Pages:
Author

Topic: 7970 mining thread (Read 20004 times)

member
Activity: 74
Merit: 10
December 28, 2012, 07:48:26 PM
#60
Any luck with that compiler beating since April?
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
April 03, 2012, 02:47:46 AM
#59
2. I did some research on the forums, and a random internet-stranger said that the 7970 is already fairly close to the maximum possible hashrate. Are any further improvements likely?

Yes. Possibly, I'm still trying to beat the compiler here.
full member
Activity: 812
Merit: 108
April 01, 2012, 12:54:11 AM
#58
Looks like I will be getting 3 or 4 7970s  Can I just assume ASUS has the best card?

Sapphire makes the best AMD cards.  

Most of the 7970s out right now are reference boards and are pretty much the same
Sapphire is just AMD's OEM branding and is mostly okay -- not great, not bad
XFX has by far the best warranty on their cards (lifetime)
Asus, Gigabyte and MSI all make good cards with decent aftermarket cooling options.  If you want a card that is going to live a long time, get something with solid cooling on the VRMs, RAM and GPU
Personally I would buy this one if you're on risers: http://hexus.net/tech/reviews/graphics/33697-gigabyte-radeon-hd-7970-oc-gv-r797oc-3gd/?page=14

1) Sapphire makes the most AMD cards in the entire world
2) Sapphire has made AMD since the dawn of AMD time
3) because of 2) Sapphire has an R&D role with AMD
4) Sapphire makes all of the FirePro cards
5) regardless of branding Sapphire makes all of the dual GPU cards

Sapphire makes the best AMD cards.  XFX is nothing more than a tier 3 card maker.  That is why their warranty has to be so good.  Their cards suck.  

EDIT:Gigabyte, Asus, and MSI all make really good cards too.  I would consider them top tier makers along with Sapphire.

XFX killed their lifetime warranty except on the 600 dollar version of their cards, the double d. Since it doesn't add a second lifetime for resale, I think I'll stick with the 2 year warranty on the 529 dollar cards, or the three year on gigabyte/asus. Just picked up a sapphire for 475 today, and got a secondhand asus for 490 last week. Slowly replacing my older 5000 series cards with mining proceeds and ebay sales of the 5000 cards, going to be a profitless couple of months, but it's worth it imo to get into cards that have full warranty.
donator
Activity: 1218
Merit: 1079
Gerald Davis
February 22, 2012, 09:49:50 AM
#57
Yes - the 5870 cores on a 5970 have to be downclocked and voltage reduced to comfortably stay with specs for PCI-e.

5970 = 1.050
5870 = 1.163



Good news is mining uses a lot less power (due to underclocked and underutilized memory) so you can restore the voltage back to 5870 levels.

The bad news is power consumption increases by the square of the voltage.  So (1.163/1.050)^2 = ~22% increase in GPU core power.  22% increase in GPU core power = 22% increase in thermal energy (heat). Sad
That is at the same clock.  If you increased the clock 15% to take advantage of that increased voltage you are looking at 1.22 * 1.15 = 41% more thermal energy than an overclocked 5970.  This is more than the cooling system can handle even at 100% fan.

Of course is you are doing something exotic:
* watercooling
* oil immersion cooling
* piping 20C air directly into air intake
Smiley
legendary
Activity: 1134
Merit: 1005
February 21, 2012, 05:59:09 PM
#56
I have no problem with Sapphire RMA's. Had to do RMA for 2 cards (not mining related)
legendary
Activity: 1666
Merit: 1000
February 12, 2012, 08:45:13 PM
#55
Yes - the 5870 cores on a 5970 have to be downclocked and voltage reduced to comfortably stay with specs for PCI-e.

5970 = 1.050
5870 = 1.163

sr. member
Activity: 378
Merit: 250
Why is it so damn hot in here?
February 12, 2012, 08:42:50 PM
#54
jim, what are your setting for your 5970's?  at 825/300 I'm getting 340/gpu

At 825/150 @ 1.05v I'm getting ~375 mhash/s from each core.

Haven't tried dropping the voltage any to see if it'd remain stable at 825/150 though.

I can bump it up to 850/150 @ 1.05v and get ~390 mhash/s but then I get stability issues where the screen will go black and I get a notification saying the driver stopped responding and it recovered but cgminer stops working until I restart it.  This only happens when I'm actively browsing with the computer at this setting though.  If I just let it run by itself overnight it's stable.

BTW - is 1.05v the default voltage for a 5970?  Tried Googling it but didn't get any concrete answers.

1.05 is the default.

That's pretty low since all my 5870s are 1.163 at stock voltage.  Downclocked to 1.100.
legendary
Activity: 1876
Merit: 1000
February 12, 2012, 08:09:51 PM
#53
jim, what are your setting for your 5970's?  at 825/300 I'm getting 340/gpu

At 825/150 @ 1.05v I'm getting ~375 mhash/s from each core.

Haven't tried dropping the voltage any to see if it'd remain stable at 825/150 though.

I can bump it up to 850/150 @ 1.05v and get ~390 mhash/s but then I get stability issues where the screen will go black and I get a notification saying the driver stopped responding and it recovered but cgminer stops working until I restart it.  This only happens when I'm actively browsing with the computer at this setting though.  If I just let it run by itself overnight it's stable.

BTW - is 1.05v the default voltage for a 5970?  Tried Googling it but didn't get any concrete answers.

1.05 is the default.
full member
Activity: 944
Merit: 101
PredX - AI-Powered Prediction Market
February 12, 2012, 07:48:34 PM
#52
jim, what are your setting for your 5970's?  at 825/300 I'm getting 340/gpu

At 825/150 @ 1.05v I'm getting ~375 mhash/s from each core.

Haven't tried dropping the voltage any to see if it'd remain stable at 825/150 though.

I can bump it up to 850/150 @ 1.05v and get ~390 mhash/s but then I get stability issues where the screen will go black and I get a notification saying the driver stopped responding and it recovered but cgminer stops working until I restart it.  This only happens when I'm actively browsing with the computer at this setting though.  If I just let it run by itself overnight it's stable.

BTW - is 1.05v the default voltage for a 5970?  Tried Googling it but didn't get any concrete answers.
donator
Activity: 1218
Merit: 1079
Gerald Davis
February 10, 2012, 04:49:26 PM
#51
posting initial results of test:

2 rigs  each has:
mobo:  crosshair V
cpu:    phenomx4
psu:   seasonic 1250
mem:  4g 1600
all defaults

4x5970
filesys:    4g corsair mini  linuxcoin cgminer
voltage:  1.05
clocks:   820/300
killwatt:  1170
Mh:        2970

5x7970
filesys:    wd green 500g  win7  diablo
voltage:  1.056
clocks:   1049/685
killwatt:  1160
Mh:        3120

I hope that is watts or you forgot a decimal. Smiley
sr. member
Activity: 369
Merit: 250
February 10, 2012, 04:44:24 PM
#50
no reason to wait..  a 5970 is a 750Mh card.  Smiley


800Mh Stable@800Mhz overclock


I have over 20 of these...  at 820/300 they get a solid  373 per core...  Please show us pics of 800Mh getting 400......stable

jim, what are your setting for your 5970's?  at 825/300 I'm getting 340/gpu
member
Activity: 62
Merit: 10
100X100111XX10
February 10, 2012, 05:24:04 AM
#49
I guess this is as good a place as any to ask a few general GPU questions, especially since they relate to my plans to buy a 7970 for my main rig that will spend a fair bit of time shuffling bits. Despite being what might very well qualify as a gamer, I've never really been on the bleeding edge of hardware purchases. As such, I know embarrassingly little about GPUs.

1. I read somewhere that there is a minimum time after the release of a new card during which AMD only allows third party manufacturers to sell reference cards. If true, how long before that changes? After my purchase of a poorly labeled 6950 that had all manner of odd quirks and was even missing the dual BIOS functionality, I've decided that if I buy a 7970, it'll be a reference model. Once burned, hella paranoid.

2. I did some research on the forums, and a random internet-stranger said that the 7970 is already fairly close to the maximum possible hashrate. Are any further improvements likely?

3. Would you recommend any particular brand? If I'm buying a reference card, it would seem like price and warranty are the only things that matter.

4. Are all reference cards built with the same fans and GPU "case"?
legendary
Activity: 1428
Merit: 1000
https://www.bitworks.io
February 08, 2012, 11:27:41 PM
#48

Sapphire makes the best AMD cards.  XFX is nothing more than a tier 3 card maker.  That is why their warranty has to be so good.  Their cards suck.  

EDIT:Gigabyte, Asus, and MSI all make really good cards too.  I would consider them top tier makers along with Sapphire.

Sadly between XFX, Sapphire, Gigabyte and MSI cards my only failures have been Sapphire cards. I used to run mostly Sapphire for the reasons mentioned above but have since jumped to XFX and have had much better luck. Might have just been my bad luck but after 3 I decided it was more than that..
member
Activity: 106
Merit: 10
February 08, 2012, 11:18:03 PM
#47
"AGGRESSION=7"

 Best I can tell, pool and local proxy is reporting accurately.
legendary
Activity: 1666
Merit: 1000
February 08, 2012, 08:35:29 PM
#46
Multicore isn't the issue - it is the additional (unneeded) power draw from a pegged CPU core.
member
Activity: 106
Merit: 10
February 08, 2012, 07:14:46 PM
#45
What is that localproxy and how do I get one ?

https://github.com/c00w/bitHopper
 
CPU issues with 11.11 are irrelevant to me as my boxes are all multicore dedicated to mining.
hero member
Activity: 896
Merit: 1000
Seal Cub Clubbing Club
February 08, 2012, 06:55:21 PM
#44
Using Catalyst 11.11 on a headless Win7 box (Absolutely nothing connected to any of the video adapters)

 Phoenix 1.7.3.

 cmdline: phoenix -u -k phatk BFI_INT FASTLOOP=false AGGRESSION=7 DEVICE=n WORKSIZE=128 VECTORS

Thanks! I'll have to upgrade my Phoenix later.  I think I'm still on 1.50.  Are you using whatever SDK came with 11.11?  I remember there being some issues with CPU load with that driver IIRC.  Sorry for veering off-topic.
-ck
legendary
Activity: 4088
Merit: 1631
Ruu \o/
February 08, 2012, 06:12:01 PM
#43
Since this is the 7970 mining thread, I'll paraphrase what I put on the cgminer thread:

Okay for those who want to get 7970 working well with cgminer on linux, I've been up most of the night since I got the card (lol) trying to get it working and then fiddled with settings to find a sweet spot.

Firstly: There is no actual "stable release" driver for this GPU yet. The driver you're directed to is a special release not really version numbered (like 12.1 etc). The 12.1 driver does NOT work with 7970 so when I put the 7970 in the machine with 6970s it didn't even show up. Then when I installed the GCN ati driver amd-driver-installer-8.921-x86.x86_64.run the 6970s wouldn't show up! Goddamn amd. I modified the xorg to add the extra devices, and xorg would start up but cgminer would just crash opencl when starting. After much searching around sharky suggested I change the card order, so I put the 7970 into slot 1 and the 6970s into slots 2-4. Again  the ati driver didn't show up the 6x devices with aticonfig --lsa, and wouldn't configure the xorg for more than just the 7970. Then I checked the GPU positions

Code:
lspci  | grep VGA
01:00.0 VGA compatible controller: ATI Technologies Inc Device 6798
02:00.0 VGA compatible controller: ATI Technologies Inc Device 6718
07:00.0 VGA compatible controller: ATI Technologies Inc Device 6718
08:00.0 VGA compatible controller: ATI Technologies Inc Device 6718

I used the PCI bus position entries and edited the xorg.conf file into the following generic version:

Code:
Section "Monitor"
        Identifier   "aticonfig-Monitor[0]-0"
        Option      "VendorName" "ATI Proprietary Driver"
        Option      "ModelName" "Generic Autodetecting Monitor"
        Option      "DPMS" "true"
EndSection

Section "Monitor"
        Identifier   "aticonfig-Monitor[1]-0"
        Option      "VendorName" "ATI Proprietary Driver"
        Option      "ModelName" "Generic Autodetecting Monitor"
        Option      "DPMS" "true"
EndSection

Section "Monitor"
        Identifier   "aticonfig-Monitor[2]-0"
        Option      "VendorName" "ATI Proprietary Driver"
        Option      "ModelName" "Generic Autodetecting Monitor"
        Option      "DPMS" "true"
EndSection

Section "Monitor"
        Identifier   "aticonfig-Monitor[3]-0"
        Option      "VendorName" "ATI Proprietary Driver"
        Option      "ModelName" "Generic Autodetecting Monitor"
        Option      "DPMS" "true"
EndSection

Section "Device"
        Identifier  "aticonfig-Device[0]-0"
        Driver      "fglrx"
        BusID       "PCI:1:0:0"
EndSection

Section "Device"
        Identifier  "aticonfig-Device[1]-0"
        Driver      "fglrx"
        BusID       "PCI:2:0:0"
EndSection

Section "Device"
        Identifier  "aticonfig-Device[2]-0"
        Driver      "fglrx"
        BusID       "PCI:7:0:0"
EndSection

Section "Device"
        Identifier  "aticonfig-Device[3]-0"
        Driver      "fglrx"
        BusID       "PCI:8:0:0"
EndSection

Section "Screen"
        Identifier "aticonfig-Screen[0]-0"
        Device     "aticonfig-Device[0]-0"
        Monitor    "aticonfig-Monitor[0]-0"
        DefaultDepth     24
        SubSection "Display"
                Viewport   0 0
                Depth     24
        EndSubSection
EndSection

Section "Screen"
        Identifier "aticonfig-Screen[1]-0"
        Device     "aticonfig-Device[1]-0"
        Monitor    "aticonfig-Monitor[1]-0"
        DefaultDepth     24
        SubSection "Display"
                Viewport   0 0
                Depth     24
        EndSubSection
EndSection

Section "Screen"
        Identifier "aticonfig-Screen[2]-0"
        Device     "aticonfig-Device[2]-0"
        Monitor    "aticonfig-Monitor[2]-0"
        DefaultDepth     24
        SubSection "Display"
                Viewport   0 0
                Depth     24
        EndSubSection
EndSection

Section "Screen"
        Identifier "aticonfig-Screen[3]-0"
        Device     "aticonfig-Device[3]-0"
        Monitor    "aticonfig-Monitor[3]-0"
        DefaultDepth     24
        SubSection "Display"
                Viewport   0 0
                Depth     24
        EndSubSection
EndSection

This would still only show up one device with aticonfig --lsa, but cgminer could detect them all.

Code:
cgminer -n
[2012-02-09 09:49:19] CL Platform 0 vendor: Advanced Micro Devices, Inc.
[2012-02-09 09:49:19] CL Platform 0 name: AMD Accelerated Parallel Processing
[2012-02-09 09:49:19] CL Platform 0 version: OpenCL 1.1 AMD-APP (844.4)
[2012-02-09 09:49:19] Platform 0 devices: 4
[2012-02-09 09:49:19] GPU 0 AMD Radeon HD 7900 Series  hardware monitoring enabled
[2012-02-09 09:49:19] GPU 1 AMD Radeon HD 6900 Series hardware monitoring enabled
[2012-02-09 09:49:19] GPU 2 AMD Radeon HD 6900 Series hardware monitoring enabled
[2012-02-09 09:49:19] GPU 3 AMD Radeon HD 6900 Series hardware monitoring enabled
[2012-02-09 09:49:19] 4 GPU devices max detected

After much playing around, I found the particular card I had would run stable at clockspeeds of 1200/1050. The 7970 has a mandatory no-more-than 150 difference between the GPU engine clockspeed and the GPU memory clockspeed much like the 6970 had a 125 difference. Note that with windows overclocking tools they can get around this using other backdoors to the devices which I can't do. Powertune of +5% provided the most boost to performance as I've seen with the 6970s and going over this made no difference.

On first successful running, I noticed the dreaded CPU usage bug was back! Luckily I heard this was not universal and

Code:
export GPU_USE_SYNC_OBJECTS=1
fixed it.

Then I played around with various combinations of gpu threads and intensities and found that there is an inflexion point in intensity where the CPU usage jumps up from the usually very low usage. With 1 GPU thread, an intensity up to 13 does not increase CPU usage. With 2 threads an intensity of 11 does not. More threads did not help hashrate.

Now bear in mind that this is an unmodified cgminer 2.2.3 so no new fancy kernels, but here is the final performance with intensity 11, engine 1200, memory 1050, powertune 5. See if you can spot the 790 amongst these Wink :

Code:
 GPU 0:  71.0C 4125RPM | 694.5/694.8Mh/s | A:355 R:1 HW:0 U:10.18/m I:11
 GPU 1:  72.5C 5107RPM | 426.4/427.4Mh/s | A:200 R:0 HW:0 U: 5.73/m I: 9
 GPU 2:  73.0C 4341RPM | 425.1/426.9Mh/s | A:214 R:1 HW:0 U: 6.14/m I: 9
 GPU 3:  72.5C 3892RPM | 434.6/436.1Mh/s | A:227 R:1 HW:0 U: 6.51/m I: 9

So that's with the now ancient poclbm kernel in cgminer that was modified recently just to work with 7970 aka GCN but has no real performance enhancements. Now I need to fiddle with kernels and wait to see what diablo comes up with as well.

Note that GCN ONLY works with the 2.6 SDK which has lousy performance on other cards, so I kept my old .bin files for the 6970s so that cgminer would just use those instead of regenerating them after I installed the new driver and sdk.
hero member
Activity: 518
Merit: 500
February 08, 2012, 05:30:44 PM
#42
Wow, Jizz that's nuts.  Would you mind sharing your Phoenix command line options?  What driver+SDK version are you using?

 Using Catalyst 11.11 on a headless Win7 box (Absolutely nothing connected to any of the video adapters)

 Phoenix 1.7.3.

 cmdline: phoenix -u -k phatk BFI_INT FASTLOOP=false AGGRESSION=7 DEVICE=n WORKSIZE=128 VECTORS

What is that localproxy and how do I get one ?

Thanks !
member
Activity: 106
Merit: 10
February 08, 2012, 03:55:07 PM
#41
Wow, Jizz that's nuts.  Would you mind sharing your Phoenix command line options?  What driver+SDK version are you using?

 Using Catalyst 11.11 on a headless Win7 box (Absolutely nothing connected to any of the video adapters)

 Phoenix 1.7.3.

 cmdline: phoenix -u -k phatk BFI_INT FASTLOOP=false AGGRESSION=7 DEVICE=n WORKSIZE=128 VECTORS



Pages:
Jump to: