A little odd?

Message boards : Number crunching : A little odd?
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · Next

AuthorMessage
Profile Bernie Vine
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 9954
Credit: 103,452,613
RAC: 328
United Kingdom
Message 1982277 - Posted: 26 Feb 2019, 11:30:57 UTC

I have just transfered my now spare GTX1060 to my number 2 machine.

It is on a riser as there is no room. Everything went OK and it is now crunching on both cards;

http://setiathome.berkeley.edu/show_host_detail.php?hostid=8669914

Obviously it says 2 1060's rather than one and a 970

However the Boinc strart up log shows this.

26/02/2019 11:13:03 | | Running under account BAV
26/02/2019 11:13:08 | | CUDA: NVIDIA GPU 0: GeForce GTX 1060 3GB (driver version 419.17, CUDA version 10.1, compute capability 6.1, 3072MB, 2488MB available, 3936 GFLOPS peak)
26/02/2019 11:13:08 | | CUDA: NVIDIA GPU 1: GeForce GTX 970 (driver version 419.17, CUDA version 10.1, compute capability 5.2, 4096MB, 3381MB available, 4045 GFLOPS peak)
26/02/2019 11:13:08 | | OpenCL: NVIDIA GPU 0: GeForce GTX 1060 3GB (driver version 419.17, device version OpenCL 1.2 CUDA, 3072MB, 2488MB available, 3936 GFLOPS peak)
26/02/2019 11:13:08 | | OpenCL: NVIDIA GPU 0: GeForce GTX 1060 3GB (driver version 419.17, device version OpenCL 1.2 CUDA, 3072MB, 2488MB available, 3936 GFLOPS peak)
26/02/2019 11:13:08 | | OpenCL: NVIDIA GPU 1: GeForce GTX 970 (driver version 419.17, device version OpenCL 1.2 CUDA, 4096MB, 3381MB available, 4045 GFLOPS peak)
26/02/2019 11:13:08 | | OpenCL: NVIDIA GPU 1: GeForce GTX 970 (driver version 419.17, device version OpenCL 1.2 CUDA, 4096MB, 3381MB available, 4045 GFLOPS peak)
26/02/2019 11:13:09 | SETI@home | Found app_info.xml; using anonymous platform


Two entries for OpenCL for each card.

Is this something to worry about or just a glitch? It is only crunching one WU per card as is set.
ID: 1982277 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14652
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1982278 - Posted: 26 Feb 2019, 11:38:47 UTC - in response to Message 1982277.  

This has been discussed before: it's a relatively minor driver glitch. You probably had to update your NVidia driver when you installed the GTX 1060 (or maybe the OS did that for you). Unfortunately, the installation process didn't clean out the remains of the previous driver properly.

At a time of your choosing - probably not in the run up to maintenance - download a fresh GTX 1060 compatible driver direct from NVidia. Install that using the 'clean install' option, and all should be tidy again.

If that doesn't completely sort it out, use the DDU tool to clean the system fully, and install the new driver again.
ID: 1982278 · Report as offensive
Profile Bernie Vine
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 9954
Credit: 103,452,613
RAC: 328
United Kingdom
Message 1982280 - Posted: 26 Feb 2019, 12:03:23 UTC

Thanks Richard

This was in fact a fresh install of the latest driver, but I didn't do a clean install.

I will try that later.
ID: 1982280 · Report as offensive
Profile Ghan-buri-Ghan Mike

Send message
Joined: 27 Dec 15
Posts: 123
Credit: 92,602,985
RAC: 172
United States
Message 1982348 - Posted: 26 Feb 2019, 23:59:37 UTC - in response to Message 1982280.  

I run into this all the time. I've just found it easier to stick a label on the case so I can keep tracks of the 30 odd GPUs I run. Some systems have multiple cards, some don't. Some of the multi GPU systems are "pure" others have a mix of cards. What shows up is usually the "priority" PCIe slot (if you have a mix of connectors), sometimes its the slot closest to the processor. At the end of the day, they all crunch in accordance with their capability
ID: 1982348 · Report as offensive
Profile Bernie Vine
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 9954
Credit: 103,452,613
RAC: 328
United Kingdom
Message 1982431 - Posted: 27 Feb 2019, 8:54:41 UTC
Last modified: 27 Feb 2019, 8:59:19 UTC

Well seems like I had a worse problem, I checked the WU's that had been done just before the outage, to see all of them were being done on device 0 the 1060.

Sure enough checked with GPU-Z and the 970 was doing nothing, both WU's were running on the 1060. (even though Bionc Tasks said they were running on d0 and d1)

So I did a clean install of the 419.17 drivers, still had the double OpenCL appearing.

So went into safe mode and removed the Nvidia drivers with DDU. Rebooted, and did another install of the 419.17 drivers. This failed before it even got to the options, tired it again same result.

So I decided to try the previous drivers, 418.81, this got to the options screen where it said I had 388.01 drivers installed, no idea where they came from Windows I assume.

I went ahead and installed 418.81 but it still had the same double OpenCL, so as this was turning into a wast of time I removed the 1060 and everything is back to normal.

So has anyone else seen this before where Boinc sees both cards and identifies them as d0 and d1, yet both WU's were running on d0, more than a little odd.

So now I have a GTX 1060 3Gb that I can't use. My only spare machines are 10 year old Dell's that cost too much to run all the time.
ID: 1982431 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34802
Credit: 261,360,520
RAC: 489
Australia
Message 1982436 - Posted: 27 Feb 2019, 9:22:29 UTC

Bernie you need a cc_config.xml with the use all GPU option set in your BOINC data folder when using mismatched GPU's.

<cc_config>
 <options>
   <use_all_gpus>1</use_all_gpus>
 </options> 
</cc_config>


Cheers.
ID: 1982436 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13739
Credit: 208,696,464
RAC: 304
Australia
Message 1982437 - Posted: 27 Feb 2019, 9:25:06 UTC - in response to Message 1982431.  
Last modified: 27 Feb 2019, 9:25:36 UTC

So has anyone else seen this before where Boinc sees both cards and identifies them as d0 and d1, yet both WU's were running on d0, more than a little odd.

I could understand the system not using the older card at all, but not it running 2 WUs on the one card unless setup to do so.
Making use of <use_all_gpus>1</use_all_gpus> in cc_config.xml would (or at least should) make the older GPU available for use by BOINC.

Editv- Wiggo beat me to it.
Grant
Darwin NT
ID: 1982437 · Report as offensive
Profile Bernie Vine
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 9954
Credit: 103,452,613
RAC: 328
United Kingdom
Message 1982438 - Posted: 27 Feb 2019, 9:28:48 UTC - in response to Message 1982436.  
Last modified: 27 Feb 2019, 9:37:14 UTC

Bernie you need a cc_config.xml with the use all GPU option set in your BOINC data folder when using mismatched GPU's.

<cc_config>
 <options>
   <use_all_gpus>1</use_all_gpus>
 </options> 
</cc_config>


Cheers.

Sorry should have mentioned that I already did that before I started.

Boinc is correctly identifying both GPU's, however the OpenCL driver appears twice for each card no mater how I install the drivers.

As my first post:

26/02/2019 11:13:03 | | Running under account BAV
26/02/2019 11:13:08 | | CUDA: NVIDIA GPU 0: GeForce GTX 1060 3GB (driver version 419.17, CUDA version 10.1, compute capability 6.1, 3072MB, 2488MB available, 3936 GFLOPS peak)
26/02/2019 11:13:08 | | CUDA: NVIDIA GPU 1: GeForce GTX 970 (driver version 419.17, CUDA version 10.1, compute capability 5.2, 4096MB, 3381MB available, 4045 GFLOPS peak)
26/02/2019 11:13:08 | | OpenCL: NVIDIA GPU 0: GeForce GTX 1060 3GB (driver version 419.17, device version OpenCL 1.2 CUDA, 3072MB, 2488MB available, 3936 GFLOPS peak)
26/02/2019 11:13:08 | | OpenCL: NVIDIA GPU 0: GeForce GTX 1060 3GB (driver version 419.17, device version OpenCL 1.2 CUDA, 3072MB, 2488MB available, 3936 GFLOPS peak)
26/02/2019 11:13:08 | | OpenCL: NVIDIA GPU 1: GeForce GTX 970 (driver version 419.17, device version OpenCL 1.2 CUDA, 4096MB, 3381MB available, 4045 GFLOPS peak)
26/02/2019 11:13:08 | | OpenCL: NVIDIA GPU 1: GeForce GTX 970 (driver version 419.17, device version OpenCL 1.2 CUDA, 4096MB, 3381MB available, 4045 GFLOPS peak)
26/02/2019 11:13:09 | SETI@home | Found app_info.xml; using anonymous platform


PS I have done this before, both my old Dells ran for the WOW event with pairs of 750ti's and never had this problem. I assume it is because they are different cards.
ID: 1982438 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14652
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1982439 - Posted: 27 Feb 2019, 9:46:35 UTC - in response to Message 1982431.  

So went into safe mode and removed the Nvidia drivers with DDU. Rebooted, and did another install of the 419.17 drivers. This failed before it even got to the options, tired it again same result.
OpenCL is a cross-platform language: you probably got the second OpenCL stack for your

AMD A8-3800 APU with Radeon(tm) HD Graphics

- I've gained OpenCL support for an NVidia card by allowing Windows 10 to download the Intel drivers for their equivalent iGPU.
ID: 1982439 · Report as offensive
Profile Bernie Vine
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 9954
Credit: 103,452,613
RAC: 328
United Kingdom
Message 1982467 - Posted: 27 Feb 2019, 15:16:53 UTC

So just to see I decided to try the 1060 in one of the old Dell XPS 420's

I had of course forgotten that they are both 32 bit and Nvidia support has stopped for 32 bit systems and driver 391.35 is the last one.

Well as you can probably guess the driver failed to install. So either it didn't like the 1060 or there is a problem with the card.

Removed the 1060 and put the 750ti back no problems Boinc saw it straight away.
ID: 1982467 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14652
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1982471 - Posted: 27 Feb 2019, 16:01:10 UTC - in response to Message 1982467.  

If you go to the 'manual select' NVidia download page https://www.nvidia.co.uk/Download/index.aspx, you can select for Windows XP drivers for the 900 series. But as soon as you select the GeForce 10 Series, the minimum operating system flips to Windows 7 64-bit.

So, 1060 won't work on XP. No reason to suspect any other conclusion or fault.
ID: 1982471 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1982476 - Posted: 27 Feb 2019, 16:47:01 UTC - in response to Message 1982431.  

Well seems like I had a worse problem, I checked the WU's that had been done just before the outage, to see all of them were being done on device 0 the 1060.


I have had this literally happen with both Nvidia and AMD drivers under Windows 10. The fix was "much older" drivers with a clean install.

Tom
A proud member of the OFA (Old Farts Association).
ID: 1982476 · Report as offensive
Profile Bernie Vine
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 9954
Credit: 103,452,613
RAC: 328
United Kingdom
Message 1983566 - Posted: 6 Mar 2019, 8:57:05 UTC
Last modified: 6 Mar 2019, 8:59:19 UTC

In an attempt to use my spare 1060, I am thinking of building a cheap and cheerful Linux box I have the case and a good PSU. I may use an existing HDD or get a small SSD

It seems the cheapest CPU is AMD, so as a cheap bundle how does this sound

AMD Bulldozer FX-6300 6 Core 3.5Ghz, ASUS M5A78L-M USB3 Motherboard & 8GB 1600Mhz DDR3 RAM - £165 - $217
ID: 1983566 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13739
Credit: 208,696,464
RAC: 304
Australia
Message 1983567 - Posted: 6 Mar 2019, 9:06:37 UTC - in response to Message 1983566.  
Last modified: 6 Mar 2019, 9:10:06 UTC

AMD Bulldozer FX-6300 6 Core 3.5Ghz, ASUS M5A78L-M USB3 Motherboard & 8GB 1600Mhz DDR3 RAM - £165 - $217

Very old system- DDR3 RAM, and Bulldozer CPU (huge power consumption, poor actual performance).
I'd suggest checking how much an original Ryzen 3 CPU (1200 or 1300X, or Ryzen 5 1400) & Motherboard with RAM would cost. Even if it's $50 more, it would $50 well spent IMHO.
For what you're intending, 4 cores (no hyperthreading) would be plenty (unless you intend to load it up with GPUs in the future).

As for storage, SSDs still cost more than HDDs, but they're worth every cent- even for the smaller capacity more basic ones.
Grant
Darwin NT
ID: 1983567 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34802
Credit: 261,360,520
RAC: 489
Australia
Message 1983568 - Posted: 6 Mar 2019, 9:15:49 UTC

Also remember that Bulldozer cores are made in modules containing 2 cores which share an FP unit between them so don't expect good performance when using more than 1 core on each module.

Cheers.
ID: 1983568 · Report as offensive
Profile Bernie Vine
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 9954
Credit: 103,452,613
RAC: 328
United Kingdom
Message 1983570 - Posted: 6 Mar 2019, 9:24:01 UTC

Thanks, yes that is what I suspected but not having kept up with AMD CPU's I was a bit in the dark,.

A similar setup with a Ryzen and no ram is £100 more and for me, really not worth it just for a crunching rig.

Trouble is INTEL CPU's start about £150 above that !!

I will have to do a bit more searching!!
ID: 1983570 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14652
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1983572 - Posted: 6 Mar 2019, 9:37:11 UTC - in response to Message 1983570.  

In the GTX 1660 thread, you mentioned buying from a firm in Bradford. I don't know if that was my favourite local store, but I had a look anyway:

AMD Ryzen 3 Motherboard Bundle for £219.99 assembled and available from stock.

CPU is a Ryzen 3 2200G which has the integrated GPU. We've had a bit of trouble with the drivers for those, but it should settle down in time. Worth including in your search, anyway.
ID: 1983572 · Report as offensive
Profile tullio
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 8797
Credit: 2,930,782
RAC: 1
Italy
Message 1983574 - Posted: 6 Mar 2019, 9:38:23 UTC - in response to Message 1983570.  

I have an A10-6700 with nominal 4 cores but the Windows 10 Task Manager says it has 2 cores and 4 logical processors. Anyway it runs only two task simultaneously, plus one GPU task. A Ryzen 5 1400 has four cores and runs 8 tasks simultaneously plus a GPU task.
Tullio
ID: 1983574 · Report as offensive
Profile Bernie Vine
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 9954
Credit: 103,452,613
RAC: 328
United Kingdom
Message 1983576 - Posted: 6 Mar 2019, 9:42:40 UTC - in response to Message 1983572.  

In the GTX 1660 thread, you mentioned buying from a firm in Bradford. I don't know if that was my favourite local store, but I had a look anyway:

AMD Ryzen 3 Motherboard Bundle for £219.99 assembled and available from stock.

CPU is a Ryzen 3 2200G which has the integrated GPU. We've had a bit of trouble with the drivers for those, but it should settle down in time. Worth including in your search, anyway.


Thanks Richard, yes it was CCL I got the 1660 from, very good service, extremely quick.

That is a better price and definitely worth thinking about
ID: 1983576 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13739
Credit: 208,696,464
RAC: 304
Australia
Message 1983579 - Posted: 6 Mar 2019, 9:45:57 UTC - in response to Message 1983576.  

AMD Ryzen 3 Motherboard Bundle for £219.99 assembled and available from stock.

That is a better price and definitely worth thinking about

While it's a bit more than you were considering, it will use a lot less power while performing a lot better than the Bulldozer system you were looking at. A bit more upfront, but well worth it over the long run.
Grant
Darwin NT
ID: 1983579 · Report as offensive
1 · 2 · 3 · Next

Message boards : Number crunching : A little odd?


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.