Bitcoin Forum

I just added a 7970 to my computer.  I have been trying different things, and I just cant get it going good.  This is a reference card.

Right now for testing, it is in the main slot, it is the only card installed, I have tried updating to GHZ bios.  That did increase stock clocks, but made no difference on the speed.  I have to manually set it to 925/1375 to get 550 again.  On my 7950 I can get 679khash overclocked, but this 7970 has to stay at 925/1375 to get 550khash and then it will not do any better than that.

I tried cgminer and reaper.  Suggestions? What other info can I provide that will help?

see sig thread, read down to 7970 settings poriton

Thanks for the link.  I was able to get it up to 575 khash, but still not anywhere near what it should be.  My 7950 is not finicky at all like this card.  Any other ideas?  I already put it up for sale a few placed, but maybe I can still get it to cooperate.

Just to let you know, my experience with Diamond 7970: Nothing gets me above 520kh/s.... I have tried almost every possible setting. The 7870s I just got are the same way. Both off brands.

Well, that is what this card is, a Diamond 7970.  I really don't know why this is happening, but it is.  Hopefully someone will buy it soon.  I'm sure it would still be a killer gaming card.

Which drivers are you using? On Catalyst 13.2 and 13.3 beta, I can pull 550MHash/s+ per card under BFGMiner. Also, make sure your power supply is sufficient and in good working order, and install MSI Afterburner to check temperatures and that the 7970 is being utilised properly. (When mining, GPU usage should be pegged at ~99%. Significantly lower values suggest that the GPU is being underused for some reason, and square waves suggest VRM throttling.)

GPU usage is right up there at the top.  Temps are fine.  I am on latest stable drivers, but I did try 12.8 as well.  I flashed to the GHZ BIOS, but am back on original.  I am now up to 590 khash, and my clocks are 965/1575.  At these clocks I would expect much more at -I 20.  This card still isn't as fast as the 7950.  I really think there is just something about this card.

increase thread concurrency? IDK I just started litecoin mining on my 7970s and cgminer reports 500 kh/s each altho the pool reports them running at 625 kh/s each. I still need to tweak my ram speed.

Try 875/1375.  I found that yields ~608 Kh/s.  Make sure power limit is @ 20%.

If you're using CGMiner, my settings are:

FYI to get close to or over 700 kh/s, you are going to be looking at RAM clocks in the 1700 to 1800 + territory.

Keep an eye on temps.  Sometimes a slight overvolt to the memory can help with stability.  I'm running 7970s with Arctic Accelero coolers for this very reason.

Thanks.  That did appear to show just over 600khash, but wasn't submitting any shares.  Weird.


At current clocks it should be doing over 600.  I expect it to far surpass my 7950, but it cant.  Card is borked somehow.  Those Accelero coolers are nice though.

"On my 7950 I can get 679khash overclocked"

But how!?  :o

I just chocked it up to GPU binning. Diamond must buy the bottom of the barrel cores.

Core 1100 MHz, mem 1850 MHz

https://i.imgur.com/qotclZT.jpg

https://i.imgur.com/f3JZ7pf.jpg

Those are my clocks.  My args: -w 256 -I 20 -g 1 and I let cgminer do the rest.

So, the 7970 sold on Amazon today. So that's cool.  NEED MORE CARDS THOUGH!  I wish those people on hardforum.com would get back to me!!!

EDIT:

Dropped core to 1140 and bumped mem to 1500, dropped vcore a little.  getting 691khash now.

My 7970 was having the same issue the OP had with being stuck in the 550s, even though I was at 1050 core/1680 memory.

Then I tried your suggested settings, and my hash rate shot up into the low 600's. I then increased memory back to 1680 and that got me another 20kh/s or so.

I started working my core and memory frequencies up and down, until I found a setting that maximized my hash rate while keeping my system stable.

My final settings are: --gpu-powertune 20 --gpu-fan 0-100 --gpu-engine 1024 --gpu-memclock 1692 -I 13 -w 256 --thread-concurrency 8192 -g 2

Final hashrate is about 710 kh/s if I'm not using the machine, down to 500 kh/s if I'm web browsing.

Thanks for the suggestion, scifimike12!

guys ive just reinstalled windows to install a clean driver.

going for 12.6 to check its performance against 13.1, but each time i open reaper OR cgminer is gets to connecting to pool then crashes.  any ideas?

could it be i need to wait on windows update?

Marss I'm working on the same issue, I'm glad you found the optimal setup. Can you please a screenshot of your clock speeds and all of that? I'm really new and just bought 2x7970 so I'm not really sure where to set those speeds and screenshot would be extremely helpful.

Thanks

On any 7970 1035/1500 is achievable and at exactly those clocks I have 4 different models, 12 cards total doing 720khs.

Try 24000 conc, gputhread 1 intensity 18 (more if it can take it)
Or 12000 conc, 2 GPU thread, intensity 13

Matauc - thanks, but i guess i dont understand what are you referring to '1035/1500'? which ones are these are below?

http://gyazo.com/819cedf4cc34e200b2b6f98758ae7f5b.png?1365684772

1035 is the core (GPU clock) and 1500 is the mem clock. On your SS it shows mem clock as effective mem clock, so its 4x. 1500 would be 6000.

Power limit is "power control settings" in CCC i assume? are we tallking + or - 20? changing these does nothing for my gigabyte 7970, currently peaking at a measly 536khash. also ANY memory overclocking at the piece of shit dies.  any thoughts? could it be overheating? can i change the point at which it kills itself in terms of heat?  im starting to think this card is shit - here are my essentials -

setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_USE_SYNC_OBJECTS 1


cgminer --scrypt -o http://xxxxxxxxx -u xxxxxxx -p xxxxxxx --shaders 2048 --intensity 13 --worksize 256 --thread-concurrency 8192 -g 1

g2 makes no difference.  higher thread concurrency kills khash to ~200.

Also, would one need to RESTART cgminer for changes to take effect? it seems even when i do OC the mem, nothing happens there and then, other than, of course, my computer dying.
 

might i just add, im running the 7970 on an old dual xeon processor mobo from about 6 years ago which has pci-e 1.0 x16.  

is there a POSSIBILITY that pci-e 1.0's power supply is the reason for not being able to overclock?  i'm sure this isnt the case as the voltages remain the same but could i be on to SOMETHING?

edit

im fairly newbie to all this.  would undervolting the card help in overclocking the memory? is it possible to undervolt the gigabyte i seem to remember reading it isnt.

thanks

Why ditch me?

You can't just take someone else's clock numbers and use them.  You need to tune your own clocks for your own card/mobo combination.

Actually, for clocks, you can. It has been demonstrated many times over that specific ratios at specific clocks, people reach similar sweet spots.

Where things vary a lot is with drivers/miner/thread conc/intensity/GPU thread. With those, you are on your own but still you should start with trying the common ones.

so frustrating, basically i can only clean install drivers with a clean install windows which requires waiting for automatic updates like .net.  are there RECOMENDED drivers for 7970's generally? im on 13.1, tried 12.6 but cgminer just wouldnt work on it, maybe due to app sdk

AFAIK, 13.1 is funky. The latest beta 13.3 (beta 3) seems to have unanimous good result.

Start at 2.6 SDK and upgrade to 2.7 if its funky, and 2.8 if that wasn't better.

so you'd recommend installing 13.3, trying that.  if not upgrade sdk to 2.7 etc to 2.8?

does 13.1 include sdk?

or is sdk downloaded independently.... < newbie

I just wanted to add my BTC0.02.  As it has been mentioned, for those new to this, you must work through the settings on your own.  I have two (supposedly) identical 7970s and they can't operate at the same clocks, but get nearly the same rates.

I have 2x Gigabyte 7970 OC edition, model GV-R797OC-3GD.  One at 1054/1500, the other at 1070/1500.  For both, --thread-concurrency=8192, -g2, and -I13  (cgminer 2.11.3, ubuntu 12.10, Catalyst 13.1 and SDK 2.8 ).  I'm getting 745 kHash/s and 750 kHash/s, respectively.  These two cards do seem to be different, as the first is locked at 1.170V and the second at 1.110V.  Well, they might not be voltage locked, but cgminer isn't able to change voltages.

Again, these settings probably won't work (exactly) for you, but they might be a good staring place as they aren't too aggressive.

The main point of my post is to say that -g2 should be tried on the poorly performing 7970s, also with lower thread concurrencies.  No matter what I do, I can't get either of my 7970s above 600 khps when attempting -g1 and high thread concurrencies.  On any version of cgminer that I've used, these settings produce nearly the exact same results.

(edited for spelling)

yeah of course. The clocks card can achieve are extremely variable factor, but I hold my point that in general, same clock/ clock ratio will often deliver similar results

I can't really agree or disagree (though my apologies for sounding as though I did disagree).  I suspect that your statement holds for same manufacturer/model cards, but not necessarily between manufacturers, even in the same series.

While this is a limited sample size, I also have an MSI Twin Frozr 7950 and an XFX Double D? 7950.  The XFX can do 1150/1750 for 680 khps and the MSI goes unstable at any memclocks past 1500.  My best settings with the MSI are 1080/1500 for 625 khps.

Keep in mind that the optimal setting is different for every card. I set my clock speeds at the cgminer command line. The full command would look like this:

i appreciate all the help and comments.  ive tried increasing each setting gradually.  if i do get ~600khash there are plenty of HW failures and it eventually bottoms out at 400  (btw 13.3 seems much more stable for OC). i find it hard to believe that even if i did hit the "sweet spot" i would get 750 khash.  just seems impossible.

seconds after typing that it seems i have hit a sweet spot....what an idiot. still only 620-650, but im not complaining

EDIT

nope computer crashed as usual with any OC

The thing about those sweet spots is they are not linear. You will usually hover around a certain value no matter what, then change the clock (or another setting) by like 5mhz and boom you jump from 600 to 720.

thanks, going to be a long evening

edit

also what are the benefits of using msi AB over CCC?

this is bonkers....1400 i get 530.  1405 i get 600. 1410 i get 530.  bonkers

right getting good hasrates now, but it seems theres a lot of stales, and my pool is returning a ~20-30% lower return than is quoted on my computer. what can be causing this?

HW errors will often be caused by too high/low thread concurrency or GPU threads. Play with those.

0 HW errors.. but ill try thread concurrency etc. increase concurrency by 128 increments? 12000 g2 and 24000 g2 dont work for me

Your pool calculates your hash rate based on the work it accepts. It doesn't count the stales.

Stales are a pool issue. Try a different one.

thanks for the info, may be a coincidence but at lower hash rates the numbers coinceded

I had a thread going about this issue as well.

It is here and may have some useful info -     
Lets Solve the unreliable 7970 issues.... (share your observations/input) - https://bitcointalk.org/index.php?topic=174036.0

Frankly the most frustrating aspects of this are two things, inability to duplicate results across different brands/versions of 7970s, and absolute non-linear results.  One card 1022/1244 might get 650k, 1022/1450 get 400k, and 1048/1720 get 720k.

They are the fastest cards but something tells me there is some sort of bug we have not found - for example most other cards show little performance variation and there are formulaic settings that can be plugged into any card to get close to max result.

Not so for 7970.  I still have 2 that are fubar.  I don't think I will ever get either one above 600k 24/7.

If its not HW errors, perhaps try different ports that your pool accepts or try a different pool (at least just to test).

24000 g2 shouldn't work for anyone. 12000 g2 doesn't for everyone, although I found a little bug for that config to work you must first run g1 then close, then g2.

If you're getting good rates and stales (not errors) you're on the right path. Means hardware config is right, now onto network :P. Thankfully, network is pretty much limited to pool config.

It's like a gearbox, with the core, memory speed, thread concurrency, gpu threads, and who knows what else, all as separate gears. If you don't have quite the ought ratio on any of the moing pieces, your hash rate suffers.

I used to believe the 7970 wasn't at full potential in the 750khs range for similar reasons, also since it seems more powerful than the difference in hash it gives. So always thought it might be miner optimization that would unlock its potential.

Recently though I have been getting hints towards the memory bus being the bottleneck. We can extropolate that from the Tahiti LE only getting 400khs because of its 256bus.

helpfull, thanks