Topic: Is it possible to get correct information on how to power a rig properly? (Read 224 times)

As said above, everybody is saying something different. I have my mobo, risers and 1 GPU on my 650 PSU. My other 2 GPUs are on my second 1200 PSU.

Anyone agree with this? I split half miners (gpus + risers) in one psu... And now i'm scared

Another problem is to split the 6-pin using Y 6-pin cable, i burned (literally) one yesterday.. any suggestion? 

Why would i need more power for 2 1060s?

And can we please not talk about risers and psus and all that ... even in this thread where it was specifically asked NOT to talk about it there are differing opinions. I made my choice!

 Don't get me started on how many "engineer mistakes" I've found and had to help fix in the years I worked as an engineering technician.
 It's bloody AMAZING the issues they don't think about on how stuff works in the real world.

 The point on "riser and MB need to be on the same PS" is usually valid on the old ribbon-type risers where there often IS a +12VDC connection between the MB and the riser, but it DOES NOT APPLY to USB-type risers where there is ZERO power connection between MB and riser.

 Also, the BIGGEST issue with using SATA connectors to power a riser is THEY ARE NOT RATED FOR THE CURRENT DRAW.
 They are only RATED for 54 watts of +12VDC - the PCI-E bus is rated for 75 watts draw, and THE RISER ITSELF ALSO HAS CONVERSION LOSSES.
 Even if the WIRING is heavy duty enough to support the draw, the CONNECTOR is not.

 USB-cable type powered risers have ZERO power connection to the MB, though they do have a ground connection in common probably for signal shielding reasons.
 The only time it matters on having the MB and the Risers on the same PS is if you are using the old-type RIBBON risers, where there MIGHT be a +12VDC link involved.

 On the other hand, while the GPU itself PROBABLY doesn't "merge" the +12VDC input from the PCI-E bus and the PCI-E power connectors, it MIGHT do so - keeping the card and the riser on the same PS is less likely to cause issues than worrying about the riser and the MB being on the same PS.

Buy better psu and replace this 450 w something much more power chieftec.

I know you said that you have "made your choice" etc, but I was once told by an electronic engineer how I should wire my rigs and he said that when using powered risers (like the typical USB riser with a molex or 6-pin 12V input) you MUST power ALL risers with the same PSU that powers the motherboard, and only use the 6-pin PCIe power from the 2nd PSU.

I know it usually works to do it the wrong way and many many people are fine for months or even years, but as @priviet  said - different PSUs are never exactly the same in their 12V output but their circuitry is attempting to "smooth" or "stabilize" their output and this gets more complicated the bigger the difference between them. So... if they both sit at say 12.2V everyone is happy, but if one of them has a problem, or is overloaded by all GPUs needing max power at once or something and it droops to 11V - you could get a lot of heat or popped capacitors.

I work as a technician with electronics (I'm definitely not an engineer) and one thing I learned a long time ago is that problems with power account for the majority of faults or problems with electronics. It's a good idea to get the power right and to give your system more power than it needs if you can afford it. I will repeat this because I know this is an issue that has caused a lot of arguments - it usually works to do it the wrong way. Usually.

Not sure... I've never wanted to run it that slow for long enough to find out! heh. XMR algo kills the driver instantly even at stock :/

Yes Hynix does sick for over clock. A few are Micron that are mostly stable at 500.. so i just run them all at 400 . I am a very unlucky guy with pretty much everything i buy. I have only ever gotten one card with Samsung and even that cant go over 700.. or even stay at 700 reliably. heh

Your OC is stable for some algos and not others.
What happens when you leave the settings on default (0 core and mem).
Are you still losing a card?
If so you will need to lower your OC.
Also Hynix memory isnt the best at overclocking.

Power limits ...

Every so often (mostly 1 maybe 2 times a week) when i go check on the rig we are talking about, one card will be down. Not always the same card. When i was still using afterburner it was connection lost to driver. I have to restart the driver with Nvidia inspector (I use this to OC now) or reboot to get it back. It's a real pain in the butt since when it does happen.. it almost ALWAYS happens right after i fall asleep! lol. So i miss 8 hours of mining on one card when it happens.

Pretty modest OC.. core +100 Mem +400 (Most cards have Hynix). I had the power at 80 but have been trying it and 86.. Seemed to last a few days longer but i just check and a card has been down for a few hours. Who knows!

Any ideas? OC, Bum risers, USB Cables? I cant be certain but i think it mostly happens when multipool miner switches algos.

^ you're on the right track here.
BUT you need to calculate power draw based on what is being pulled at the wall (use a kill-a-watt meter to determine the power draw based on your own system and not a news article). What are your power limit settings in afterburner?
Try to get good quality splitters if you can and  connect your PSU's with an add2psu if you can.

Heh

Im gonna rewire it anyways

TX750M comes with

1 x ATX (motherboard)
1 x EPS (CPU)
1 x floppy connector (useless)
7 x molex (on 2 rails)
4 x PCIe (on 2 rails)
8 x SATA (on 2 rails)

400W EVGA has:

2 x PCIe
2 x molex

TX750M will be the "master" PSU and the EVGA the "slave" PSU. All the basic (non GPU/riser) components will be hooked up to the master:

1060's are powered by a single 6-pin connector. They draw about 61W along the PCIe and 60W along the riser (http://www.tomshardware.com/reviews/nvidia-geforce-gtx-1060-pascal,4679-6.html). This means im capable of powering 2 cards or 2 risers from a single (8-pin) PCIe power connector (max 150W per 8-pin PCIe).

So using the TX750M, split PCIe #1 to power card #1 and #2. Split PCIe #2 to power riser #1 and #2.

Then split PCIe #3 to power card #3 and #4. Split PCIe #4 to power riser #3 and #4.

So all total, mobo/CPU/SSD, 4 cards and their corresponding risers are powered via the TX750M. At full power, the cards will draw 4 x 121W, plus the other components (~85W ish) = 570W. Fantastic, because a PSU runs optimally at 80% load (750W * 0.80 = 600W) which you are very close to.

Now that leaves the EVGA 400W.

Using the EVGA 400W,  split PCIe #1 to power card #5 and #6. You then split PCIe #2 to power riser #5 and #6.

So all total, that PSU is drawing 2 x 121W = 242W.

Good to go?

No it is not possible.
This is the internet, the font of all knowledge AND bulls&*t at the same time.

There is no ABSOLUTE safety, but if it hasn't burned down the apartment yet, it probably is hooked up correctly and probably wont.

And that is a firm an answer as you can expect.

Baz

When you use multiple power supplys then you have current from one 12V rail to another because the rails are not exact the same voltage on both psu. If you power riser and GPU from different PSU then in the worst case this generates heat at the wrong place because of the differential voltage and the short-circuit current. You can measure this with a multimeter connected to the different 12V rails.

This multi PSU setups could be safe, they are probably safe if properly distributed.

SATA to PCIE 6 Pin adapter have another problem. Fake chinese wires and no proper clamping in the SATA connector... They claim to be AWG18 but they are thin as f... I only use PCIE to 4 Pin Molex and test one adapter of each bulk order by destroying it and checking the diameter of the wire.

I generally prefer, on multi-PS setups, to power the riser and the GPU power connections from the same power supply.
I'm PRETTY SURE that the cards themselves don't "merge" the inputs though, so it should be safe to power them from different supplies - just more likely for the card to get "confused" if one power supply starts up faster than the other and not be sure if it HAS proper power.

I don't think it is. It is beyond frustrating lol. Every single person you ask, every single video you watch, every single forum post you read, everybody says something different.

Say i had a few 1060 rigs set up similarly. One example .. 6 GPUS and 2 power supplys. One PSU is a Corsair TX750M (2017 version) and the other is a 400w EVGA I had laying around. Yes i am planning on replacing it. I pay the equivalent of 6 cents USD per k/h so i haven't really cared.

Corsair TX750M is powering the MB/CPU/HDD

Corsair TX750M has 2 PCIE cables with a total of 4 6/8 pins - Each lead powering 2 cards

Corsair TX750M has 2 molex leads - Each lead powering 2 molex risers

400w has 1 PCIE cable with a total of 2 6/8 pins - Powering 2 cards

400w has 1 molex lead - powering 2 molex risers

Now, at the time pretty much everything most of what i read and every person i asked said this was ok. The more i read now, the more paranoid i get that i'm gonna burn down my apartment when i'm not here. So, is it safe, or not? I mean i have seem people powering 6 1060 with one 750w. Using 2 SATA (using SATA to 6pin adapters) risers per chain!

If not, since i am buying a new PSU to replace the EVGA, what would be the ideal way to do this without spending a lot? ELI5 even. lol

P.S. I don't want to get into powering the GPU risers with the same PSU as the motherboard or not. Been there done that every body has a different opinion. I made my choice.