Bitcoin Forum

Hi,
Will give $50 worth of btc to anyone who can help me out. Assembled my 4 x 1070ti mining rig about 3 weeks ago has a 1200w corsair platinum psu, asus h270 plus mobo, 2 x 4g ddr4 ram, running windows 10

So my first problem is when i benchmark each singular gpu on its own i get good hashrates, about 460 m/hs on equihash. But when i run more then one gpu the hashrate of the others drops significantly. When i try run all 4 together i get like 720 m/hs which is trash for all 4 gpus. I have tried uninstalling and reinstalling the latest drivers for my gpus and a clean install of windows. Does anyone have any tips for this?

My second problem is with all 4 gpus running any mining software i use crashes after a few minutes of mining and just freezes my pc. I tried getting 4g more of ram to bump up my system to 8gb (improved overall stability but still get crashes). I really am stumped here and have no idea what to do with this one, have tried searching through forums but no solution has helped. Thought it was a nicehash issue, then started using nemosminer and mining pool hub. Still crashes and shit hashrates...

So, i think this may be an issue with faulty hardware. My 2 guesses are either my powersupply or my motherboard. I am guessing one of them cant support to provide enough power across the rig with more then 1 gpu. It cant be my gpus because when i plug each in separately  and benchmark them they all give me good hashrates individually. Just not with more then 1 plugged in...

I have spent the past 3 weeks trying to solve this and its doing my head in, alot of lost revenue aswell so if any one has any ideas please let me know.

Thanks,

Nic, I see from your other posts that you have already adjusted your pagefile size. My first suggestion would be to make sure you are using the 4 pin molex connectors to power your riser cards and NOT the SATA rail. Try to only use one molex rail or wire for each riser card, but two risers on each wire should work. Don't use SATA plugs on the riser cards, they cant supply enough power and will cause all kinds of issues like you're describing such as locks ups and crazy unexplained issues. I found out the hard way that the SATA cables don't carry enough wattage to power the riser cards. The graphics cards can pull as much as 75 watts from the motherboard, but since the GPU's are plugged into the riser cards instead, the riser cards need to supply this power. From what I understand, the SATA cables can only supply around 50-55 watts (depending on your PSU). How I discovered this was that the rig kept crashing when I would begin to mine. Once I dropped the GPU voltage (TDP), the rig did better but still had issues. Once I identified the problem and plugged in the molex connectors instead, the issue went away. I was told that the molex connectors supply up to 135 watts (more or less depending on your PSU), so try to only plug in one or at most two cards on one rail. Also, DO NOT use the SATA to 6 pin adapters supplied with some of the riser cards. You are still only getting the max wattage supplied by the SATA plugs and any extra wiring or plugs only add resistance and reduce power and create more heat. Use as few plugs as possible and don't use excessively long wires for anything. I run my GPU's and risers on separate rails and have no issues at all. I reset my rig every 4 or 5 days just out of caution but haven't had any issues since I learned how to setup my rig properly. My rig seemed a little slower with the risers compared to when the cards were plugged directly into the motherboard, but the open air case allowed me to increase voltage and overclock to achieve higher hash rates than I was seeing with my tower case as well as provided room for more than 3 GPU's that i was limited to in the tower case. I spaced the cards about two inches apart from each other to help with cooling. The 1080ti FTW3 I have happens to vent some hot air out the back, so I ended up moving it to the end of the case so it doesn't blow hot air right onto the 1080 FE cards. This helped with cooling as the 1080 next to the ti originally was running warmer and slower due to getting all the ti's hot exhaust. When the cards were originally too close together, they would slow their hash rate tremendously after a couple minutes of heating up. I was originally running two 1080's and one 1080ti on a single 850 watt PSU with no issues. I now have two PSU's with a dual PSU adapter from Amazon. I run Windows 10 with no issues at all and get hash rates just as high as guys running Linux and other OS's. There is nothing wrong with Windows 10 if your rig is setup correctly. Windows is not the cause of your issues and there is no need to change OS's unless you simply prefer one over the other. If you have issues getting Windows to see all the cards, plug them in one at a time and boot Windows before installing the next card each time so Windows can see each card and install the drivers one at a time.

Also, mining clients such as ccminer can have issues if you are overclocking your cards. Overclocking the memory on my Nvidia rig was causing alot of issues with ccminer crashing. I set all of my cards to the default settings, and they ran fine. After seeing this, I overclocked one card at a time and ran the rig for about 8 or 10 hours to make sure there were no issues before overclocking the next card. In the end, I had one card that I could not overclock the memory at all without causing issues, but I was able to get the highest GPU core speed out of that card for some reason. I run 1080's and one 1080ti on my current rig. I bought a 1070 yesterday for testing but haven't added it to the rig yet.

Firstly i would recommend using nvOC, but with windows make sure you have changed windows power settings to "high performance" and do the following. Click “Change advanced power settings” and look for “PCI Express” -> “Link State Power Management“, make sure it is set to “OFF“.  Your motherboard may have a specific mining BIOS made specifically for mining as well, i would research that first.

Hi. I have been mining for 2 years so I do have a lot of experience with it so maybe I can help....

Firstly I know it's a pain in the butt but I recommend you get rid of Windows 10 and try 8.1 or even better (if you can wrap your brain around it) use one of the Nvidia specific linux mining distros - nvOC is the first obvious choice https://bitcointalk.org/index.php?topic=1854250.0 (https://bitcointalk.org/index.php?topic=1854250.0) but it will require a steep learning curve if you have no Linux experience. It is worth it in terms of stability especially but like I said - if you don't know any Linux it will take a lot of work.

My next suggestion is much easier to implement - make sure that your Windows virtual memory is set to more than the total of vram on your GPUs. So in your case, I think the 1070ti has 8GB of vram? If so 4 x 8 = 32GB so maybe set your virtual memory to 40 GB. If you don't know what I mean....

right click on "my computer" icon -> properties -> advanced system settings ->  in the "advanced" tab of system properties click the "settings" button in the "performance" box -> in performance options choose the advanced tab -> click the "change" button in the "virtual memory" box -> uncheck the "automatically manage ....etc" box -> click the "custom size" point and set the minimum and maximum to 40960 MB, or maybe minimum to 32768 and maximum to 49152. Then apply the changes and you may have to reboot. Also obviously the virtual memory has to be on a drive (or drives - you can have multiple virtual memory blocks) that has that much space spare on it. If you are only using one drive or partition for Windows it will obviously need to be on C: but will only work if C: has 40 or 50 GB of spare space. It's a very good idea to put the virtual memory on your fastest drive, so if you have a hard-drive and an SSD, put it on the SSD if you have the space.

The next thing that I will say is that 80% of the time when I have had problems, it has been because of a bad riser. Risers are made very cheaply in China and it seems they don't test them before shipping, because I have probably bought nearly twice as many risers as I have ever used because they either didn't work when I got them, or failed withing 1 or 2 months of using them. I prefer to use the USB 3 risers that are powered via a 6-pin PCIe 12V connector. You did say though, that you have hash-rate tested them all individually so the riser problem seems unlikely.

I'm not familiar with that particular motherboard, but I think your power supply is easily big enough, but I haven't used 1070ti before... I'm guessing they use more power than a regular 1070 but one of my rigs is 4 1070s and a 1080 and that uses around 1000W. An important point there though is that I have set the powerlimit on those cards to 80% using MSI Afterburner. It's very inefficient mine equihash with your GPUs at 100% powerlimit unless you have free or very cheap electricity. Are you using any overclocking/underclocking software like Afterburner or Nvidia Inspector or whatever? I'm guessing not since it would be obvious that you should try to start mining at factory clock settings and under/overclock later when you are stable, right?

What software are you mining with? Have you tried both EWBF and DSTM for equihash? Have you tried mining any other algorithms like ETH, Lyra2rev2, neoscrypt, etc?

Anyway - I think the most likely one of my suggestions is your virtual memory needs to be much higher, especially considering you have only 8GB RAM. I know RAM is expensive, and everyone says use as little as possible, but I have always found that you get much more stability on a rig with 4 or more GPUs when you have at least 8 GB RAM. Most of my rigs have 12 GB on them, one 8 GB stick and one 4 GB stick. It doesn't matter that it's not dual channel - it's the extra amount that seems to help. Unfortunately since you are running Windows 10, there is a lot of crap running in the background tracking everything you do so it's using a lot of RAM.

Good luck, I hope this helps and please ask me specific questions if you have any.

Here is what I would suggest.  If you try it and it works, you can tip me then...

1) Ditch windows, download ethos, run it.  If it fixes your hash rates, then buy a license for it on gpushack.  I have guys who don't even know how to power on a laptop building and running mining rigs with ethos, it just works.  Plus you get remote monitoring built in.

2) Make sure you have good quality cables, using the 6 pin adapters, to your risers.

3) Try not to run the riser cables right next to each other, there can be cross talk and cause pcie bus signal integrity issues.

4) If problems persist, it's likely your risers, buy a different brand.  I've used most of the ones for sale on newegg with good luck.

Hope this helps.  If you have any questions or need help with your ethos config, feel free to pm

In windows 10 do the following as a starting point...
1. Increase page file size to at least 16GB
2. Change power profile to high performance
3. Set afterburner tdp to 65-70%, default clocks, once stable then +100 core, +500 mem as a starting point after crashes are stopped
4. Report back the cpu usage when mining 2 cards then all cards, +cpu specs please, what is it?
5. Pcie power cable from PS to cards and riser should be on the same 12v rail.  If you use 12v rail 1 for all risers then other rails for cards it can possibly create problems.  Not sure if your PS has a single or multiple rails.
6. Bad power wiring...never exceed 300w per pcie cable is my rule of thumb.  I typically use a single pcie power cable for each riser + gpu using a splitter.  I use server PSU's so all wiring is pcie cables.  Please ensure your loads are balanced and proper, not some BS like all risers on one cable.

My thoughts of why the crash...
1. If you didn't turn down the tdp in afterburner you might be causing issues on the PS.  Did you split the loads across each 12v rail?
2. Virtual memory/page file size does help and is necessary
3. If the cpu is a terd it will certainly choke throughput, but shouldn't crash
4. Possible bad pcie power plug.  I had a burnt plug on one of my 13 gpu rigs.  Slightly burnt and melted causing all kinds of havoc.

I suspect tdp + page file combination causing issues.  If your power wiring is jacked, another real possibility.

Are your pcie x1 cards touching each other at the mb? Slide a piece of paper between them for a test.

A photo of the rig wiring  would help.

I would highly suggest you don't try and run all your risers and GPU's on one rail...I must be misunderstanding the post above? I dont know about the 1070's yet, but my 1080's use 180 watts each. Putting just two of them on one rail would be overloaded.

I agree that ethos is also a good alternative to Windows, and probably much easier to set up than nvOC. When I used to mine ETH, and then ZEC (which is obviously equihash) - ethos was always very stable and as @rem26 said - it "just works". It's pretty easy to set up but I haven't used it in quite a while because I swapped from AMD R9 290s to GTX 1060s, 1070s and (unfortunately) 1080s. Wish I had bought 1080ti's instead, but the 1080s are good at neoscrypt at least... I might even spin up one of my old ethos SSDs and see how they are with Nvidia GPUs nowadays.

If he has 1200W - what is the problem with one rail? I haven't ever bothered with multi rails and I have no problem running 4 x 1070 and 1 x 1080 with 2 x 750W PSUs. Using dual PSUs is much more complicated for a beginner though.

Card 1 riser - rail 1, card 1 gpu - rail 1

NOT Card 1 riser - rail 1, card 1 gpu - rail 3 or some other combo.

Each 12v rail has regulation.  Each card is pulling on the rail.  Don't be a dumb ass and put 4 cards on rail 1 and 4 risers on rail 2.  Keep them together if possible.   Overloaded rails cause regulation issues is my point.

GPU's pull a lot of power when mining begins.  It can cause the 12v rail to pull down then pump up...imagine a ball bouncing. This is very true when the tdp hasn't been turned down.

You do realize some PSU's have multiple 12v rails right?  A single PSU can have 2-4 12v rails.  I said rails, not PSU.  Please read.  I am not a mind reader and have no clue what PSU model he has.  I made the statement just in case.  It needs to be looked at. 

Yes, that was my point. Pulling everything through one rail is a bad idea. It would work fine with two or three 1070's but I would spread the amperage across multiple rails when possible. I guess I misunderstood your post.

OP, do you have corsair PSU HX1200?  Flip the damn switch from multiple rails to single rails as a test.  See how the rails work?  This particular bad ass PSU can switch between the 2.  I would choose multiple rails with proper loading.  As a test flip the switch if you have this PSU to prove the wiring load is unbalanced on multiple rails.  PS it come default to multiple

Corsair HX1200 DC Output Ratings in Single +12V Rail Mode
DC output   +3.3V   +5V      +12V   -12V   +5VSB
Max Load   30A            30A.  100A   0.8A   3.5A
Maximum Combined Wattage   150W   1200W   9.6W   17.5W
Total Power: 1200W

Corsair HX1200 DC Output Ratings in Multiple +12V Rail Mode
DC Output   +3.3V   +5V   +12V1   +12V2   +12V3   +12V4   +12V5   +12V6   +12V7   +12V8   -12V   +5VSB
Max Load   30A         30A   40A            40A   40A.     40A    40A    40A    40A    40A    0.8A   3A
Maximum Combined   150W   1200W   9.6W   15W
Total Power   1200W

Correct, 2 cards max at 1070 or ti per rail.  1 x 1070ti = 200 to 230 watts at 100% with peaks to 290 watt for less than 50ms.  2 of these per rail max, personally I would do 1 per rail.  16 awg cables = 300 watts per cable in my opinion, or 1 card.

OK that makes a lot of sense now that I understand.

I don't really have to worry about this stuff personally, because I use dual PSUs on both my rigs, and they are both using 1 x 850W ATX PSU for mobo, risers and 1st 2 GPUS with all the extra GPU PCIe 12V connections from a modded 1200W server PSU. 4 x 1060 6GB, 4 x 1070, 2 x 1080 on 2 mobos, etc all pulling around 1400-1500W @ 240vac mining with everything at powerlimit 85% (currently mining Bulwark - NIST5 algo).

I think the OP must be at factory clocks, powerlimit, etc - so yes I agree that balancing outputs over 12V rails would be a good idea.

I really want to know if changing the pagefile size helps, but I can't remember if equihash uses a lot of RAM or if the equihash mining software advises using a big pagefile - I suggested that because I remember when I swapped from Linux back to Windows (to get easier to use ccminer releases) the first thing I had to do was give Windows a huge pagefile - but I can't remember what I was mining at the time.

Thanks for all your help:

My virtual memory is already set to 50000 so thats not the issue.

I have ordered some molex powered risers to see if potentially thats the issue.

I have tried using a linux based OS but still get shit combined hashrates.

I am running the cards at stock settings so its not that

I am using a corsiar hx1200 powersupply, i have flicked it to single rail and will report the results. The other information is a bit confusing as i dont know how to manually assign things to specific rails in my psu.

I have tried some testing and it seems my rig just will not go above roughly 750 m/s if i plug 2 cards in, first one will hash at 450 and then next at 300, then if i plug all 4 in they will do some some random arrangement of hash rates to equal roughly 750. Has anyone ever heard of a problem like this?

When my rig just crashed then i got the error Video TDR Failure, after some googling it states the error is driver related, yet im running the latest drivers on all my gpus...

Not having enough power for the risers can cause alot of random crashes and errors. From what you have said so far, I am betting it is the risers and lack of power.

If you start again from very beginning.

2 cards on mobo, does it mine well? If it mining well, then perhaps you can take another riser and slot in the third card and see how it goes.

Start everything with stock performance first.

IMHO, sounds like a power issue to me. when gpu cards dont get enough power it will run slower. I see this a lot when I undervolt my cards. some cards respond better to undervolting and can clock at a higher speed, some just default to a lower speed because it doesnt get enough power. probably silicon variation.

Nic, did you ever get this sorted out? If not, let me know as I might have another trick or two you can try. Let me know if all of the cards show up in device manager, and if they all say working properly when you view the properties of each card.