Author Topic: Is my cgminer hardware error high? (newbie)  (Read 21179 times)

Offline netvope

  • Newbie
  • *
  • Posts: 16
Is my cgminer hardware error high? (newbie)
« on: March 16, 2013, 08:56:08 AM »
I've just completed my very first hour of litecoin mining today (and this is my first post). Feeling great! Thanks you for all the useful information on this forum.

My GPU is a 7870 XT at 750/1500 (core/mem). I'm using cgminer-2.11.2-win32 with options "--scrypt --shaders 1536 --intensity 13" on Windows 8 x64. Here is the statistics after one hour:

Code: [Select]
cgminer version 2.11.2 - Started: [2013-03-16 03:29:00]
--------------------------------------------------------------------------------
 (5s):350.0K (avg):331.9Kh/s | Q:132  A:604  R:0  HW:0  E:458%  U:9.2/m
 ST: 2  SS: 0  DW: 260  NB: 23  LW: 546  GF: 0  RF: 0
 Connected to notroll.in diff 32 with stratum as user netvope.01
 Block: a802df9c893b3a07...  Diff:5.52M  Started: [04:34:40]  Best share: 81.3K
--------------------------------------------------------------------------------
 [P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
 GPU 0:  65.0C 1604RPM | 343.4K/332.3Kh/s | A:604 R:0 HW:52 U: 9.23/m I:13
--------------------------------------------------------------------------------

After reading cgminer's README file, I understand most of the fields. I'm a little bit concerned about "HW:52" in the last line. Given "A:604" and "R:0", my hardware error rate is about 8%, which sounds a bit too high for me. Is that normal? Is there anything I should do to lower "HW"?

I have not fully optimized the parameters. Running the GPU core underclocked is because the default clock (975) results in a lower hash rate. Any suggestions are welcome ;)

Additional question: I know that hash rate is not the only thing that affects actual useful work output. Which number best represents the useful output rate? The "U:9.2/m" field? (number of shares / minute)

Offline Necroscope

  • Jr. Member
  • **
  • Posts: 104
Re: Is my cgminer hardware error high? (newbie)
« Reply #1 on: March 16, 2013, 04:48:06 PM »
Your HW rate is too high. Ideally you should have 0 hardware errors. Hardware errors mean you are overworking or providing bad parameters for your card to work in and you may actually harm your card.

Your clocks seem a little weird. You may want to change those as well. Some good numbers were 920 clock 1360 mem.
http://sdrv.ms/1011LK4 - Most recent mining contract
http://sdrv.ms/ZN6CkA - Sample reporting sheet

Offline DiCE1904

  • Full Member
  • ***
  • Donator
  • ***
  • Posts: 444
  • legit
    • LTCDice
Re: Is my cgminer hardware error high? (newbie)
« Reply #2 on: March 16, 2013, 06:08:49 PM »
what happens when you take off shaders and use
Code: [Select]
--auto-fan --auto-gpu   ?

Offline shyliar

  • Full Member
  • ***
  • Posts: 385
Re: Is my cgminer hardware error high? (newbie)
« Reply #3 on: March 16, 2013, 06:20:46 PM »
Yes just leave out shaders....it's my understanding that a new version of cgminer is about to be released that takes care of those issues for the 7000 series cards and the intensity 13 limit. So download the next version when it comes out.
Please support the Litecoin Dev team here https://litecointalk.org/index.php/topic,3874.0.html

Offline netvope

  • Newbie
  • *
  • Posts: 16
Re: Is my cgminer hardware error high? (newbie)
« Reply #4 on: March 16, 2013, 08:38:41 PM »
Thank you for the advice.

With "--scrypt --shaders 1536 --intensity 13":
At 750/1500, the average hash rate is 345 kH/s
At 920/1360, the average hash rate is 305 kH/s
At 975/1500 (default boosted clock), 330 kH/s
All these speeds gives some hardware error.

With "--scrypt --intensity 13 --auto-fan --auto-gpu":
The card runs at the default boosted clock of 975/1500, and the average hash rate is 350 kH/s. Zero hardware error.

I left all other cgminer parameters default. Temp is below 70 deg C in all cases.

Given that removing "--shader 1536" resolves the hardware errors, I guess it chose a thread size that's too big for my card?

I'm definitely looking forward to the next version  :)

Offline DiCE1904

  • Full Member
  • ***
  • Donator
  • ***
  • Posts: 444
  • legit
    • LTCDice
Re: Is my cgminer hardware error high? (newbie)
« Reply #5 on: March 16, 2013, 09:16:36 PM »
Thank you for the advice.

With "--scrypt --shaders 1536 --intensity 13":
At 750/1500, the average hash rate is 345 kH/s
At 920/1360, the average hash rate is 305 kH/s
At 975/1500 (default boosted clock), 330 kH/s
All these speeds gives some hardware error.

With "--scrypt --intensity 13 --auto-fan --auto-gpu":
The card runs at the default boosted clock of 975/1500, and the average hash rate is 350 kH/s. Zero hardware error.

I left all other cgminer parameters default. Temp is below 70 deg C in all cases.

Given that removing "--shader 1536" resolves the hardware errors, I guess it chose a thread size that's too big for my card?

I'm definitely looking forward to the next version  :)

great to hear its fixed! :)

Offline LinuxNoob

  • Full Member
  • ***
  • Posts: 289
Re: Is my cgminer hardware error high? (newbie)
« Reply #6 on: March 16, 2013, 09:19:00 PM »
{
"pools" : [
   {
      "url" : "http://mining-foreman.no-ip.org:10341",
      "user" : "xxxxxx",
      "pass" : "xxxxxx"
   }
]
,
"intensity" : "18",
"vectors" : "1",
"worksize" : "256",
"kernel" : "scrypt",
"lookup-gap" : "0",
"thread-concurrency" : "3336",
"shaders" : "0",
"gpu-engine" : "960",
"gpu-fan" : "10-90",
"gpu-memclock" : "1250",
"gpu-memdiff" : "0",
"gpu-powertune" : "0",
"gpu-vddc" : "0.000",
"temp-cutoff" : "95",
"temp-overheat" : "85",
"temp-target" : "75",
"api-port" : "4028",
"auto-fan" : true,
"auto-gpu" : true,
"expiry" : "120",
"gpu-dyninterval" : "7",
"gpu-platform" : "0",
"gpu-threads" : "1",
"log" : "5",
"no-pool-disable" : true,
"queue" : "1",
"scan-time" : "120",
"scrypt" : true,
"temp-hysteresis" : "3",
"shares" : "0",
"kernel-path" : "/usr/local/bin"
}

Thats whats written in my cgminer.conf

XFX R6770 1GB using CGminer 2.10.5 version

Getting average 225k hashes with 0 HW errors. GPU and FAN both on auto

Suggestions would be appreciated on various methods to improve.
« Last Edit: March 16, 2013, 09:21:08 PM by LinuxNoob »

Offline shyliar

  • Full Member
  • ***
  • Posts: 385
Re: Is my cgminer hardware error high? (newbie)
« Reply #7 on: March 16, 2013, 09:20:40 PM »
Thank you for the advice.

With "--scrypt --shaders 1536 --intensity 13":
At 750/1500, the average hash rate is 345 kH/s
At 920/1360, the average hash rate is 305 kH/s
At 975/1500 (default boosted clock), 330 kH/s
All these speeds gives some hardware error.

With "--scrypt --intensity 13 --auto-fan --auto-gpu":
The card runs at the default boosted clock of 975/1500, and the average hash rate is 350 kH/s. Zero hardware error.

I left all other cgminer parameters default. Temp is below 70 deg C in all cases.

Given that removing "--shader 1536" resolves the hardware errors, I guess it chose a thread size that's too big for my card?

I'm definitely looking forward to the next version  :)

No the current version of cgminer has some issues with the 7000 series cards shaders and thread-concurrency. The new version coming out is suppose to resolve that issue.

Read post from ckolivas here:

https://bitcointalk.org/index.php?PHPSESSID=cifmlrni11dhdc7o14nd7kcls2&topic=117221.msg1627195#msg1627195

And what he's getting on his 7970 here:

https://bitcointalk.org/index.php?topic=28402.msg1627553#msg1627553

I downloaded from the GIT but still have to play with it. Initially I had some problems so looking forward to the finally release.
« Last Edit: March 16, 2013, 09:32:32 PM by shyliar »
Please support the Litecoin Dev team here https://litecointalk.org/index.php/topic,3874.0.html

Offline netvope

  • Newbie
  • *
  • Posts: 16
Re: Is my cgminer hardware error high? (newbie)
« Reply #8 on: March 16, 2013, 10:17:50 PM »
After extensive experiments, the best I can get with the current version (2.11.2) is about 385 kH/s, with zero hardware errors. (My original untuned setting gave 345 kH/s and 8% hardware errors.)

Code: [Select]
Parameters: --scrypt --intensity 13 --thread-concurrency 8192 --gpu-threads 2 --worksize 256 --vectors 1
Core speed: 840 MHz (further increase will drop the hash rate)
Memory speed: 1500 MHz

At intensity 13, optimum thread concurrency is 8192. At 6144 or below, I get hardware errors. At 7680, it works without hardware errors but is slower than 8192. Does anyone know why a lower concurrency would result in hardware error?

Maximum gpu threads that can fit into 2 GB VRAM is 2. 1 gpu thread results in a much lower hash rate, presumably because it cannot fully exercise the gpu.

Default worksize is 64 and that limits the peak performing core clock to about 780 MHz (355 kH/s). Increasing the work size to 128 or 256 allows further increase in core clock before performance drops. The optimum clock is now about 840 MHz (385 kH/s). I also tried a worksize of 192 and that worsens the hash rate per clock.

Next is to try the new build  ;D

« Last Edit: March 16, 2013, 10:34:35 PM by netvope »

Offline netvope

  • Newbie
  • *
  • Posts: 16
Re: Is my cgminer hardware error high? (newbie)
« Reply #9 on: March 16, 2013, 10:52:30 PM »
I've just tried the new build in this post (thanks shyliar!)

First I tried raising the intensity and thread concurrency. However, it appears that the 512 MB memory limit is still there. If I try "--intensity 14 --thread-concurrency 10240", I get this error:
Code: [Select]
Maximum buffer memory device 0 supports says 536870912
Your scrypt settings come to 671088640

So the maximum possible is still 8192.

Then I tried the setting optimized for the old build, but that resulted in very low hash rate and a unresponsive desktop, resembling the situation when the number of gpu threads is too high.

Lowering gpu threads to 1 works, but the hash rate is below 300 kH/s.

I experimented with a couple other parameters but I cannot get anything near the hash rate of the old build.

Any idea what parameters to use with the new build? I've read all the replies in that thread (after the new build post) but nothing seems to be useful. For the time being I'll continue to use the old one.
« Last Edit: March 17, 2013, 12:49:57 AM by netvope »

Offline shyliar

  • Full Member
  • ***
  • Posts: 385
Re: Is my cgminer hardware error high? (newbie)
« Reply #10 on: March 17, 2013, 05:20:34 AM »
I didn't have time to experiment with the new .exe today....and well back to work tomorrow after a week off  >:(

When I do i'll post my results and hopefully increased kh/s. The brief few minutes I experimented with it I got hardware errors.....Maybe still a work in progrees; but, it is being worked on.
« Last Edit: March 17, 2013, 05:22:32 AM by shyliar »
Please support the Litecoin Dev team here https://litecointalk.org/index.php/topic,3874.0.html

Offline MKEGuy

  • Jr. Member
  • **
  • Posts: 115
Re: Is my cgminer hardware error high? (newbie)
« Reply #11 on: March 17, 2013, 05:55:31 AM »
If you go to the CGMiner thread on bitcointalk.org you can download a windows binary of CGMiner.exe that has an update to allow higher thread concurrency values.

Google or search bitcointalk.org
If I helped you, please consider donating to the following address.  LNarfMNLyymkS9WYQFLcmr46AeMYNP8Qj2  ATTENTION LITECOIN MINERS!  Do not forget to donate to CKlovias if you are using CGMiner!  The man never receives any grattitude for the money he has enabled us to make!!!!

Offline DiCE1904

  • Full Member
  • ***
  • Donator
  • ***
  • Posts: 444
  • legit
    • LTCDice
Re: Is my cgminer hardware error high? (newbie)
« Reply #12 on: March 17, 2013, 06:30:33 AM »
I didn't have time to experiment with the new .exe today....and well back to work tomorrow after a week off  >:(

When I do i'll post my results and hopefully increased kh/s. The brief few minutes I experimented with it I got hardware errors.....Maybe still a work in progrees; but, it is being worked on.

Tried it just now, mining LTC with 2.11.3 now gives me HW errors. Only with LTC, no HW errors mining BTC  >:(

Offline netvope

  • Newbie
  • *
  • Posts: 16
Re: Is my cgminer hardware error high? (newbie)
« Reply #13 on: March 18, 2013, 07:22:56 AM »
Tried 2.11.3. With the old parameters it gives about the same hash rate as 2.11.2.

With a higher thread concurrency (>8192) I still get this error as I got in 2.11.2:

Code: [Select]
[2013-03-18 03:06:09] Maximum buffer memory device 0 supports says 536870912                   
 [2013-03-18 03:06:09] Your scrypt settings come to 671088640                   
 [2013-03-18 03:06:09] Creating scrypt buffer sized 671088640                   
 [2013-03-18 03:06:09] Error -61: clCreateBuffer (padbuffer8), decrease CT or increase LG               

Increasing LG (lookup gap) will allow a higher thread concurrency, but the hash rate will be much lower. Reducing the number of GPU threads from 2 to 1 won't solve the error.

So while I highly appreciate their effort in bringing the new version, unfortunately I'm unable to benefit from it (or I don't know how to set it)

Here is a very useful guide for 2.11.3 written by the author of cgminer. Sadly the automatic detection his suggested would give me a thread concurrency of 4032 only...

Edit: Resolved. See http://forum.litecoin.net/index.php/topic,1420.msg8740.html#msg8740
« Last Edit: March 19, 2013, 02:43:07 AM by netvope »