13 GPU rig unstable - any suggestions?

danielo89

Commendable
Oct 15, 2016
34
0
1,530
Hello.
I am experimenting with the ASrock H110 mobo which has 13 PCIe slots.
I have connected 13 gtx 1070s to it using risers. All cards are detected in SimpleMining OS and it starts to mine, but usually in less then 1 hour it freezes and I have to reboot it manually with the power switch.
I have tried lots of different things to solve it but the problem still persists. In the mining console there is no error message or anything prior to it freezing.
Anyone with experience from this mobo? What might the problem be?

One curious thing that I have noticed is that 2 of the cards does not get applied the OC settings, which makes me suspect some problem with the power. Maybe those two cards dont get enought power? I use 2 of those pci splitters who splits PCI 6 pin into 2x 8 pin. I know that is blasphemy around here to use such adapters but I have had good experience with these so far. I connect 1 of each those 8 pins to a different card, i.e 1 cable from the PSU supplies 2 cards, and they are set to draw about 105-110 W per card. Is this too much for one of those PCI cables? Or might the problem be that the cards draw asymmetrical amount of power and that causes some kind of error in the PSU?
I use 2 x 850W PSUs to power the rig so I have a total of 1700W and the rig draws about 1550-1600W when i use powerlimit 105-110W so it should not be a problem with the total power draw...

Anyways, thoughts on what the problem might be?
 
Solution
To avoid a giant multi quote:

1. You need a new PSU for those GPUs, I wouldn't go any lower than 1000-1200w to ensure you have both enough connections and that your power needs are met.

2. It depends on the PSU and how much power the rails in it put out the the PCIe connectors. That said as I mentioned that guy didn't account for resistance in the connectors. I would not over draw each connector too far from 150w, the risk is that the connectors will melt (or you will damage the PSU).

3. There IS headroom to over draw a cable some what, but only so much, you clearly exceeded it. They do have failsafe systems, somehow you got lucky with your EVGA BQ. EVGA does sell some good PSUs, the GS, PS, G2, P2, T2 lines are all excellent...

maruniverse7

Notable
Jan 6, 2018
266
0
860


oh my.. Anyways what im seeing is that you are probably using adapters to connect each GPU but one? And with all those the motherboard don't support them all because you would have to connect them all with SLI bridges
 

maruniverse7

Notable
Jan 6, 2018
266
0
860


To make sure you have full power delievery, remove some cards and slowly add them back till you reach your same problem.
 

danielo89

Commendable
Oct 15, 2016
34
0
1,530


Yes that might be a good idea. At least trying it without the PCI splitters might be the first step
 

Rogue Leader

It's a trap!
Moderator


So you're splitting 1 PCIe 6 pin to 2 8pin for multiple cards. Considering the 6 pin is rated for 75w, and 8pins rated for 150, even though you aren't drawing that kind of power, you may be running into amperage issues depending on the PSUs you are running. What brand and model are they?
 

danielo89

Commendable
Oct 15, 2016
34
0
1,530
1 x Corsair RM 850x
1 x EVGA GQ Hybrid 850W
Both are Gold-rated. (yes I know....another blasphemy using 2 different types of PSUs)

I have a hard time believing the 6-pin is only 75 W. Besides, it is actually a 8-pin cable, but the adapter can only make use of 6 of the pins.

Yes I am splitting 2 PSU-cables into 4 x 8pin that I use in 4 different cards

Oh yes - another thing. The PCI cables are actually split themselves from 1 PSU socket into 2x 6+2 pin cables, before entering the adapter, like you can see in this picture http://www.corsair.com/~/media/corsair/product%20photos/psu/hx-series/hx850/large/detail_contents_hx850.png

So one PSU cable is actually supllying 3 cards (unless I use both adapters on the same PSU cable, which would mean 4 cards on 1 cable, I have to check that up)
 

Rogue Leader

It's a trap!
Moderator


You may have a hard time believing it, but thats what it is, the PCIe spec for a 6 pin connector is 75w. Splitting it to 2 8 pin connectors, essentially allowing the card to draw up to 300w through it is a bad idea. That said the Corsair is a very good quality PSU, the EVGA GQ not so much. But I like maruniverse's testing idea, I would start with the Corsair PSU, and test adding cards till it craps out. I think theres a chance that EVGA can't handle the amperage draw.
 

danielo89

Commendable
Oct 15, 2016
34
0
1,530


But the cards are set to draw 105 W so it´s far from 300W anyway. And the cable is still actually a 8-pin cable doesnt that count even if I use only 6 pins?
And actually the Corsair is not so good quality. Several of the SATA slots have melted when using 3 risers on one cable. That has happened on a lot of Corsairs PSU even the Platinum rated 1200W one. So i am never buying that Brand ever again

I am not using the splitters on the EVGA PSU. I use that for the mobo and CPU and risers, and most of the GPUs (like 8 or 9 of them) are supplied by Corsair. I have used a Watt-meter to calibrate the power draw so that it is split evenly (and below 850W) between the PSUs
 

danielo89

Commendable
Oct 15, 2016
34
0
1,530


According to this post the modern PCIe power cables can handle much more than 75/150W, i.e 192/396W for 6/8-pin, respectively.

http://www.tomshardware.co.uk/forum/274631-28-power-spec-power-plug#2674391
 

Rogue Leader

It's a trap!
Moderator


So you are basing its quality on overdrawing the crap out of a cable. And if you did that isn't it conceivable that you damaged that PSU? Either way certain Corsair PSUs (the RMx family in particular) are very good quality while the EVGA GQ series has been tested to be poor quality.

That 6 pin connector is rated for 75w, yes of course in theory more can be drawn out, but a card won't do that (it needs to stay within specifications), in your case splitting it to 2 8 pins, and running 2 gpus off that, with your settings I'd guess you are drawing around 180w per 6 pin connector (remember the PCIe slot provides per spec about 75w of power, but good GPUs are below that and offload the power demands to the connector, I was generous and gave you 65w from the slot). So you're overdrawing the connector.

Your PSU has 3 PCIE connectors with each cable providing 2 6+2 connections. So its is expecting to provide a max of 300w per cable, per spec. You are running 360w per cable. That is well over spec. Then multiply it by 3 youre at 1080w, that PSU peaks out if I had to guess around 930w (like in terms of over current shut down). Now most likely you're using a bit less wattage here else it would be shutting down immediately all the time. But you are still driving the hell out of that PSU.

People running 13 GPU mining rigs are running 2000w PSUs, this whole system is severely underpowered.



You're citing an 8 year old post from some random guy who did the math based on the theory of what a wire should be able to handle. All while missing a number of factors such as resistance, in his mathematical equation. Anyone dumb enough to try and draw 400w out of a single 8 pin connector is higher than a kite. Not only that a PSU would never provide that much power through a single PCIe port.

This is easy to prove as you can look at any test of the initial run of AMD RX 480 GPUs which only had a 6 pin connector which severely hampered its overclocking, as it could not draw enough power.
 

danielo89

Commendable
Oct 15, 2016
34
0
1,530
I have now tested without the pci-splitters. Same problem still.

So you are basing its quality on overdrawing the crap out of a cable. And if you did that isn't it conceivable that you damaged that PSU? Either way certain Corsair PSUs (the RMx family in particular) are very good quality while the EVGA GQ series has been tested to be poor quality.
I think this might be part of the problem, yes


You're citing an 8 year old post from some random guy who did the math based on the theory of what a wire should be able to handle. All while missing a number of factors such as resistance, in his mathematical equation. Anyone dumb enough to try and draw 400w out of a single 8 pin connector is higher than a kite. Not only that a PSU would never provide that much power through a single PCIe port.
So how many watts can I actually get from a 6/8-pin in the real world? Obviosuly it´s not 75/150W either


So you are basing its quality on overdrawing the crap out of a cable. And if you did that isn't it conceivable that you damaged that PSU? Either way certain Corsair PSUs (the RMx family in particular) are very good quality while the EVGA GQ series has been tested to be poor quality.
Well yes I base my quality evaluation on this.
The reason is. There should be some headroom to experiment with different cables and e.g by mistake draw too much power from an outlet; the PSU should have fail safe systems built into it so that it shuts down - and not burning up. Obviously even the "bad quality" EVGA GQ has this sort of wiring in it since none of them have burnt or meltet. But almost all of my Corsair - even the platinum ones - have done that. So yes, in my experience and for my use, EVGA is better quality than Corsair.
 

Rogue Leader

It's a trap!
Moderator
To avoid a giant multi quote:

1. You need a new PSU for those GPUs, I wouldn't go any lower than 1000-1200w to ensure you have both enough connections and that your power needs are met.

2. It depends on the PSU and how much power the rails in it put out the the PCIe connectors. That said as I mentioned that guy didn't account for resistance in the connectors. I would not over draw each connector too far from 150w, the risk is that the connectors will melt (or you will damage the PSU).

3. There IS headroom to over draw a cable some what, but only so much, you clearly exceeded it. They do have failsafe systems, somehow you got lucky with your EVGA BQ. EVGA does sell some good PSUs, the GS, PS, G2, P2, T2 lines are all excellent quality. The new G3 and B3 have had some quality issues, and the GQ and BQ are their lower end stuff. Neither Corsair nor EVGA manufacture their own PSUs, both use various OEMs such as Superflower, Seasonic, CWT, FSP, to build them. CWT is generally cheap stuff (thats your EVGA GQ).

If I were you I'd ditch both brands and Buy this

https://seasonic.com/prime-1300-w-gold#cables

As you can see it has 12 6+2 connectors, Very safe for your situation, the best in the business.
 
Solution

danielo89

Commendable
Oct 15, 2016
34
0
1,530


A new? Like in one? Cause that would surely be too little power for a 13 gpu rig. It draws like 15-1600W
I suppose you mean 2 x 1200 W? 2400W? That seems way too high. I am thinking maybe 1000W + 850W would be nice. Good and powerful PSUs are really expensive so I don´t want to buy more stuff than I need.

That Seasonic 1300W seems nice but is not available in my country it appears (scandinavia)
 

Rogue Leader

It's a trap!
Moderator


No I didn't say 1 gpu for the whole rig. That EVGA GQ you have while I'm not a fan works for you, so keep it (unless you still have issues). I'm surprised that you don't have Seasonic they are available everywhere. So can you get this:

https://www.evga.com/products/product.aspx?pn=120-G2-1300-XR

This will give you 12 GPU connectors, then run 1 off your other PSU and you are set.
 

danielo89

Commendable
Oct 15, 2016
34
0
1,530


Yea That 1300 G2 is nice. I already have a couple of those and they are the best PSUs I have ever owned. Such solid build and stable run.
Btw they have 8 GPU-connectors not 12.

Unfortunately they are sold out until march everywhere (suprise...)

Or did you perhaps mean this one https://www.komplett.no/product/839803/datautstyr/pc-komponenter/stroemforsyning/enheter/evga-supernova-1600-g2-1600w-psu

It has 10 PCIe connectors
 

Rogue Leader

It's a trap!
Moderator


The 1300 should have 6 connectors on the PSU, with 2 8 pins on each connector giving you 12 8 pin connectors.

If not then yeah bump up to the 1600
 

danielo89

Commendable
Oct 15, 2016
34
0
1,530


I own like 2 or 3 of them and running 8-cards rigs on them so I can ASSURE you that they have 8 GPU connectors. Anything else and I would of course have had more than 8 cards on them (knowing me by now...)
Besides, as you can see in the link I gave the 1600W has 10 PCIe connectors, so it is obvious that the 1300W can´t have 12.

It is true that it has 6 VGA outlets on the back side but there are 4 single cables and 2 split cables - giving a total of 8 PCIe connectors in the other end. The first versions actually had the 2 split cables as 8 + 6 pin which was terrible. Nowadays you get only 8-pin ports.
I only have the old version with two 6-pins so I have to use adapters to get 8-pins to the card. Another blasphemy I know. But it has worked fine for over 6 months now

Edit: You can see the specs here that it has 8 connectors https://www.evga.com/products/product.aspx?pn=120-g2-1300-xr
There it still says two 6-pin ports but I am sure the last one I saw at a friend´s had purely 8-pin connectors (as we connected 8 cards and we didn´t use any adapters)
 

danielo89

Commendable
Oct 15, 2016
34
0
1,530



I pick this as the solution as I will have to get at least one of the PSUs replaced by a more powerful one, so that I get enough native 8-pin PCIe ports and don´t have to use any adapters - as well as being able to split the power bill between the two PSUs without overcharging either of them (which I have to do now as they have 6 GPU connectors each but one also has to carry to load of Risers and MOBO+CPU (the riser load can be shared but the Mobo+CPU can´t))

I think I will get an 1000W PSU as all the 1200W and above are either impossible to get right now or insanely overpriced. I have measured the wattage for 6 gtx 1070s + mobo/cpu and it is below 1000.

(actually i have ran an 8 gtx 1070s-rig on less than 1000W before)
 

AnonymousAndy

Reputable
Dec 18, 2014
361
0
4,960
For future reference, the 6pin / 8pin CABLE is not rated for 75w / 150w. The 6pin connection is mandated on the GPU / device side to not PULL more than 75w for the six pin or 150w for the 8pin. So your GPU will not pull more than 75w via a six pin connection. This ATX spec limitation is NOT enforced in anyway by your PSU. The six pin cable itself can support over 200w of current. The 8pin can support over 300w. So if your single six pin cable that splits into dual 8 pins ever drew max power draw from the GPUs, it would overload the actual current limitation of the cable, which is over 200w. That shouldn't happen, as you have your GPU's undervolted so much, but software settings don't always work like you've already experienced with your OC settings. That is not safe, I would never run my systems like that.

I'm glad you're getting new PSUs :p [strike]But I don't think it's gonna fix your stability problem. [/strike]

 

danielo89

Commendable
Oct 15, 2016
34
0
1,530


Thanks for the info!
But why don´t you think it will fix the stability problem? Obviously it is most likely caused by power insuffiency
(although it might also be caused by the burnt sata-ports which has damaged the entire PSU in some way)
 

AnonymousAndy

Reputable
Dec 18, 2014
361
0
4,960


Sorry, I only skimmed the first half when I decided to respond. I didn't get to the part where you said exactly which PSU's you had. Yea, you totally don't have enough power lol. Amperage is the problem not wattage. Both of your PSUs supply 70a on the 12v rail... Lets assume you're pulling 120w per GPU to be on the safe side, (to save us from calculating for heat/resistance/etc) that equates to 11amps per GPU. That's 143amps out of your available 140, not even including your CPU/mobo.

Further, do you have both of these PSUs connected to the same outlet? I'm assuming it's a 20a, but is it on its own circuit or are there other devices running on the circuit? That would further reduce the quality of power. Even though your PSUs are only supplying 1500-1600w of power, they're pulling closer to 1800w+ from the wall due to efficiency loss. That's a lot of power for a standard household outlet, and coming close to the maximum of the 20a circuit.

And what's this about burnt SATA ports? I missed that entirely.
 

danielo89

Commendable
Oct 15, 2016
34
0
1,530


Some of the SATA-ports on the Corsair PSU have melted when I used it for 3 x risers. Do you think it affects only that very port or the enitre PSU?

I have a wattmeter directly in the wall socket so I know it is only pulling 14-1500W from the wall.
We have 16A breakers in the house, which would amount to 3680 watts right (on 220V outlets)? Only a router is on the same circuit as the rig.
The main breaker is 60A, so that´s certainly not a problem.
Yea I will get more powerful PSUs. Atleast for the main one who supplies mob0+cpu aswell as gpus
 

AnonymousAndy

Reputable
Dec 18, 2014
361
0
4,960


Oh, you're not in north america... nvm about the outlets, we're limited pretty heavy over here.
Depending on where the SATA connections melted, I doubt you damaged the PSU beyond the physical plastic connection on the PSU if thats the part that melted. I'm sure you've come to this conclusion already, but don't connect more than two risers per SATA cable. :p
 

Rogue Leader

It's a trap!
Moderator


This sounds like a good plan, and I do agree since you are mining and not gaming, and controlling the power usage of them you can get away with somewhat less wattage, in fact I think my back of the napkin calculations above came out around 1000w. I mean I would prefer 1200w, but a good 1000 will do it.



I did say that, and it was also illustrated in that linked post (some of the good info that was also otherwise bad). The limitation for them is the connectors and the rails in the PSU. The standard is of course the draw of the connectors, and the PSU is built around the expectation that the attached accessories will not overdraw those connectors. Which is yet another reason we hate adaptors on this site. If it was meant to work with that type/number of ports, the manufacturer would have included such ports on the PSU.
 

TRENDING THREADS