HD7950 Crossifre very low Intensity / Massive HW errors above -I 13

[mgoldfinger]

Need some help here. I was previously running a rig with HD 6950 at a nicely tweaked level of about 500K/hs. I now build a new rig with the following specs:
-   2 x XFX HD7950 Double Dissipation (FX-795A-TDFC)
-   MSI Z87-G45 Gaming,s1150  
-   Intel Core i3-4130,s1150
-   1 x TNAV6R00 Corsair RM1000 (Gold standard)
-   ICIDU Value DDR3 1333-2GB (2x4GB)
-   Windows 7 Home Premium 64Bit

All new hardware.

Using CGMiner 3.7.2. for scrypt mining

Did a clean install of Windows 7. After I installed all the latest AMD drivers, I input setx GPU_MAX_ALLOC_PERCENT 100  and I start cgminer
I get massive HV issues when trying to increase the intensity above 12!

currently running at stock with 900/1250 settings for GPU/RAM. Voltage is untouched. Both GPUs are found by CGminer and both have this issue.

I have tried:
1   CGMiner settings,  adjusting –G, -W, --Thead Concurrency (8192 is the max I can run without adjusting the lookup-gap) --Lookup-gap, blank settings (I then get intensity 8  )
2   Various AMD/catalyst drivers
3   Varios CGMiner versions < 3.7.2
4   Run a single card, instead of crossfire.
5   Disabling crossfire, run both cards
6   Disabled performance tuning in the Catalyst software
7   Swapped the system RAM
8      mine a different pools / different Scrypt cons

I have found posts with similar problems, but none of the solutions seem to work for me. I also have the feeling that CGminer settings is not the issue here. More like a conflict somewhere happening in the rig?
Any ideas? did you also have this issue, and how did you solve it?

[motd2k]

I have a very similar setup (with a 3rd 7950) and have them tweaked to produce around 650KH/s with this config:

Code:

"auto-fan" : true,
"auto-gpu" : true,
"temp-cutoff" : "90",
"temp-overheat" : "80",
"temp-target" : "70",
"gpu-fan" : "0-100",
"intensity" : "20",
"vectors" : "1",
"worksize" : "256",
"kernel" : "scrypt",
"lookup-gap" : "2",
"thread-concurrency" : "22400",
"shaders" : "0",
"api-port" : "4028",
"expiry" : "120",
"gpu-dyninterval" : "7",
"gpu-platform" : "0",
"gpu-threads" : "1",
"gpu-engine" : "1020,1020,1100",
"gpu-memclock" : "1500",
"log" : "5",
"no-pool-disable" : true,
"queue" : "1",
"scan-time" : "60",
"scrypt" : true,
"shares" : "0",
"kernel-path" : "/usr/local/bin"

These hash rates were fairly consistent between Windows, Debian, and BAMT installations.

[mgoldfinger]

I have a very similar setup (with a 3rd 7950) and have them tweaked to produce around 650KH/s with this config:

Code:

"auto-fan" : true,
"auto-gpu" : true,
"temp-cutoff" : "90",
"temp-overheat" : "80",
"temp-target" : "70",
"gpu-fan" : "0-100",
"intensity" : "20",
"vectors" : "1",
"worksize" : "256",
"kernel" : "scrypt",
"lookup-gap" : "2",
"thread-concurrency" : "22400",
"shaders" : "0",
"api-port" : "4028",
"expiry" : "120",
"gpu-dyninterval" : "7",
"gpu-platform" : "0",
"gpu-threads" : "1",
"gpu-engine" : "1020,1020,1100",
"gpu-memclock" : "1500",
"log" : "5",
"no-pool-disable" : true,
"queue" : "1",
"scan-time" : "60",
"scrypt" : true,
"shares" : "0",
"kernel-path" : "/usr/local/bin"

These hash rates were fairly consistent between Windows, Debian, and BAMT installations.

thanks, will try. Did you initially have similair problems like I am having?

[mgoldfinger]

I have a very similar setup (with a 3rd 7950) and have them tweaked to produce around 650KH/s with this config:

Code:

"auto-fan" : true,
"auto-gpu" : true,
"temp-cutoff" : "90",
"temp-overheat" : "80",
"temp-target" : "70",
"gpu-fan" : "0-100",
"intensity" : "20",
"vectors" : "1",
"worksize" : "256",
"kernel" : "scrypt",
"lookup-gap" : "2",
"thread-concurrency" : "22400",
"shaders" : "0",
"api-port" : "4028",
"expiry" : "120",
"gpu-dyninterval" : "7",
"gpu-platform" : "0",
"gpu-threads" : "1",
"gpu-engine" : "1020,1020,1100",
"gpu-memclock" : "1500",
"log" : "5",
"no-pool-disable" : true,
"queue" : "1",
"scan-time" : "60",
"scrypt" : true,
"shares" : "0",
"kernel-path" : "/usr/local/bin"

These hash rates were fairly consistent between Windows, Debian, and BAMT installations.

not working i am afraid. still can't get intensity above 13.....

[mgoldfinger]

I think i am zooming in on the problem, but i have no idea how to solve it / what is actually causing it?

I noticed stability is very much dependent on thread concurrency , in my setup

I can not get thread concurrency above 8192 (this should be way to low for a 7950??), without changing the lookup gap. @gap 2 i can run stable on intensity 13 @ 8192 concurrency. lowering concurrency results in HW errors @ I 13, gap 2

If i change lookup gap to 3, i can blast intensity to 19/20. However, performance is then gone.... +/-650 khz combined (remember i run a crosffire setup).

really getting stuck here......

[mgoldfinger]

FYI: solved.

Flashed the bios on the card using VBE7, which got me down to 1.080V and removed boost

Reinstalled windows and can now reach TC>40000 (which is pointless btw)

Tweaked the clocks and can now hash at +/-660 per card with GPU/RAM 1120/1500, stable and all.

btw.

http://www.holynerdvana.com/2013/12/finding-optimal-clock-speed-for-cgminer.html

great tool (credits to the developer)!

[sc2btc]

mgoldfinger, would you be able to disclose what BIOS you've flashed too and/or where you obtained it?

I found this thread after Googling a bit and I'm having the exact same problem: fresh build, fresh Win7 64 install, (single) XFX HD 7950 DD has ever increasing HW failures at difficulty 13+ (it also rises in temperature pretty quickly after as well). Everything is stock. I've yet to toy with TC values.

GPU-Z lists the current BIOS as 015.024.000.001.000000

Glad you got yours working!

[andycee]

TBH, I found it was way easier to let CG Miner decide its own Thread Concurrency on multiple card setups. I leave it out of the .conf file and get 650Kh/s every time, with minimal tweaking on my Gigabyte 7950's.

[6Fallen]

you need a new rig thought