STX with m.2 GPU fun build

[TheGooseIsHere]

The downclocking issue doesn't happen at all from heaven stress testing, but happens again after playing one game of SC2 which isn't a very graphics intensive game. Weird.

It could be a power saving feature going wrong or something, gonna try using the EVGA Precision X feature to maintain a stable clock rate.

EDIT: Just realised I had an old nvidia driver installed which apparently caused this issue. Updated to newest drivers and it seems to be fixed

 

[TheGooseIsHere]

Small update on testing the 24v-12v step-down voltage converter, which is outputting 12.4V according to my multimeter from a 19.5V laptop charger (before being connected to the GPU adapter).

I have currently got it powering an ATI Radeon X1550 graphics card (which is only a 35W TDP card mind you) through the P4SM2 and it's currently running without issues!

The wiring goes like: 19.5V Laptop charger -> 12V voltage converter -> P4SM2 (by a floppy disk connector). Note, only the 12V and ground wires are connected to the P4SM2 (no 5V).


The next step will be to move onto a slightly more powerful 150W card (also found an old GTS 250 that I'm willing to sacrifice).

EDIT: tested with GTS 250 and it's working fine as well (note, the GTS250 needs a 6-pin power input, so I am using an ATX PSU to supply the 6-pin connector to the gpu, the laptop charger is only being used to power the p4sm2 atm since it's only a 120W charger and I don't want to overload it).

 

[TheGooseIsHere]

The next step is powering both the motherboard and the graphics card with a single charger. To do this, I will have to make a splitter by splcing the voltage converter and a 5.25mm plug for the mobo into the laptop charger. I have no idea what will happen when I do this and it will be putting the motherboard at risk, but this is for science so it's ok, right?!

 

[jeshikat]

12.4V is on the high side, is that measured at idle or load? Usually they'll set the voltage a bit high at low load so as the voltage drops under load it'll end up closer to the nominal voltage.

But if that's at load that'd be higher than I'm comfortable with.

 

[TheGooseIsHere]

That was measured at the converter without even being connected to the P4SM2. I measured it again at the FDD connector just now while the gpu is under load and it's showing 12.30V, so presumably that extra resistance lowers it a bit.

 

[jeshikat]

Ok, that should be fine then.

 

[TheGooseIsHere]

I'm now pleased to say that I've got it working off one AC adapter now!

I'm hesitating to load test this though as I'm using just a 120W adapter for both the CPU (65W tdp) and P4SM2 adapter (which could also draw 75W), meaning that in theory the system could draw up to ~150W under full load from the adapter.

EDIT: just realised the adapter is actually 130W, so I've disabled 2 cores on the CPU which should hopefully lower CPU power consumption and will do some medium load testing to see if it is stable.

 

[TheGooseIsHere]

Small update!

I have ordered a EVGA GTX 1060 SC (ACX 2.0) graphics card to use for this build, as it will provide plenty of power for any game at 1080p and the dimensions are small enough that it can fit alongside the STX board fairly nicely (although it is a few centimeters longer, this should still be ok).

Using my professional CAD modelling skills (Kappa), I have done a rough layout of how an acrylic or aluminium enclosure could be made to protect the graphics card and attach it onto the STX case.

 

[jeshikat]

You can model it in 3D all you want but nothing can beat actually mocking it up in reality, even if it's very rough

 

[TheGooseIsHere]

You can model it in 3D all you want but nothing can beat actually mocking it up in reality, even if it's very rough

Yeah

Anyway regarding powering the system, I think I will end up going the route of using 2x power adapters instead of using the voltage converter. The reason being that the converter itself is quite large and would add bulk to the system (even though an adapter is larger, it can at least be hidden away under the desk), plus it is another point of potential failure (unsure whether it will run consistently in the long term). So the dream of running the whole thing off a single adapter will be put on hold for now.

I have ordered a 192W 12V Leicke power supply which conveniently uses a molex P4 connector as the output so it will be convenient for custom wiring. I will splice in a FDD and PCI-E 6-pin connector to use for the graphics card.

 

[TheGooseIsHere]

Ugh.. Turns out the 192W adapter was no good, it made a terrible whiny noise when under any significant load so I will be returning it. I will stick to using the 150W adapter instead (which should be enough for a 1060 with a slight power limit decrease).

I spent most of the evening make a custom set of wires from DC to 6-pin pcie + fdd connector, and it seems to be running ok with the 1060. For some reason I can't see the GPU's core voltage in afterburner, not sure why that would be. Apart from that there don't seem to be any major issues with it, but going to leave it stress testing for a couple of hours to make sure it's stable.

 

[TheGooseIsHere]

Time for another update!

I have installed a Noctua NH-l9i heatsink on to the CPU and changed the riser to a 4x -> 16x one (since the cable is less wide).

Still haven't made much progress on building a custom enclosure for the GPU, but I do have some parts ordered so may make progress on that soon.

 

[aquelito]

Do you think there is a real difference between the Bplus p4sm2 and some cheap generic adapters, apart from the fact it is powered ?

 

[jeshikat]

My research into M.2 to PCIe slot adapters showed that it was the most reliable one among people doing eGPU NUC mods.

Cheaper adapters looked to be much more inconsistent on whether they worked reliably or not.

 

[aquelito]

Ok, thanks. Same issue with the generic powered flexible riser ? Some can run in Gen 3, some only Gen 2 ?
Problem is the BPlus adapter costs 35€ more than the motherboard :/

 

[jeshikat]

How much does the BPlus cost there?

 

[aquelito]

Retail price on their website is $125.

 

[Kmpkt]

125 dollars for the Bplus M.2 to PCIe x4 adapter? That can't be right.

Edit: I just looked back at my receipt for the same unit purchased in January of 2015 and it was 50.00 USD. I wonder what has changed since then (my guess is nothing and they're price gouging something fierce)

 

[aquelito]

Found this one for 20€, from a trusted brand sold on a trusted website :

https://www.amazon.fr/DeLOCK-62584/dp/B00UQSBIOS

It is only PCIe Gen 2 though ; don't know for the BPlus adapter.

 

[Kmpkt]

PCIe 2.0 is about half as fast as 3.0. Considering 3.0 is already a slight bottleneck on current generation GPUs, I'd be reluctant to go that way. The BPlus unit seems to be going for around 65 on eBay which might be a better option.