GPU Thin Mini-ITX with discrete graphics card

[EdZ]

I stumbled on this article and it made me think of the 25W PCIe slot issue :

http://cryptomining-blog.com/3757-h...-target-limit-on-geforce-gtx-970-and-gtx-980/

Following their BIOS mod guide, do you think you could lower the PCI-E slot power value to 25W and balance the extra power onto the PCI-E 1 value ?
If such a mod could work, you wouldn't need a powered riser, provided you have GPU with a PEG connector.

I happen to have a PEG-equiped 750 Ti to test such a setup.

That looks promising. 66W to 25W is a hefty power cut though, and depending on how the power phases on the card are physically connected could result in odd issues if one/two phases are only putting out a quarter of what the other phases are.

 

[aquelito]

Do you mean that the PCI-E connector couldn't cope with an extra 39W ?

I thought less power-hungry like a GTX 960 would offer more margin but power tables are the same, except it has one less PCI-E connector :

 

[EdZ]

New Do you mean that the PCI-E connector couldn't cope with an extra 39W ?

No, I was thinking of power delivery to the GPU chip itself:
To turn 12V supply to the 1.X-ish V the chip itself uses, the card has a DC-DC converter on-board. Because the chip needs a very smooth (no variations) power delivery, it uses a 'multi phase' design: each DC-DC converter will 'pulse' its power delivery, as it uses an oscillator as part of its design for efficient conversion, so by using multiple phases those pulses can be smoothed out along with the aid of capacitors. e.g. in a '6 phase' design there would be 6 separate DC-DC stepdown converters, each 'peaking' their delivery one after the other (the oscillations are 'phase offset'). When all phases are providing the same proportion of the power, this works fine. When one or more phases are not, there is now the possibility of a 'dip' in available power.
Now, variations in phase power is expected and GPU (and CPU, located on the motherboard) power conversion circuitry is designed to compensate for this. This is why there are capacitors on the board: they smooth the power delivery by 'buffering' the power (to greatly simplify things). Where a problem could occur is if the amount of power the capacitors have been specced to be able to 'buffer' is sufficient to handle one phase suddenly supplying 1/4 of what it was designed to do. If it can, you're living on the edge of stability but the card should still work. If it can't then, dropping the PCIe slot power draw that far may result in instability as power delivery is no longer smooth and always available.

Because this is dependant on how the power delivery circuitry on the card itself was designed, this is not really something that can be determined beforehand.

 

[aquelito]

Thanks a lot @EdZ for the simple and clear explanation

Anyway, I solved the issue by sourcing, thanks to @Kmpkt, what appears to be a reliable flexible powered riser !

http://vary.technology/photo/RP22-WebShop.html

I have been in contact with their team and they accepted to sell a powered version of their riser on their webshop
Very nice people to deal with.
I ordered a couple of those and I hope to get them to use soon enough.

No. As I said previously, it looks like the internal 12V input/ouput is AFTER the initial DC regulator that converts the 19V-12V range to a solid 12V. So trying to power a big GPU from that might kill that regulator. This sort of thing worked with older boards because the internal and external sockets were connected in parallel directly.

@iFreilicht I contacted Gigabyte support, and their very short answer strongly suggests the internal ATX_12V connector is acting as a direct passthrough :

- when using a 12V AC adapter plugged into the DC jack, what is the maximum power the connector can output ?
- If 19V DC_IN is connected, ATX_12V will also be 19V.

To me, it means there is no voltage regulation prior to the ATX_12V connector (which name is confusing), such as the more "standard" Asrock H110 TM-ITX.
Which is a great news

 

[iFreilicht]

@iFreilicht I contacted Gigabyte support, and their very short answer strongly suggests the internal ATX_12V connector is acting as a direct passthrough :

- when using a 12V AC adapter plugged into the DC jack, what is the maximum power the connector can output ?
- If 19V DC_IN is connected, ATX_12V will also be 19V.

To me, it means there is no voltage regulation prior to the ATX_12V connector (which name is confusing), such as the more "standard" Asrock H110 TM-ITX.
Which is a great news

Awesome, that's great to hear! Looks like you're all set then

 

[aquelito]

@iFreilicht, to follow up on the possibility to feed both motherboard and GPU with a single 12V Dell DA-2 modded brick (5.5 barrel connector), do you think it would be risky to create the following wire harness ?

4-Pin (motherbord header) > 6-Pin (GPU power connector) + 2-Pin Molex (Riser power connector)

Essentially : 2 x 12V pins + 2 x GND pins > 4 x 12V pins + 4 x GND pins, to power a GTX 1060 for instance.

 

[iFreilicht]

Depends, what is the maximum power the 1060 you're planning to use is pulling?

 

[aquelito]

In average, the load distribution of a stock GTX 1060 at full tilt is the following :

- PCIe slot : around 60W, which means around 35W on the Molex wiring.
- 6-Pin PCIe : around 60W.

The Dell brick would also have to power a 35W i5 6400T and a mSata drive.

The limit here may be the 5.5mm barrel connector ? What is its current rating ?

 

[iFreilicht]

That sounds reasonable, the cable harness itself wouldn't be a problem at all. Yes, you're right that the barrel connector seems to be the bottleneck here. I don't know what the exact part number of the one on the gigabyte board is, so I have no idea about the current rating. Maybe shoot them a mail and ask? Or see whether you can find a number printed on the connector in the product pictures.

 

[aquelito]

Just asked Gigabyte support waiting for their answer.

According to this Akitio mod (I remember seeing this on this forum as well), no issue powering a GTX 1060 through a 5.5 x 2.1 barrel connector.

That's also true that I will be adding 50W to the equation...

EDIT : found the SFF post

 

[aquelito]

That sounds reasonable, the cable harness itself wouldn't be a problem at all. Yes, you're right that the barrel connector seems to be the bottleneck here. I don't know what the exact part number of the one on the gigabyte board is, so I have no idea about the current rating. Maybe shoot them a mail and ask? Or see whether you can find a number printed on the connector in the product pictures.

According to Gigabyte, the DC connector is rated for 10A...
It might be just enough for a GTX 1060 alone or a GTX 1050 Ti and a 35W CPU.

The connector itself bears the serial number "160414+f5" and has a sea horse logo engraved on it.

 

[iFreilicht]

According to Gigabyte, the DC connector is rated for 10A...

120W then at 12V and 190W at 19V. That's not much, unfortunately, but with 19V you could internally connect a 19V-to-12V DC supply and power the GPU off of that. Might barely be enough, and the connector probably has a little bit of tolerance upwards, so it could work.

The connector itself bears the serial number "160414+f5" and has a sea horse logo engraved on it.

Hm, that doesn't seem to give me anything on digikey or element14, so it's probably a serial number as you say. But if Gigabyte is truthful, we already got all the information we need anyway.

 

[aquelito]

120W then at 12V and 190W at 19V. That's not much, unfortunately, but with 19V you could internally connect a 19V-to-12V DC supply and power the GPU off of that. Might barely be enough, and the connector probably has a little bit of tolerance upwards, so it could work.

The whole interest of that board was to use a 12V AC adapter (a 18A modded Dell DA-2 in my case) to skip the need for an extra DC board.
120W is still OK for a 35W CPU and a GTX 1050 Ti though.

However, if you want to power something like a GTX 1060, then you need to use that kind of adapter with the unmodded Dell DA-2.
Not as clean as you end up with more cables and you have to jumper the brick, but it should be good enough for a GTX 1060 and 65W CPU.

If you choose to go the 19V road with the on-board DC connector and a dedicated DC board, then it's better to use the Asrock H110TM-ITX board, which has a 7.4mm DC connector, while the Gigabyte has a 5.5mm (and is also much cheaper).

This is already what I have in my current system and it works great.

Any idea about a brand with a sea-horse logo ?

 

[iFreilicht]

Any idea about a brand with a sea-horse logo ?

No, sorry :/

 

[aquelito]

@iFreilicht, I found the DC connector brand in Intel's Thin Mini-ITX guidelines : it is called Singatron.

Hereafter the link towards the DC power jack connectors catalog :

http://www.singatron.com.tw/ftp/3C/Singatron-DC.pdf

Gigabytes 5.5mm DC connector seems to be the 2DC-G213 reference and is indeed rated for 10A @ 20V.
Asrock 7.4mm DC connector seems to be the 2DC-S060 reference and is rated for 10A @ 42V.

@QinX could run a 4790K and a GTX 970 off those DC connectors though (with a single Voodoo 350W brick).

 

[iFreilicht]

Gigabytes 5.5mm DC connector seems to be the 2DC-G213 reference and is indeed rated for 10A @ 20V.
Asrock 7.4mm DC connector seems to be the 2DC-S060 reference and is rated for 10A @ 42V.

Good find! That would be something to add to the wiki if you find the time.
So 10A either way. Remember that the important part is current, so when you're limited to 19V, you won't get more power through the bigger connector.

@QinX could run a 4790K and a GTX 970 off those DC connectors though (with a single Voodoo 350W brick).

Yeah, you can certainly try it and it might work, but be prepared that you might fry your board. I wouldn't recommend it.

 

[aquelito]

Yes, seems the reasonable thing to do

As I'm using a GTX 1060 and a 35W CPU, I'm still within the Asrock ratings.

Is there any reference of 5.5 DC power connectors than can handle more than 10A ?
Could find a Switchcraft ref @ 11A but nothing more.

It's a bit frustrating knowing that there are 12V bricks with a 5.5 DC jack providing 24A.

 

[iFreilicht]

11A seems to be the maximum, at least for barrel connectors. Everything above should use something much beefier.

 

[Fireside]

Is there motherboards with a different connector to handle something beefier? Would be interesting if there were.

 

[aquelito]

Nope.

To my knowledge, the only thing that would fit the bill was @QinX elusive PCB to redistribute power from a single 12V brick :

BUT

Check @Kmpkt Dynamo board, which should work with both 19V and 12V boards.

It will allow you to redistribute power the same way :

- one V out for the motherboard 2-Pin Mini Fit Jr 19V input
- one PCIe 8 pin for the GPU

Due for may hopefully