Topic: PCI Riser Power Test, Results, and Recommendations (Read 538 times)

Actually they should, PCIE requires a 3.3v pin by standard but almost no videocard uses it.

Test Results are listed below:

SATA                3.3v*                        5v*                        12v               
GPU Pwr    Amps    Watts      Amps    Watts     Amps    Watts      Neutral Current   Efficiency (%)       Core Clock    Memory Clock
75.00%     0.01     0.033       0.015     0.075    3.88     46.56           -2.5                 64.0%                1930             4050
100.00%    0.015   0.0495      0.006     0.03     4.05     48.6            -3.39                83.3%                2100             4050
75.00%     0.011   0.0363      0.005     0.025    3.76     45.12           -2.59                68.6%               no-OC           no-OC
100.00%    0.01     0.033        0           0         3.88     46.56           -3.5                 90.0%               no-OC           no-OC
* Appears no direct load on the 3.3v and 5v rails, this is likely cable loss only.
                              
Molex                  5v*                    12v                     
GPU Pwr      Amps   Watts   Amps   Watts   Neutral Current   Efficiency (%)     Core Clock   Memory Clock      
80.00%       0.003   0.015   4.73    56.76        -2.98                 63.0%               1860         4050      
100.00%       0         0       4.85    58.2          -3.8                 78.4%                 2025         4050      
80.00%         0         0       4.55    54.6          -2.95                 64.8%               no-OC       no-OC      
100.00%       0         0       4.65    55.8          -3.82                 82.2%               no-OC       no-OC      
* Appears no direct load on the 5v rails, this is likely cable loss only.
                              
PCI                     12v                           
GPU Pwr     Amps    Watts   Neutral Current   Efficiency (%)    Core Clock    Memory Clock            
75.00%       4.62    55.44          -4                   86.6%            1800                4050            
100.00%     4.8      57.6          -5.3                  90.6%             2035                4050            
75.00%       4.44    53.28          -4                   90.1%            no-OC              no-OC            
100.00%     4.56     54.72        -5.3                  86.0%            no-OC              no-OC            
                              
GPU Pwr Limit    Total Rig Power    Other                        
      75%                  465               w/ OC                        
     100%                 610               w/ OC                        
      75%                 465                no-OC                        
     100%                 595               no-OC

Looking at these results, all power cable types draw nearly the same current and power as they all supplied by 12v.  SATA appears to have the lowest power draw (unsure why).  Note that i noticed differences in riser power when mining different algorithms.  However, this was within probably +/- 10% of other algorithms.  GPU power limits had the largest effect on riser power draw.

When i switched to Molex, i had to up my power limit to 80% otherwise one of my GPUs would throw errors.  I believe this was my GTX 1070 mini-ATX which is more likely the reason for that anomaly.

One note is riser power increases with higher GPU power limits and only slightly higher with overclocking.  My power cables are 18 AWG rated at 80C which are rated for 15 amps in short lengths.  However, you need to consider the ratings of the connectors and their pins!  Below are connector ratings for SATA, Molex, and PCI.

SATA connector rating = 4.5 amps (54 watts total for 12v pins)
Molex peripheral connector rating = 5 amps  (60 watts total for 12v pins)
PCI connector rating = 6.25 amps (75 watts total, only 12v pins)

Source: https://www.moddiy.com/pages/Power-Supply-Connectors-%26-Pinouts.html

Based on this test, you shouldn't daisy-chain more than two risers, especially if you are running high power graphics cards (i.e. 1080 ti) as you may exceed the amp rating of the connector pins (or the cable).  However, if you are running your GPUs at reduced power limits using nvOC for instance, than you may be able to get away with powering two risers with one cable.  However, in general I would not recommend it, as typically everyone would have the ability to run an extra molex peripheral cable or PCI cable from their power supplies.  

Hope this was helpful and prevents some burnt connectors

1 amp at 3.3 volts would be 3.3 watts - but even at 80% conversion efficiency that would be 0.66 watts of conversion loss, not enough to worry about.

dude, you need to do some reading on how PCIe works.

power comes from 12v ONLY!

3.3v is only used for signal sending! you are not pulling any large amount of current from 5v or 3.3v lines.

Agree with QuintLeo. And just to add...

I have used some cheap-ish "pure Molex" risers which were making 3.3V from that 5V rail coming from "the Molex". 3.3V power line should not see much load if any. I measured it on a couple different GPUs and it was well under 1A, which is less than 0.3W.

Those cheap step-down LDO voltage regulators they put on risers still have decent efficiency at low currents. Say, 80-90% if not more. Do the math and get 0.3W - 85% = 45 mW of power loss on the riser!

I'm not so much "die hard anti-sata" as I am "insistent on pointing out that it CAN be a risk" since the SATA power connector is NOT rated to handle the max power that can be drawn from the PCI-E bus and still be in spec on that bus.

I recognize that a lot of cards don't pull the full 75 watts (plus voltage conversion inefficiency) the PCI-E bus allows - but unless you can actually measure the power draw of YOUR cards at the settings YOU use, there is no way to be sure that you are staying within the 54 watt max power draw rating for the 12VDC connections on a SATA power connector.

Daisy chaining - depends on the specific power supply and cables, but most CABLES for SATA use are not designed to handle 100+ watts that 2 risers commonly pull, much less the over 150 watts they COULD possibly pull.
I've seen SOME SATA chains that don't use heavy enough wiring to handle 75 watts - and rare cases of MOLEX chains that have that light of wiring as well.

None of the risers I have examined use 5V input at all, much less 3.3V input - ALL of their power comes from the 12VDC lines - but I don't have any "pure SATA" or "pure MOLEX" risers to check, those in theory COULD be different.

Note that the PCI-E bus itself only supplies +12VDC and +3.3 VDC - if a SATA-based PCI-E riser used the 3.3v feed from the SATA connector, it wouldn't need any power conversion AT ALL and would only have probably 2 electrolytic caps on it AT MOST - one for 12VDC one for 3.3 VDC.
3.3 VDC on the PCI-E bus is also stated as "standby power", indicating that the load would be very small when it is used at all and that the power conversion circuitry on a riser probably doesn't eat even a watt.

You have no idea what you’re talking about lol. GPUs can pull up to 75W from the slot. That’s the spec. You think any of that comes from 5 or 3.3v??? Think again.

Just because there’s a 3.3v wire on the SATA connector, doesn’t mean it’s used or even to connected to anything on the riser PCB. I haven’t found any devices that use 3.3v from the SATA connector

You people need to relax, why is everyone so quick to jump to conclusions.  So maybe i left out some details.  That is correct both SATA and molex connectors utilize multiple voltages but where do you think the 3.3v and 5v are coming from.....lets think....maybe its the 12v rail right??? No its the 5v or 3.3v rails!!  Dammit people use your head...

Not factoring in the load demand of the P/S originally will change things based on how many cards are plugged in, I'm going to measure current and voltage directly with my Arduino board to get actual riser power and run the cards with SATA, Molex, and PCI power cables at different GPU power levels.

This is what i plan to do next to put this to bed for good.

Nope, didn't try it, been using 6 pin risers ever since it came out and before it I was mostly using molex

Have you tried it?
But still, we need the different object, a few characters of cards. With one card only on test seems less accurate result. A different card may draw different power from PCIe slot or Risers.

If you want to check the riser efficiency .

Plug one GPU into the motherboard 16x and compare the power consumption with 6 GPUs on risers VS 5 GPUs on riser and one GPU on Mobo .

If you want to check the  Power draw from the riser by the GPU then the easiest way is to use Dual PSU setup with the 2nd PSU only powering the risers .

i use two risers in two Molex connection per cable. Since Risers can get up to 75W , and a molex cable can give max 150W, you are perfectly fine like this. This is my priority config.

I use one riser in one sata connection per cable. Sata uses like 54W per cable that's why only one riser per Sata cable. It can give you problems, specially if you are running a high power input GPU like RX Vega or 1080ti among others. This is the config when you need to add another card and have no molex to connect it.

If you do your research, you can see that 1060,1070 and all RX 470/480/570/580 and below, if well done (downclock and undervolt), will use less than 150W from the 8 pin connector so, even if your sata cannot provide 75W max for the riser, you should be fine.

if you want to mine Zcash, for example a Sata will give you problems with almost all cards specially those from AMD (since Nvidia cards are more efficient)

That's not how it works, your conclusions are totally wrong. SATA and Molex risers use both 5V and 12V rails, combo and 6-pin risers use only 12V rails. NVIDIA reports completely accurate power draw, which includes the slot draw. You are actually measuring the difference between AC (wall) and DC (cards) power draw, which is all about the PSU's (in)efficiency.

The proper way to accurately measure riser power draw is by using clamps or run your multimeter on power draw (A) mode and connect the leads in series.

If you want to measure your power consumption the right way, measure your current directly.

Get a cheap clamp meter capable of picking up DC currents like the one I have...

It's amazingly accurate for the price (less than 1% off), so I even bought a second one for the lab.
https://www.amazon.com/gp/product/B00O1Q2HOQ

-SCSI

Based on your statement:


Mean one SATA rail was powering one Risers is okay right? I've never tried to measure how much power would draw by GPUs from a Risers, but based on my experience one SATA rail powering one Risers seems safe.

I've got this kind of discussion before. About how to powering the Risers. You have to use more than one object for each part for a more accurate result, IMO.

A different card may be draw different power from the Risers, each card has different power management, and we have zero control on that. S different PSUs have a different specification in power producing.

Overall, I like what you do, keep going!


Same as you, I'm using SATA for more than one years without issue. One SATA rail / Risers seems fine for me.
PCIe cable (8,6+2,6) will more useful for powering GPUs.

your conclusion makes no sense.

1st, sata doesnt "run" off the 3.3v rail, and Molex doesnt "run" off the 12v rail.

MOLEX contrains both 12v and 5v power sources
SATA contains 12v, 5v, and 3.3v power sources (but the 3.3v is rarely used by anything, i doubt the risers even use it)

most risers with a 6-pin connector come with either molex or sata to 6-pin adapters. guess what, they ONLY use the 12v lines and have Vregs on the riser to drop it to lower voltages.

people heres are die hard anti-sata for some reason. yeah you can have some problems with low quality components, but daisy chaining and using SATA adapters isn't insta-death like some would have you believe.

my first rig has been running for MONTHS in the perfect storm of "bad" configs. 8 GPUs, V006c risers (6-pin power) using SATA->6-pin adapters. running 2 SATA cables from the PSU, with 4 risers on each line (from a very high end 1200w plat PSU with metal barrel connectors). it's fine *mic drop*

the biggest problem is getting enough 6-pin connectors in most cases.

You made a good point, the power supply losses will be different with different load, thanks for pointing that out.  But measuring mobo vs. riser would only determine the difference in power.  The goal is to find the total consumed power of each individual riser.  I think the only true way to do this is to measure the voltage and current directly to take the power supply loss out of the equation...perhaps i'll try and rig something up to measure that directly.

I dont see anywhere that you are calculating the inefficiency of the PSU. When looking at those numbers it looks more to be majority inefficiency of your PSUs as your 80+ platinum PSU should have an inefficiency of around 10%. If you really want to test the power pull you should be placing 1 unit directly on the Mobo, then test the power draw under load, then put it on a riser plugged into that same slot then test under the same load, or even under no load with no monitor plugged into the unit.

So i wanted to do a test to validate the power consumption of PSI risers and determine what is safe for powering them based on their consumption.  I've heard many claims that you cannot use SATA cables, Molex, etc. due to the high power consumption of these risers.  Some posts I found stated as high as 65 W of power which is around 5.5 amps at 12V supply.  This power draw seemed unusually high too me.  Below is the parts list applicable to the test.

- Gigabyte Z270p-D3 Motherboard
- Intel G4400
- 6 x GTX 1070
- Zijiez PCI risers (excellent quality, highly recommended)
- 2 x Seasonic Focus Platinum 750W
- Newforshop electricity power monitor
- Running nvOC v0019

The test was simple, I determined the power draw of my normal rig.  I have a separate power supply which provides power to 3 GPUs including the PCI risers.  I determined the steady state power usage of my rig with all cards running and removed one GPU and recorded the difference.  I also ran the GPU at a different power limit settings to see if that had any effect on riser consumption.  Below are the results.

Rig running 6 x GTX 1070s with 115W power limit = ~ 800 Watts @ 0.99 - 1 P.F.
Rig running 5 x GTX 1070s with 115W power limit = 670 Watts* @ 0.99 - 1 P.F.
* Additional PCI riser still powered

Rig running 6 x GTX 1070s with 115W power limit = ~ 800 Watts @ 0.99 - 1.0 P.F.
Rig running 6 x GTX 1070s with 125W power limit = 860 Watts @ 0.99 - 1.0 P.F.

It should be noted that I noticed no difference in power consumption with the additional PCI riser card being powered or non-powered with no GPU inserted.

Based on the above, the difference in wattage between my 5 and 6 card rig was 130 watts while using a 115 watt GPU power limit.  I verified the GPU power consumption on my power monitor and in the driver settings.  This equates to a PSI riser consumption of around 15 watts.  Since I was running these on the 12V rail this is a little over 1 amp.  I noticed when I increased the card power that riser consumption did not change.  Perhaps i can re-run the test at higher power limits to see the effects.  

Even though the test shows that the power consumption is fairly low, you need to confirm your power supply can handle the load demand based on what connector you are using.  In my opinion it is best to power risers from the 12 V rail with a PCI power cable or a molex cable (both run off the 12 v rail).  SATA is not really meant to power risers and i'm no even sure why they design risers with this option.  Since they use 3.3 V rail this increases the current to about 4-5 amps (15 watts / 3.3 V =  4.45 Amps) per SATA power cable.  Based on the specifications below, if you are powering 5 risers from the 3.3 V rail you will be exceeding the rail power limits (4.45 x 5 = 22.7 Amps) which can damage the power supply or catch it on fire.  So in conclusion, power your risers with INDIVIDUAL (not daisy-chained) PCI power cables or molex cables and you should be okay.  PCI and molex cables both run off the 12V rail and this rail can handle the highest power demand.  

Seasonic Platinum Focus 750 Specifications
DC Output    Rail        +3.3 V       +5 V    +12 V    -12 V    +5 VSB
Maximum Power           20 A         20 A     62 A     0.3 A    3 A
                                      100 W          744 W    3.6 W    15 W
Total continuous power    750 W