Advanced search

Message boards : Graphics cards (GPUs) : GTX1080Ti

Author Message
Profile caffeineyellow5
Avatar
Send message
Joined: 30 Jul 14
Posts: 225
Credit: 2,658,976,345
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwat
Message 45131 - Posted: 3 Nov 2016 | 9:48:51 UTC

Waiting on the EVGA 1080 TI (maybe the Classified Edition) to get a Paschal card.
____________
1 Corinthians 9:16 "For though I preach the gospel, I have nothing to glory of: for necessity is laid upon me; yea, woe is unto me, if I preach not the gospel!"
Ephesians 6:18-20, please ;-)
http://tbc-pa.org

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,972,990,085
RAC: 161,576
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45134 - Posted: 3 Nov 2016 | 11:22:02 UTC
Last modified: 3 Nov 2016 | 11:27:47 UTC

Waiting on the EVGA 1080 TI (maybe the Classified Edition) to get a Paschal card.

Might be announced in Jan 2017 & arrive between Feb and April. So, realistically it could be 4 to 6 months before it arrives. NV could make one right now based on an NV GTX Titan, but what's the point? There is no competition and probably won't be for a while. It's also a flagship card (limited demand) and for here the big question would be, does it scale?

14nm Fin-Fet is presently being used for the GTX1050 and GTX1050Ti, while the other Pascal's are 16nm, so there is a possibility a GTX1080Ti will arrive based on the 14nm fabrication. That would fit in with a new GPU in 6months time and might mean a ~10 to 15% improvement in itself; justifying it's existence. NV are using two factories so it's conceivable that one will scale up while the other scales down and starts adapting for new Fabrication.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2006
Credit: 14,643,098,694
RAC: 1,405,364
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51059 - Posted: 21 Dec 2018 | 0:09:06 UTC
Last modified: 21 Dec 2018 | 0:15:43 UTC

Now that we can fully utilize the GPU under Linux with SWAN_SYNC=1, I can compare the performance of my cards for real.

GPU RAC GTX 1080Ti (Pascal) 1.450.000~1.550.000 GTX 1080 (Pascal) 1.150.000~1.200.000 TITAN X (Maxwell) 1.050.000~1.150.000 GTX 980Ti (Maxwell) 950.000~1.050.000

If there was a month with continuous work, these numbers could be more precise.

rod4x4
Send message
Joined: 4 Aug 14
Posts: 49
Credit: 1,381,773,744
RAC: 1,892,364
Level
Met
Scientific publications
watwatwatwatwat
Message 51062 - Posted: 22 Dec 2018 | 2:54:50 UTC - in response to Message 51059.
Last modified: 22 Dec 2018 | 3:26:11 UTC

Impressive.
What is the power draw (reported by nvidia-smi) and GPU/Mem clocks used to achieve these stats?

PappaLitto
Send message
Joined: 21 Mar 16
Posts: 486
Credit: 4,113,925,626
RAC: 4,380,831
Level
Arg
Scientific publications
watwat
Message 51063 - Posted: 22 Dec 2018 | 16:33:49 UTC - in response to Message 51062.

Impressive.
What is the power draw (reported by nvidia-smi) and GPU/Mem clocks used to achieve these stats?

How do I access this xml file on linux? I can tell you once I find it.

PappaLitto
Send message
Joined: 21 Mar 16
Posts: 486
Credit: 4,113,925,626
RAC: 4,380,831
Level
Arg
Scientific publications
watwat
Message 51064 - Posted: 22 Dec 2018 | 16:41:39 UTC
Last modified: 22 Dec 2018 | 16:50:52 UTC

I just typed in "nvidia-smi" into the terminal and it gave me a basic overview, I have a watercooled evga hybrid 1080ti and it is reporting with SWAN_SYNC enabled:

224 watts
53C
655 MB/ 11171MB RAM usage
99% utilization
2% BUS interface load

I too get around 1.5 million RAC when it is sufficiently supplied with WUs.

I have another system with two evga hybrid 1080tis and I get:

243 watts
53C
868 MB/ 11171MB RAM usage
99% utilization
2% BUS interface load

213 watts
45C
678 MB/ 11171MB RAM usage
99% utilization
2% BUS interface load

This basic overview is unfortunately not detailed enough to show clocks but in "Nvidia X Server Settings" I find that all GPUs are running at 1949Mhz clock and 10010Mhz Memory.

rod4x4
Send message
Joined: 4 Aug 14
Posts: 49
Credit: 1,381,773,744
RAC: 1,892,364
Level
Met
Scientific publications
watwatwatwatwat
Message 51068 - Posted: 23 Dec 2018 | 0:15:05 UTC - in response to Message 51064.

Thanks for the stats. A more detailed view is available using "nvidia-smi -a"
nvidia-smi also works in Windows (execute in c:\program files\nvidia corporation\nvsmi folder at a prompt)

In Linux -
"nvidia-smi --help" will give a lot more options
"nvidia-settings" is also another handy commandline utility. Good tool for scripting although the graphic version (nvidia x server settings) is easier to use.

The power options you provided is of great interest as I always try to produce the best output for the least possible power.

PappaLitto
Send message
Joined: 21 Mar 16
Posts: 486
Credit: 4,113,925,626
RAC: 4,380,831
Level
Arg
Scientific publications
watwat
Message 51071 - Posted: 23 Dec 2018 | 17:19:28 UTC

Linux with SWAN_SYNC enabled is the highest performing out of any OS. With SWAN_SYNC enabled you can achieve a true 100% usage on seemingly any CPU/GPU combination. I have noticed that the BUS interface load has significantly decreased with SWAN_SYNC enabled, most likely allowing this true 100% utilization.

Aurum
Send message
Joined: 12 Jul 17
Posts: 86
Credit: 7,193,903,068
RAC: 219,522
Level
Tyr
Scientific publications
wat
Message 51099 - Posted: 26 Dec 2018 | 18:44:51 UTC - in response to Message 51071.

How do I prove I have SWAN_SYNC enabled???

/lib/systemd/system/boinc-client.service
First line under Service: Environment="SWAN_SYNC=1"
Reboot: printenv


And SWAN_SYNC is not shown.

Zalster
Avatar
Send message
Joined: 26 Feb 14
Posts: 172
Credit: 4,013,368,076
RAC: 1,433,445
Level
Arg
Scientific publications
watwatwat
Message 51100 - Posted: 26 Dec 2018 | 19:06:17 UTC - in response to Message 51099.

Some of your computers have it enables, others don't.

Look at the stderr reports of the work units of the different computers using Linux. Unfortunately in Windows, can't tell if it's running or not.


<core_client_version>7.9.3</core_client_version>
<![CDATA[
<stderr_txt>
# CUDA Synchronisation mode: BLOCKING
# SWAN Device 1 :
# Name : GeForce GTX 1080 Ti
# ECC : Disabled
# Global mem : 11178MB
# Capability : 6.1
# PCI ID : 0000:02:00.0
# Device clock : 1683MHz
# Memory clock : 5505MHz
# Memory width : 352bit




____________

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2006
Credit: 14,643,098,694
RAC: 1,405,364
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51101 - Posted: 26 Dec 2018 | 19:32:03 UTC - in response to Message 51099.
Last modified: 26 Dec 2018 | 19:40:33 UTC

How do I prove I have SWAN_SYNC enabled???

/lib/systemd/system/boinc-client.service
First line under Service: Environment="SWAN_SYNC=1"
Reboot: printenv


And SWAN_SYNC is not shown.
The printenv shows the environment of the current user, and the boinc client service is running under a different account (named boinc; that's why we couldn't apply SWAN_SYNC for a very long time to a BOINC manager from the repo), so you can't check it this way. The correct method (beside checking the "CUDA syncronisation mode" in the stderr.txt file in /var/lib/boinc-client/slots/X folder, where X is the number of the GPUGrid app's slot, or as Zalster described) is by checking the CPU usage of the acemd binary in the task manager (either under Windows or Linux too). If the acemd binary consumes 1 CPU thread (4 CPUs: 25%, 6 CPUs: 16-17%, 8 CPUs: 12-13%, 12 CPUs 8%) then it's OK. If you see lower CPU usage, then SWAN_SYNC is not applied. The other method is to check the task list of the given host on the GPUGrid website: if the CPU time is (nearly) equals to the GPU time, then SWAN_SYNC is applied.

Post to thread

Message boards : Graphics cards (GPUs) : GTX1080Ti