42) which gives a work unit time of 300 seconds divided by 7 work units = 42.857142857 seconds/work unit.
ID: 69297 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Diego

Send message
Joined: 26 Dec 10
Posts: 12
Credit: 1,855,030
RAC: 0
Message 69299 - Posted: 24 Nov 2019, 21:43:37 UTC - in response to Message 69297.  

Hi Vester, it's better said by you. Ok I said 42 Wu thinking at the total WU onto all your GPUs. I forgot a S. More or less I think we are saing the same thing.

In any way.

I set 0,5 for both cpu and GPU in the app_config file.

Now my system should do 2 Wu with my only card and more or less it took 170-180 seconds instead of doing 1 Wu in 100 Second perhaps resources are better allocated in this way.

I hope I'll have no problem with normal use of my system....

I'd like to ask you another thing. How much do you spend more or less for KW there in USA. You surely would have good rates.

Here more or less people spend 0,25 € per Kw (taxes included (taxes....are regolar VAT, cost of the transport and managing of energy, system's charges)).

Is soo foolish even there?

Have a nice sunday.

Diego
ID: 69299 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Vester
Avatar

Send message
Joined: 30 Dec 14
Posts: 34
Credit: 909,988,687
RAC: 918
Message 69300 - Posted: 24 Nov 2019, 23:50:41 UTC
Last modified: 24 Nov 2019, 23:51:17 UTC

One Euro is equal to 1.10 US dollars. I pay about 9.86 cents per kWh, or $0.0986. Round that number to $0.10 and I am still paying less than 0.10 EUR.

I am still trying to find the best number of work units per GPU and I am now running three per GPU.
ID: 69300 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Diego

Send message
Joined: 26 Dec 10
Posts: 12
Credit: 1,855,030
RAC: 0
Message 69306 - Posted: 25 Nov 2019, 22:00:30 UTC - in response to Message 69300.  

Hello Vester,

Nice you are lucky with cost of energy, I see a statistic about the cost of energy here in Europe and I wondered how much cost energy also in other european nations....

I hope you will find the number of wu per CPU that is good for your system..

I'm not to sure 2 wu are good for mine I experiment some limited freeze for some seconds and I think to return to 1 Wu at a time...

There is a strange warning onto my elaborations log that says my estimated dp power is 325 GFlops but in realty I do 70 GFlops....

here an extrapolation of elaboration log....
3 warnings generated.
Estimated AMD GPU GFLOP/s: 351 SP GFLOP/s, 70 DP FLOP/s
Warning: Bizarrely low flops (70). Defaulting to 100....

In the folder of app_config.xml i see exists also txt files of main elaboration tasks of milkyway do you have any idea of their use?

Bye and see you soon.

Diego
ID: 69306 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Vester
Avatar

Send message
Joined: 30 Dec 14
Posts: 34
Credit: 909,988,687
RAC: 918
Message 69307 - Posted: 26 Nov 2019, 0:48:51 UTC

I am sorry that I am unable to help you, Diego. You can post this problem in the "Number crunching" forum. I have not seen a message in my app_config.xml file and I do not have any nVidia GPUs.
ID: 69307 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,943,471
RAC: 22,405
Message 69312 - Posted: 26 Nov 2019, 12:05:31 UTC - in response to Message 69306.  

Hello Vester,

Nice you are lucky with cost of energy, I see a statistic about the cost of energy here in Europe and I wondered how much cost energy also in other european nations....

I hope you will find the number of wu per CPU that is good for your system..

I'm not to sure 2 wu are good for mine I experiment some limited freeze for some seconds and I think to return to 1 Wu at a time...

There is a strange warning onto my elaborations log that says my estimated dp power is 325 GFlops but in realty I do 70 GFlops....

here an extrapolation of elaboration log....
3 warnings generated.
Estimated AMD GPU GFLOP/s: 351 SP GFLOP/s, 70 DP FLOP/s
Warning: Bizarrely low flops (70). Defaulting to 100....

In the folder of app_config.xml i see exists also txt files of main elaboration tasks of milkyway do you have any idea of their use?

Bye and see you soon.

Diego


Try rebooting the pc, your pc's are hidden so I can't tell what OS you are using but in Windows a reboot is the only way to fix that.
ID: 69312 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Diego

Send message
Joined: 26 Dec 10
Posts: 12
Credit: 1,855,030
RAC: 0
Message 69314 - Posted: 27 Nov 2019, 22:44:18 UTC - in response to Message 69312.  

mikey wrote:

Try rebooting the pc, your pc's are hidden so I can't tell what OS you are using but in Windows a reboot is the only way to fix that.


I finally returned to 1 Wu per Gpu

Temperature are more stable and seem I can use pc.

I will help project in this way that is surely better than before...

Nice to see many of you are supervolounteer...

Thank you to all for every help given to me.

Faithfully and take care..

Diego
ID: 69314 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Questions and Answers : Windows : General thinking about GPU computation

©2024 Astroinformatics Group