chrysophylax
Legendary
Offline
Activity: 2814
Merit: 1091
--- ChainWorks Industries ---
|
|
February 10, 2018, 04:33:22 PM |
|
Ccminer is very unstable with 19 gpu's ..
I just wander why would anyone create system like that and get a constant problems. Comparing to total system cost, an extra motherboard+cpu+memory would not be that big investment, but release complexity by times. Well ... Following the strict rules that are required to run such a machine, there really isn't much issues, let alone problems. Once a design is set, it is very easy to duplicate. That is the reason you would go that route. Density plays a massive part if you are doing this on a larger scale. That is why, at least for CWI. #crysx
|
|
|
|
|
|
|
|
|
The forum strives to allow free discussion of any ideas. All policies are built around this principle. This doesn't mean you can post garbage, though: posts should actually contain ideas, and these ideas should be argued reasonably.
|
|
|
Advertised sites are not endorsed by the Bitcoin Forum. They may be unsafe, untrustworthy, or illegal in your jurisdiction.
|
|
|
|
buzzkillb
|
|
February 10, 2018, 04:43:56 PM |
|
Ccminer is very unstable with 19 gpu's ..
I just wander why would anyone create system like that and get a constant problems. Comparing to total system cost, an extra motherboard+cpu+memory would not be that big investment, but release complexity by times. Well ... Following the strict rules that are required to run such a machine, there really isn't much issues, let alone problems. Once a design is set, it is very easy to duplicate. That is the reason you would go that route. Density plays a massive part if you are doing this on a larger scale. That is why, at least for CWI. #crysx Maybe the price of adding the extra mb/cpu/ram/ssd needs to be spelled out, thats like an extra $400-500. Fit as many GPU's as you can for the farms.
|
|
|
|
joblo
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
February 10, 2018, 06:11:27 PM |
|
If it gets worse with more cards it's not likely a software issue, probably a resource issue like power.
man i have enough power i am using server psu's, don't waste your time Power was just an example of a resource issue. But I guess I am wasting my time with someone with a closed mind.
|
|
|
|
gzubeck
Jr. Member
Offline
Activity: 74
Merit: 1
|
|
February 10, 2018, 07:56:15 PM |
|
Ccminer is very unstable with 19 gpu's ..
I just wander why would anyone create system like that and get a constant problems. Comparing to total system cost, an extra motherboard+cpu+memory would not be that big investment, but release complexity by times. Well ... Following the strict rules that are required to run such a machine, there really isn't much issues, let alone problems. Once a design is set, it is very easy to duplicate. That is the reason you would go that route. Density plays a massive part if you are doing this on a larger scale. That is why, at least for CWI. #crysx Maybe the price of adding the extra mb/cpu/ram/ssd needs to be spelled out, thats like an extra $400-500. Fit as many GPU's as you can for the farms. Not worth it...I've killed two motherboards trying to get past six graphics cards on normal motherboards...I think if you want to get 12 gpu's you should just buy a special motherboard with 12 pci express slots. If you want to save power buy more powerful graphics cards if you can get them. I can buy surplus hard drives for $10 and I can buy a ton of amd cpus and motherboards for like $150 or less...the more crap you add to a system the more likely it's going to fail by crashing consistently. also, the more power you draw on any circuit will reduce the performance at some point. you definitely don't want to put 8 1080 tis on one circuit unless you have a special electrical output setup by an electrician that can handle that situation. I've even been hooking up power supplies on different circuits when I'm daisy chaining them for added amperage and power stability. just my two cents...
|
|
|
|
DumaxFr
Newbie
Offline
Activity: 49
Merit: 0
|
|
February 10, 2018, 08:42:25 PM |
|
Ccminer is very unstable with 19 gpu's ..
I just wander why would anyone create system like that and get a constant problems. Comparing to total system cost, an extra motherboard+cpu+memory would not be that big investment, but release complexity by times. Well ... Following the strict rules that are required to run such a machine, there really isn't much issues, let alone problems. Once a design is set, it is very easy to duplicate. That is the reason you would go that route. Density plays a massive part if you are doing this on a larger scale. That is why, at least for CWI. #crysx Maybe the price of adding the extra mb/cpu/ram/ssd needs to be spelled out, thats like an extra $400-500. Fit as many GPU's as you can for the farms. Perhaps in terms of initial costs but maybe not in terms of availabilty management (maintenance costs). If you split your single 19 gpu rig into multiple smaller rigs (let's say 3 rigs), you will probably remove many annoying factors. For example, you will probably remove PSU coupling, ease thermal dissipation, etc. You will probably have less stress with complete downtime; 2 rigs still working while one is on fault/repair. As stated by crysx, the only valuable reason to build huge rigs is "limited" space
|
|
|
|
p1r473
Copper Member
Newbie
Offline
Activity: 10
Merit: 0
|
|
February 10, 2018, 10:25:59 PM Last edit: February 10, 2018, 10:48:47 PM by p1r473 |
|
Getting problems with Tribus with the x86 build. Im using default intensity, I am not overclocked, and I am using latest drivers (390.77) Im having this issue on 2 rigs It doesn't happen immediately always, sometimes it takes 5-10 minutes to start Only started noticing it when I upgraded my drivers to 390.77 There is also a memory error hidden amongst the CPU errors Could this be due to CUDA 9.1 drivers? Using latest, 390.77... Notice in the github page it says: CUDA 9.1 required drivers are still in a early stage and unstable in my opinion. Can someone else please try Tribus with Tpruvot X86 with Nvidia driver 390.77? Let it run for 10-20 min.
|
|
|
|
firemylasers
Newbie
Offline
Activity: 3
Merit: 0
|
|
February 11, 2018, 12:41:17 AM |
|
I can't get any recent version of ccminer to compile on OSX. I have previously successfully compiled 1.7.6-r10-nanashi, but multiple attempts at compiling several versions of v2.2.4 (windows source, linux source, windows release source) and someone else's fork of v2.2 have all failed with the following errors: equi/cuda_equi.cu(2040): error: no instance of constructor "rt_error::rt_error" matches the argument list argument types are: (char [512]) detected during instantiation of "void eq_cuda_context<RB, SM, SSM, THREADS, PACKER>::solve(const char *, unsigned int, const char *, unsigned int, fn_cancel, fn_solution, fn_hashdone) [with RB=9U, SM=1248U, SSM=12U, THREADS=640U, PACKER=packer_cantor]" (2124): here
equi/cuda_equi.cu(2042): error: no instance of constructor "rt_error::rt_error" matches the argument list argument types are: (char [512]) detected during instantiation of "void eq_cuda_context<RB, SM, SSM, THREADS, PACKER>::solve(const char *, unsigned int, const char *, unsigned int, fn_cancel, fn_solution, fn_hashdone) [with RB=9U, SM=1248U, SSM=12U, THREADS=640U, PACKER=packer_cantor]" (2124): here
equi/cuda_equi.cu(2061): error: no instance of constructor "rt_error::rt_error" matches the argument list argument types are: (char [512]) detected during instantiation of "void eq_cuda_context<RB, SM, SSM, THREADS, PACKER>::solve(const char *, unsigned int, const char *, unsigned int, fn_cancel, fn_solution, fn_hashdone) [with RB=9U, SM=1248U, SSM=12U, THREADS=640U, PACKER=packer_cantor]" (2124): here
equi/cuda_equi.cu(1966): error: no instance of constructor "rt_error::rt_error" matches the argument list argument types are: (char [512]) detected during instantiation of "eq_cuda_context<RB, SM, SSM, THREADS, PACKER>::eq_cuda_context(int, int) [with RB=9U, SM=1248U, SSM=12U, THREADS=640U, PACKER=packer_cantor]" (2124): here
equi/cuda_equi.cu(1967): error: no instance of constructor "rt_error::rt_error" matches the argument list argument types are: (char [512]) detected during instantiation of "eq_cuda_context<RB, SM, SSM, THREADS, PACKER>::eq_cuda_context(int, int) [with RB=9U, SM=1248U, SSM=12U, THREADS=640U, PACKER=packer_cantor]" (2124): here
equi/cuda_equi.cu(1968): error: no instance of constructor "rt_error::rt_error" matches the argument list argument types are: (char [512]) detected during instantiation of "eq_cuda_context<RB, SM, SSM, THREADS, PACKER>::eq_cuda_context(int, int) [with RB=9U, SM=1248U, SSM=12U, THREADS=640U, PACKER=packer_cantor]" (2124): here
equi/cuda_equi.cu(2001): error: no instance of constructor "rt_error::rt_error" matches the argument list argument types are: (char [512]) detected during instantiation of "eq_cuda_context<RB, SM, SSM, THREADS, PACKER>::eq_cuda_context(int, int) [with RB=9U, SM=1248U, SSM=12U, THREADS=640U, PACKER=packer_cantor]" (2124): here
equi/cuda_equi.cu(2002): error: no instance of constructor "rt_error::rt_error" matches the argument list argument types are: (char [512]) detected during instantiation of "eq_cuda_context<RB, SM, SSM, THREADS, PACKER>::eq_cuda_context(int, int) [with RB=9U, SM=1248U, SSM=12U, THREADS=640U, PACKER=packer_cantor]" (2124): here
equi/cuda_equi.cu(2003): error: no instance of constructor "rt_error::rt_error" matches the argument list argument types are: (char [512]) detected during instantiation of "eq_cuda_context<RB, SM, SSM, THREADS, PACKER>::eq_cuda_context(int, int) [with RB=9U, SM=1248U, SSM=12U, THREADS=640U, PACKER=packer_cantor]" (2124): here
equi/cuda_equi.cu(2108): error: no instance of constructor "rt_error::rt_error" matches the argument list argument types are: (char [512]) detected during instantiation of "void eq_cuda_context<RB, SM, SSM, THREADS, PACKER>::freemem() [with RB=9U, SM=1248U, SSM=12U, THREADS=640U, PACKER=packer_cantor]" (2124): here
equi/cuda_equi.cu(2111): error: no instance of constructor "rt_error::rt_error" matches the argument list argument types are: (char [512]) detected during instantiation of "void eq_cuda_context<RB, SM, SSM, THREADS, PACKER>::freemem() [with RB=9U, SM=1248U, SSM=12U, THREADS=640U, PACKER=packer_cantor]" (2124): here
I've been unable to find any explanation online for these errors. A look through the source code was unproductive outside of revealing the origin of rt_error (equi/eqcuda.hpp appears to remap std::runtime_error to rt_error, not that this tells me much of anything). I've tried using different CUDA versions on the off chance that this was related to that, but 8.0, 9.0, and 9.1 all give the same error... I'm honestly not sure how the heck to resolve the issue at this point. Does anyone have any ideas? Some help would be greatly appreciated. I've gone through the past 30 pages of the thread looking for answers and have found nothing helpful. I found a solution. Here are the instructions for anyone else with the same issue in the future... Open equi/eqcuda.hpp in a text editor and find the following lines: #ifdef WIN32 #define rt_error std::runtime_error #else class rt_error : public std::runtime_error { public: explicit rt_error(const std::string& str) : std::runtime_error(str) {} }; #endif
Replace them with this: #ifdef WIN32 #define rt_error std::runtime_error #elif __APPLE__ #define rt_error std::runtime_error #else class rt_error : public std::runtime_error { public: explicit rt_error(const std::string& str) : std::runtime_error(str) {} }; #endif
Explanation: I had a hunch that OSX wasn't handling the std::runtime_error redefinition correctly, so I tried commenting out the original alternative redefinition and replacing it with the windows-type redefinition, which worked on OSX. In order to keep the code compatible with Linux and other similar platforms, I then modified it again to only use that redefinition type for OSX/Windows. In limited testing with versions compiled using CUDA 8.0 only I noticed no obvious issues caused by my change, but it's possible that some could pop up in the future. I'm not sure if I want to submit this as a PR to the repo yet. I'll probably want to conduct further testing before I do that.
|
|
|
|
bensam1231
Legendary
Offline
Activity: 1750
Merit: 1024
|
|
February 11, 2018, 10:39:12 AM |
|
6+ gpus are nothing but problems, new miners haven't figured that out yet so they aim for some efficiency by strapping as many GPUs to a system as possible. Every miner had the same exact thought when they first started mining, I've revisited the idea a few different times over the years, but the headaches just aren't worth it, just like running nix. I'll spend a little bit more money for a system that will maintain 99.9% uptime and when it goes down it wont be the same as three miners being taken offline at the same time. Nor will I waste a bunch of time troubleshooting frivolous issues that don't plague anyone else other then 6+ GPU machines and almost always require miner devs to fix (who usually don't and are lazy to begin with when it comes to real problems).
|
I buy private Nvidia miners. Send information and/or inquiries to my PM box.
|
|
|
chrysophylax
Legendary
Offline
Activity: 2814
Merit: 1091
--- ChainWorks Industries ---
|
|
February 12, 2018, 12:16:03 PM |
|
Ccminer is very unstable with 19 gpu's ..
I just wander why would anyone create system like that and get a constant problems. Comparing to total system cost, an extra motherboard+cpu+memory would not be that big investment, but release complexity by times. Well ... Following the strict rules that are required to run such a machine, there really isn't much issues, let alone problems. Once a design is set, it is very easy to duplicate. That is the reason you would go that route. Density plays a massive part if you are doing this on a larger scale. That is why, at least for CWI. #crysx Maybe the price of adding the extra mb/cpu/ram/ssd needs to be spelled out, thats like an extra $400-500. Fit as many GPU's as you can for the farms. This is very true. There is also the density loss involved with the smaller GPU/MotherBoard method as well. #crysx
|
|
|
|
chrysophylax
Legendary
Offline
Activity: 2814
Merit: 1091
--- ChainWorks Industries ---
|
|
February 12, 2018, 12:27:40 PM |
|
Ccminer is very unstable with 19 gpu's ..
I just wander why would anyone create system like that and get a constant problems. Comparing to total system cost, an extra motherboard+cpu+memory would not be that big investment, but release complexity by times. Well ... Following the strict rules that are required to run such a machine, there really isn't much issues, let alone problems. Once a design is set, it is very easy to duplicate. That is the reason you would go that route. Density plays a massive part if you are doing this on a larger scale. That is why, at least for CWI. #crysx Maybe the price of adding the extra mb/cpu/ram/ssd needs to be spelled out, thats like an extra $400-500. Fit as many GPU's as you can for the farms. Not worth it...I've killed two motherboards trying to get past six graphics cards on normal motherboards...I think if you want to get 12 gpu's you should just buy a special motherboard with 12 pci express slots. If you want to save power buy more powerful graphics cards if you can get them. I can buy surplus hard drives for $10 and I can buy a ton of amd cpus and motherboards for like $150 or less...the more crap you add to a system the more likely it's going to fail by crashing consistently. also, the more power you draw on any circuit will reduce the performance at some point. you definitely don't want to put 8 1080 tis on one circuit unless you have a special electrical output setup by an electrician that can handle that situation. I've even been hooking up power supplies on different circuits when I'm daisy chaining them for added amperage and power stability. just my two cents... Well, We have built HUNDREDS of systems, and the experience tells us that this smaller GPU to MotherBoard ratio is more of an 'issue' than not. So for us who have more than 7 or 25 or 70 GPU's to place, density plays a big part - unfortunately. Agreed there is slightly more headache than normal, but that is with ANYTHING you put your hand to when it comes to building something out of the norm. TWO GPU's were a massive things we I first started mining, and 4 GPU's was a monster machine. Now look at the 'minimum' spec you will need to even start mining well. So it is worth it for us to attack the density factor with gusto, and (in our case) develop a MUCH better cooling solution that magnifies the mining effort by almost 3 times. So your two cents are worth much more than that and I would agree in that respect also. I agree also with the power - which is also why we have designed (and have our first prototype) CWI-Power system fully functional, which we hope to make available for those that WANT to mine in a much more extensive fashion. Circuits in most houses are limited to a small amperage, so our newest design does away with the singular circuits and raises the bar for spreading the circuits within ONE system. We are still improving the MK I model, and MK II will be much more desne again and able to cater for small farms easily. BTW - ALL our cards are a mix of Aorus 1080TI Xtreme WindForce and recently Aorus 1080TI Xtreme WaterForce systems, so they do draw a lot of power depending on what Algo is mined, hence the need for density, power availability, and soon we will be working on efficiency, though that is not the priority currently. #crysx
|
|
|
|
chrysophylax
Legendary
Offline
Activity: 2814
Merit: 1091
--- ChainWorks Industries ---
|
|
February 12, 2018, 12:32:19 PM |
|
Ccminer is very unstable with 19 gpu's ..
I just wander why would anyone create system like that and get a constant problems. Comparing to total system cost, an extra motherboard+cpu+memory would not be that big investment, but release complexity by times. Well ... Following the strict rules that are required to run such a machine, there really isn't much issues, let alone problems. Once a design is set, it is very easy to duplicate. That is the reason you would go that route. Density plays a massive part if you are doing this on a larger scale. That is why, at least for CWI. #crysx Maybe the price of adding the extra mb/cpu/ram/ssd needs to be spelled out, thats like an extra $400-500. Fit as many GPU's as you can for the farms. Perhaps in terms of initial costs but maybe not in terms of availabilty management (maintenance costs). If you split your single 19 gpu rig into multiple smaller rigs (let's say 3 rigs), you will probably remove many annoying factors. For example, you will probably remove PSU coupling, ease thermal dissipation, etc. You will probably have less stress with complete downtime; 2 rigs still working while one is on fault/repair. As stated by crysx, the only valuable reason to build huge rigs is "limited" space Space and power availability. Two of the main reasons people go this route. We are 'prototyping' a number of these motherboards filled with these cards and we are looking at the maintenance side very closely. As you mentioned, the bigger you grow in the singular unit, the more headaches you open yourself up to. However, the flipside of that also, is that once you have a tried and tested system in place (which we now do), the maintenance time is reduced to such a small factor, that it is more of an issue to go the smaller GPU/More MotherBoard way. We are finding that out literally as we chat about it. Hence why we have also purchased another 7 x 19GPU motherboards, and our 'standard' way of building them, which are a lot easier to maintain. One only needs a system in place and a method maintenance, and it becomes easy. When all these 7 MotherBoards are populated, we will let you know what 'other' headaches may arise. So far - the first two are not only doing well, but running like wildfire. BTW - only 16 GPU's are in each due to the nVidia driver limitation. Duplicated Linux and you are running in an hour including the physical build. The first build/prototype took a full weekend. So a little over an hour from unboxing to full build and mining is a massive achievement in comparison to a full weekend. #crysx
|
|
|
|
pallas
Legendary
Offline
Activity: 2716
Merit: 1094
Black Belt Developer
|
|
February 12, 2018, 12:52:09 PM |
|
Sorry for the off topic, but since we are at it... What do you guys think of these PSUs you can find on amazon, very powerful and cheap, with the right cables for a 6 gpus rig: https://www.amazon.co.uk/dp/B0787BPNQ8(price is 100-150$ for a 1600W gold rated unit). I would get one just for curiosity, given the price, but I fear that it might fry some components.
|
|
|
|
chrysophylax
Legendary
Offline
Activity: 2814
Merit: 1091
--- ChainWorks Industries ---
|
|
February 12, 2018, 01:43:55 PM |
|
Sorry for the off topic, but since we are at it... What do you guys think of these PSUs you can find on amazon, very powerful and cheap, with the right cables for a 6 gpus rig: https://www.amazon.co.uk/dp/B0787BPNQ8(price is 100-150$ for a 1600W gold rated unit). I would get one just for curiosity, given the price, but I fear that it might fry some components. We looked at a number of those Pallas. What we found were three things. 1 - The PSU are very cheaply made, as are the components. This lends itself to all sorts of things going wrong. 2 - The power delivery is subpar as opposed to say SEASONIC style of PSU's (like Corsair for example) where the power delivery is not tainted and steady. Spikes happen, but the better PSU units subdue these internal issues MUCH better than the any of these cheaper units. 3 - Warranty returns are a major pain, and sometimes costly for the amount of time these cheaper units are actually IN warranty. These factors are why we choose the highest platinum series PSU that are local AND a 24hour replacement turnaround. Not three weeks like those cheaper ones. Apart from that, Algos used in the ccminer-tpruvot and other quality mining apps that have variable hashfunctions active (like the latest X16R and TimeTravel/TimeTravel10) push the PSU really well with regards to power draw. A simple PSU made of poor quality components tend to die much easier, and I have had MANY (when theFARM was first started years ago) actually burn and melt the black Molex connectors TO the PSU as well as in some cases, the GPU connectors themselves. #crysx
|
|
|
|
SparkyU
|
|
February 13, 2018, 02:18:11 AM |
|
Why can't I force intensity? For some reason it's setting it to whatever it wants when I'm mining this scrypt coin, I haven't had these issues with any other algo. The --intensity flags worked just fine to throttle my gpu before. I can't have it running at full blast, at that's what it sets it to.
This is my troubled config ccminer-x64-75 -d 0 -a scrypt -o stratum+tcp://lycheebit.com:3433 -u D6rhrJqCK3DJuNZftuEEo181vjdPvhtLDF -p c=DFS --cpu-priority 5 --no-autotune --intensity=10
I've tried playing muscial chairs and moving the parameters around and switching --intensity=10 to -i 10 and so on. I'm confused why this one is giving me a headache. No matter what I try the intensity is always set high causing 100% gpu load.
|
|
|
|
mangoo
Newbie
Offline
Activity: 23
Merit: 0
|
|
February 13, 2018, 05:56:22 AM |
|
Is it possible to add the following feature for solo mining?
--no-getwork disable getwork support
|
|
|
|
bensam1231
Legendary
Offline
Activity: 1750
Merit: 1024
|
|
February 13, 2018, 12:40:18 PM |
|
Apparently Epsylon feels the need to police everyones rig, sad days. Now we get to wait till someone forks Epsylons branch changing only one thing, which is allowing intensity changes.
|
I buy private Nvidia miners. Send information and/or inquiries to my PM box.
|
|
|
|
gzubeck
Jr. Member
Offline
Activity: 74
Merit: 1
|
|
February 15, 2018, 07:31:01 AM |
|
Why can't I force intensity? For some reason it's setting it to whatever it wants when I'm mining this scrypt coin, I haven't had these issues with any other algo. The --intensity flags worked just fine to throttle my gpu before. I can't have it running at full blast, at that's what it sets it to.
This is my troubled config ccminer-x64-75 -d 0 -a scrypt -o stratum+tcp://lycheebit.com:3433 -u D6rhrJqCK3DJuNZftuEEo181vjdPvhtLDF -p c=DFS --cpu-priority 5 --no-autotune --intensity=10
I've tried playing muscial chairs and moving the parameters around and switching --intensity=10 to -i 10 and so on. I'm confused why this one is giving me a headache. No matter what I try the intensity is always set high causing 100% gpu load.
Given how many new and old coins are using neoscrypt it would be nice for tpruvot to actually have an update that creates stability with newer nvidia drivers. The last update I've seen is in may of 2016. maybe I've missed a few as I haven't really been paying attention to neoscrypt coins for the last six months and I think this has been part of the problem in general. it would be easier to mine anything but neoscrypt coins including ethash, equihash, lyra2v2 etc.
|
|
|
|
tomos3k
Newbie
Offline
Activity: 5
Merit: 0
|
|
February 18, 2018, 05:51:16 PM |
|
Hi, I have an issue with my CCminer 2.2.4. I'm using it to mine XVG and it's been working just fine until I swapped my GTX 1050 to 1060 for a couple of days, and then when I installed the 1050 again the miner stopped showing the power consumption of the card. I reinstalled the drivers (performed a clean install), installed the card itself again on the motherboard and still nothing - it shows 0W all the time.
Any idea what could've caused it and how to fix it?
|
|
|
|
k0stas
|
|
February 18, 2018, 07:24:27 PM |
|
Hi, I have an issue with my CCminer 2.2.4. I'm using it to mine XVG and it's been working just fine until I swapped my GTX 1050 to 1060 for a couple of days, and then when I installed the 1050 again the miner stopped showing the power consumption of the card. I reinstalled the drivers (performed a clean install), installed the card itself again on the motherboard and still nothing - it shows 0W all the time.
Any idea what could've caused it and how to fix it?
Downgrade the driver to 384.90, the newer drivers have this problem.
|
|
|
|
|