Bitcoin Forum

There has been a lot of speculation about the GTX 680's hashing power, but no real results with optimized mining code. There are many brands being sold at Newegg now.
http://www.newegg.com/Product/ProductList.aspx?Submit=Property&Subcategory=48&Description=&Type=&N=100007709+600315498&IsNodeId=1&IsPowerSearch=1&srchInDesc=&MinPrice=&MaxPrice=&PropertyCodeValue=679%3A289341

  Mine should arrive Tuesday.  Will post results when I can.

Well, lets be real about this:

More than 24 hours have passed since Kepler appeared in the market. The 7970 at launch already had tests performed on mining, and results were already published in reviews. This is probably the fourth (+fifth?) topic on this card in this forum, yet no data is available for its performance in mining on the internet. Only speculative theoretical numbers have appeared in this forum, though we know the card is "castrated" in at least one sense, and "leaner" than its previous generation for graphical operations.

nVidia site and online reviews talk about games, games, games, and some potentially interesting/relevant synthetic benchmarks.  There are reviews that discuss Kepler "GPGPU functionality" to be mainly in Quadro and Tesla.

Seems convincing the the lack of publicity on mining might indicate this card is best suited for killing dragons and firing/reloading weapons?

Economics of PW cracking, nvidia & AMD:
http://www.youtube.com/watch?v=CKNW5AUo-dM#t=13m22s

I have someone with a card who should be benching this with RPC-miner shortly

Please try with both RPC-miner and Ufasoft (https://bitcointalk.org/?topic=3486.0) (-t 0).

That's because AMD outright said in their slides that you can mine better using their cards.  Since we already know you can mine very well with AMD, it's not surprising people had tested it akin to a program in benchmarks or very soon after launch.

Even if we know the 680 isn't a miner's card in a sense, I know people just want to know what it'll put up, myself included.  Since the community as a whole doesn't really equate Nvidia cards to being good for mining, it's easy to ignore it.

Once people test, I'm sure #s will come soon enough.

The person who I know tried to get it running with rpcminer-cuda but is having trouble getting the executable to run correctly (problems with the CUDA modules).  I guess we'll have to wait.  :(

tacotime,

Ask him to try with Ufasoft, should be hassle free. Use it like: bitcoin-miner-64.exe -t 0 -o http://worker_name:worker_password@mining_pool:port

Info about recommended driver versions and the CUDA toolkit update:
http://forums.nvidia.com/index.php?showtopic=225731 (http://forums.nvidia.com/index.php?showtopic=225731)

Still no answer as to how this card can perform under bitcoin mining huh? This card appears to be sold out in a number of places, but still no actual figures, surely someone must have done some experimentation with Kepler by now at the very least. Still curious about this brand new piece of hardware, anyone have any results?

I'd like to see some real world numbers too.

Me 3.

From the Newbie section: https://bitcointalk.org/index.php?topic=74215.0

Try RPCMiner-CUDA with the CUDA dev kit, something doesn't sound right.

  Just got my card in today ;).  Going through the paperwork part (registering etc. / grabbing drivers) and going to be installing it shortly.  Should have it benched by tonight.

watching closely!

Apparently I can post in here now. I'll see if I can get this going. I agree the number doesn't sound right, but that's what I got with what I did (ufasoft 0.28 binary, win7 x64, 301.10 drivers).

  Hey guys,

  Sorry for the delay but had errands this evening while trying to also sit and benchmark lots of stuff.

  Apparently Kuro is indeed correct in his results.  I figure stuff simply is not optimized for it just yet or that there must be some difference in how the 680 utilizes CUDA compared to the 580.

  Since I'm no command line expert, I just did the following:

  1.  Launch GUIMiner (my version was 2011-8-24) which I'd normally use just fine.  Tried the same settings for my 680 that worked with my 580.  Nothing.  
  2.  Updated GUIMiner to the latest I think (2012-2-19) and tried the settings again.  Nothing.
  3.  Started launching miners by attempting each of the miner options.
  
  OpenCL = Works but might blank the display.  Crashed the system once I stopped it but did get mining results.  System seemed stable while mining.
  Phoenix = Launched but sat at "Connection Problems"  No luck despite being configured correctly.
  CG = Same as above.
  CUDA = Same as above.
  Ufasoft = Same as above but without any "Connection Problems" at all.  Start mining button did nothing.
  Other Miner .. not sure what to try here.

  Results with the one or two miners I tried (forgot to name the tab so Phoenix may work as well as OpenCL):

  http://dfullerton.smugmug.com/photos/i-hQKTvnQ/0/X3/i-hQKTvnQ-X3.jpg (http://dfullerton.smugmug.com/photos/i-hQKTvnQ/0/X3/i-hQKTvnQ-X3.jpg)
  http://dfullerton.smugmug.com/photos/i-hQKTvnQ/0/M/i-hQKTvnQ-M.jpg

  Indeed, we're at ~110MH/sec.  This is stock w/GPU boost.  I have not yet attempted to even mess with power boosting and overclocking the card but either way it doesn't really matter at this point in time.

  Here's hoping that a proper update to things will eventually happen!  Thanks for the patience everyone.  As far as games go outside of BTC mining, this card rocks.  But alas, that's not why we're here.  I'm sure once some optimizations are made, it might work better.

  Drivers are the Nvidia 301.10 launch (only ones so far) and running Windows 7 x64.

Ew, This is such a double edged sword for me, Im an AMD fanboy, And i was going "HA nvidia sucks a mining ahaha amd ftw"
And then the gtx 680's were announced, And i thought "oh shit, 1536 CUDA cores (assumed SPcore equivalency) looks like it'll get around 300ish! Good job nvidia! now AMD will try harder!"
annnd 110mh/sec...
It just doesnt seem right... I mean like, Yeah i was hoping for failure, but not the kind of business wrecking failure

Lots of stream processors for mining - right idea. Lousy integer performance: Fail.

Darn and I hate ATI so much.

  Yeah, as you guys said it just doesn't seem right in theory.  Whether the CUDA miner or something else needs to be adjusted or re-written for the Kepler architecture is just my guess, it's too bad to not see it mine well.

  FWIW, Folding @ Home (Stanford) has also said they have to recompile their program to work with the new architecture.  Thus while normally Folding is a strong suit for Nvidia, it also is "broken" on the Kepler cards for the moment.

  Might also just need a newer driver.  Hopefully if that's the case, Nvidia won't take 3 months to release one.  Normally it seems like they get new cards' drivers out very fast (updated ones I mean, rather than the stock).  Here's hoping!

Folding @ Home is broken because nVidia broke it.  The cards have way less DP/SP floating point performance than a 560 Ti.  It's part of nVidia's new tactic of making their lower end architecture useful only for gaming.

just means amd is gonna get that much more money from people that want to use gpus for gpgpu since amd gpgpu is hugely better than anything nvidia has ever offered

Bumping this because I want some REAL results this time around.

I think that people have given up on the Kepler for mining just because the kernels we have now are totally NOT optimized for it and that is why it gets only 100 MHash/s.

With more kernel development I am pretty sure that this card will be a killer mining card.

Maybe someone is holding back the code and mining at over 400 mhash/s ???

Too bad I am not a coder otherwise I would have done this myself.

I just find it really hard to believe a GTX 580 gets 140 mhash/s and this POS Kepler gets lower at 100 mhash/s ...

Also, while we are on the subject, how is this doing for F@H : did they develop that new Kepler-optimized client that Nvidia was promising or not ?

F@H has not released a Kepler aware core with optimizations yet. Using the currently available cores GTX680 is performing on par with GTX 560 Ti.

And despite that there are ppl who love nvidia and hate amd. Nonsense.

Seriously, who want to buy a nvidia? Not me.

I have tried to recompile the Ufa-miner and the rpcminer-cuda, but for various reasons it didn't work.
Now I have written a very simple miner using the latest CUDA toolkit, just to see what the speed is like.
Since I have written the kernel from scratch, and the program is not doing longpolling, there's some room for improvement.

If somebody with a GTX680 would like to try it you can download it there:
http://www.sonstiges.org/bitcoin/ICU-Miner.zip (http://www.sonstiges.org/bitcoin/ICU-Miner.zip)
Usage:
(devicenumber, gridsize and blocksize are optional)
For example:

Here's the ugly source code:
http://www.sonstiges.org/bitcoin/ICU-Miner-source.zip (http://www.sonstiges.org/bitcoin/ICU-Miner-source.zip)

Too bad I don't have a GTX 680.

Somebody needs to try this ASAP !

I just tried it now but it doesn't seem to work very well.  Desktop is completely unresponsive and the GPU usage doesn't stay pegged at 99%.

http://oi40.tinypic.com/3359urn.jpg

Nvidia just cut all integer and double precision support to fit 1536 crappy shaders. We have to wait for the high end Kepler.

Care for a link ?

As much as I love AMD vs Nvidia mining competition it really does look like Kepler totally SUCKS for mining :'(

Or maybe the GTX 680 is the card that is for gaming only and designed to crush 7970 prices ?

Thanks !

YES... please! Then again, there's the 7990 down the pike.

Then you'd have to decide between a $499 7970 (if the price drops) or $849 7990 for each slot.

Please let the 7990 hit 1.3Ghash/s.

GTX 680 mining does work but the performance is very disappointing at the moment. I think there may be some gains from drivers in the future though, since I'm only showing ~70% power usage. However, I highly doubt the GTX 680 will ever be competitive as a mining card.

Showing 123.40 Mhash/sec in Phoenix 2.0.0 with the following settings:

Vectors seem to reduce performance. Enabling goffset will cause the miner to error out and exit after about a minute. GPU clock was a constant 1241MHz for the test. (using +144 offset) Maximum load temperature was 67 C.

I think we are going to be watching for a long time because it seems nobody is interested in getting Nvidia to be competitive in mining against AMD  :'(

It's not that they're not interested. They just know that they will only be wasting their time and think "Why bother?".   :)
 

Apparently nVidia is releasing a Kepler version of Tesla on May 14.  What seems very interesting is that it is a 7 billion transistor chip, the GTX 680 is 3.5 billion transistors.

http://www.brightsideofnews.com/news/2012/4/20/nvidia-to-launch-a-7-billion-transistor-kepler-gpgpu-tesla-boards-on-may-14.aspx

Interesting. With the number it would be tempting to think 2xGK104, but given how crippled GK104 is in FP64 performance I can't see that happening in a Tesla product. If this actually is BigK, it's way ahead of when most people expected it to ship.

Actually NVIDIA is most likely just crippling in the drivers like they did for the GTX480 vs Quadro6000/Tesla C2050 by hobbling the drivers. Anyways I dont think GK104 has ECC support, so it is likely this new card will be GK110.

You're basing this on what? Every review that talks about it is pretty emphatic that FP64 performance just isn't there, which is one of the reasons NVIDIA was able to pack so much performance in such a little die.

From Anandtech

Mining doesnt use FP -_- Also not sure why you quoted the part on GF114. The GTX480 and Quadro/Tesla are GF100, not GF114.

Edit: Anyways if it is right that 1 of the 3 blocks cannot do anything but FP mathematics, that means that there are still 1024 cores which can. Coupled with the halved speed, it should still match the GTX580. I wonder why the performance is so poor then.

I dont buy the new architecture = no tuned miner theory. It should be relatively easy to write a compiler for a MIMD architecture like the GTX680 and get 99% utilization with data parallel tasks, like hashing.

Only the first sentence really talks about GF114, and it's just there to give context as to the difference in approach between GF114 and GK104.

As for mining, that's true. However, Telsa cards are not sold for mining. Most scientific computing uses FP, so when discussing whether the new 7B transistor Telsa card is 2x3.5B GK104 or GK110, floating point performance is a huge factor.

It's not 1/3 blocks. There's 192 normal CUDA cores in an SMX and only 8 FP64 cores. That's why performance is so poor.

Ahh, I misread and equated "blocks" with "SMs". Apologies.

But as far as I know, mining is integer performance, which should go with the normal CUDA cores right? Why is the performance so poor then?

As for FP performance, most scientific applications should use FP32 to save memory/time (although with the amount of lazy programmers I deal with everyday, that may not be the case....)

Many scientific simulations require the precision of FP64, so FP32 isn't really an option. That's why you pay thousands for Telsa instead of just buying a GTX580.

As for AMD vs NVIDIA mining performance, there are other factors there. AMD supports a couple instructions that significantly improve hashing performance. You can't directly compare shaders*clocks between the two.

Except 6970 FP64 performance is even higher than Tesla.

People don't buy Tesla for its FP64 performance. They buy it because it is a dynamically-scheduled architecture (which tends to extract a ton of thread-level parallelism), with ECC support (important if you are into HPC), and TCC mode.


I'm talking about Fermi vs Kepler

That, and the better support NVIDIA offers to HPC developers. Interestingly, the hardware scheduler is gone in GK104.

For that, I have no idea. I would imagine it's a lack of optimizations for the new arch, but I don't know much about the CUDA miners.

Does anyone have a GTX680 here? I would be interested in the results of CUDA-Z.

http://cuda-z.sourceforge.net/

If the integer performance there is good, it's probably an optimization issue. If it is bad, then more than likely the hardware is just not as good as Fermi in some way.

7900 serie has ECC support

and updates on 680 hashrates?

Nope it looks like Nfail again.

Too bad because I would have loved to get rid of this shitty ATI mining monopoly game.

>:(

the thing that got me excited was the quad-GPU VGX

edit: the VGX (http://www.nvidia.com/object/vgx-boards.html) has only 786 cuda cores total across all 4 GPUs?  SAD!  no longer excited.

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

and the endless racks FULL of GPUs

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

and since they offer enterprise cloud computing, wondering if we could get all those GPUs hashing?  EXCEPT the fact that nVidia sucks balls at mining.

It seems more reasonable to me that it would be 768 core per each of the GPU.

Yeah but it says

GPU Specifications
Number of GPUs    4
Total NVIDIA CUDA® Cores    768
Shader Perf (TFLOPS)    1.3
Power (W)    150

So.. Total CUDA cores = 768?  What a rip!

I'm getting set up for doing some test mining on my dual 680's now.


Tested, Here are the results:
Gives more stable performance on the second card in my SLI Setup.
http://dl.dropbox.com/u/20040127/CUDA-Z.txt

I'll be interested to see performance with Cuda 5

Yup, it's the integer performance that is killing the hashing results, GTX580 is capable of almost 1000000 Miops and it does ~180 MH/s.

http://forums.nvidia.com/index.php?showtopic=192076

Fail again most likely. Nvidia are a bunch of retards for letting AMD and their shitty drivers be the only mining GPUs.

There's a table of operations per clock cycle per SMX (http://forums.nvidia.com/index.php?showtopic=229473&view=findpost&p=1411769)

Given that K20 has 15 SMX and K10 has 16 SMX and GTX680 has 8 SMX and we all know how fast GTX680 is at mining.... it doesn't look hopeful.

I think the mining community is too small for nVidia to focus on.  Thankfully!  If the whole world was mining the difficulty would be pretty darn hard.