arclord

I just got my Akitio Thunder2 and I'm trying to test it on my G46vw with Thunderbolt using a GTX 760 and Corsair CX750M power supply. Basically,the same setup as here except with a 760 instead of 780 Ti (and of course G46 instead of Mac Mini). Just as he does, I have the enclosure powered with it's PSU, and the GPU powered with the 6+8 pins from the ATX PSU. I'm on Windows 8.1, but cannot get it to recognize the card at all, after trying every possible order of plugging in, turning on PSUs, restarting etc. Should I have to do anything in the BIOS or drivers before you could get the device recognized in Windows?

It sounds like I will need a powered riser, which I have on order, but entzoe was able to make a 780 Ti work without a riser. I'm probably missing something, since I'm new to all of this. Any thoughts on where I go from here?

Are you using Win8.1 with UEFI?

Thanks much appreciated

Here you go:

Quote

I always use the free standard reaolutions ~720p. It's gonna be awesome seeing some gaming on that little beast!

Quote

Can you do a Unigine Heaven Benchmark (v4) and also Cinebench(vR15) please. Would like to compare gaming performance and power consumption for a personal GRID Server In the works. My old eGPU Setup 2014 15" Top of the Line MacBook Pro + GTX 780Ti My cpu throttled even after getting 2 different units and undervolting so I returned the MacBook Pro. Now I want to build an always on GRID Server with as little hit in the power consumption department as possible. Nuc's have very low idle and load power consumption vs Mac Mini which has low idle but 3-4 times higher load power consumption and desktop CPU's are out of the question as far as idle Power Consumption goes (unless someone could point me in the right direction not looking for atoms lol).

@arclord

The GPU score of 3Dmark11 seems pretty good right?

Oh, and nice work with the enclosure buddy!

The GPU score for 3DMark11 is similar to those in the HD7870 reviews so it isn't too bad.

Thanks! By the way, what resolution should I use for the game benchmarking utilities?

Hi all,

I've read through the two Akitio threads, but I haven't been able to answer this question:

Is it possible to use the Akitio with a low profile (mini) GPU *without* any modding/splicing/etc?

I know that the max output is 25W... I'm wondering if there are any mid-range/decent GPUs that will fit which only require that much power?

If not, is there a clean/simple solution that gets decent performance?

Thanks in advance - sorry if these questions have been answered before!

I think this is the fastest GPU that fits the 25W power requirement:

http://www.geforce.com/hardware/desktop-gpus/geforce-gt-730/specifications

The GT 730 (Kelper core) with GDDR5 is the one to get. However, it is a mid to low end card.

Nice, thanks buddy!

I'll just post some games that I find interesting, maybe you happen to have one or two of them:

• Borderlands 2
  
• Chivalry: medieval warfare
  
• Counter-strike GO
  
• DOTA 2 (free)
  
• Skyrim
  
• Metro: Last light
  
• Far cry 3
  

I'm looking at your scores for 3DMark06 and I guess they are a bit low due to the CPU. I wonder if 3DMark11 would be more suitable since the test is more CPU independent than 06, would you mind do a test?

I know I might ask a lot of stuff from you, but you can use all this info in your write-up later =)

After all you are the pioneer of this kind of implementation

The i3-3217U is probably the limiting factor since it is dual-core and has a maximum speed of 1.8 Ghz.

Here is the 3DMark11 score:

http://www.3dmark.com/3dm11/<wbr>8749410

This test was done with the final setup in the Elite 130 casing. Essentially, there are 3 cables connected to the setup. The power cable to the ATX power supply, the Thunderbolt cable to the TB board and the HDMI cable to the HD7870.

I don't have the games you listed. Will try Bioshock Infinite and RE5 benchmarking utilities soon.

Forgot to add, you might try M-low's rEFInd solution posted here to deal with the error 12 under UEFI mode with the GT 730.

Thanks, will give it a try once I have time.

- - - Updated - - -

@arclord

This some interesting stuff right here, good job buddy!

Please post the links to the new 3Dmark scores when you remove the redundant lanes of the riser.

This concept is pretty awesome imo, I'm really interested to see some gaming tests too if you don't mind?

Here are the 3DMark and 3DMark06 with the last ribbon cable removed from the PCIe riser:

http://www.3dmark.com/3dm/<wbr>4083085

http://www.3dmark.com/3dm06/<wbr>17657202

The scores are quite similar. Which gaming tests should I run?

Thank you for posting this guide. The are the first pioneer to release a public Thunderbolt eGPU implementation on a NUC.

Only tweak to your config I can suggest is you could remove the last ribbon cable of your PCIe riser. That's because there is only a x4 2.0 electrical link to the video card so the last ribbon cable that services higher lanes (probably x9.. x16) isn't sending traffic.

You can confirm that removal of that ribbon cable has no side effects by running GPU-Z and confirming it's still a x4 2.0 link as it shows currently.

Thanks for the suggestion. Will remove the last ribbon cable of the PCIe riser.

Here is a short guide on setting up the HD7870 for the Intel NUC DC3217BY (with Thunderbolt 1) using the AKiTiO Thunder2 PCIe Box.

Hardware:
•    Intel NUC DC3217BY with Thunderbolt 1 + Win8.1/64 (EFI) 
•    Crucial M500 480GB mSATA
•    2 x 4GB DDR3 1333 ram (old ram from my laptop)
•    AKiTiO Thunder2 PCIe Box + Thunderbolt cable
•    Seasonic X 760W
•    Powercolor Radeon HD7870
•    PCIe 16x riser extender with power molex
•    SWEX (ATX power supply on switch)

As mentioned in other guides, the AKiTiO 60W power supply is not required as power is supplied via the molex on the PCIe 16x riser extender to the Thunder2. If power is provided by the AKiTiO power supply instead of the molex, the HD7870 is unstable (not detected by Win 8.1 at times). This might be due to the HD7870 drawing more power from the PCIe slot than what the AKiTiO power supply can provide.

 

Software: 
The M500 was partitioned using GPT. Installed Win 8 and upgraded to Win 8.1 without any major issues. Thunder2 was detected by Win8.1 after connecting to the NUC. Detection of the HD7870 required a reboot and Catalyst drivers was installed after that. Both Catalyst 14.4 WHQL and 14.7 RC3 drivers work for the HD7870.

 

Benchmarks:
3DMark (with and without HD7870)




3DMark 06



Other graphics cards tested:
Powercolor R9 290X (using PCIe 16x riser extender) – 290X detected by Win8.1 and Catalyst drivers installed. However, artifacts appear on the desktop screen after a few minutes and Win8.1 crashes.

EVGA GTX 750Ti SC (using PCIe 16x riser extender) – 750Ti detected by Win8.1 and GeForce drivers installed. However, Win8.1 hangs after a reboot.

PNY GT 730 (inserted directly into the Thunder2) – 730 detected by Win8.1 and GeForce drivers installed. However, Win8.1 reports error 12. Tried the DSDT override but error 12 still remains. By the way, this card uses 25W power only and has 1GB GDDR5 ram.

I am currently migrating the Thunder2 PCIe + TB boards to the Cooler Master Elite 130 mITX casing.



Here are the 3DMark scores with the last ribbon cable removed from the PCIe riser:

http://www.3dmark.com/3dm06/17657202
http://www.3dmark.com/3dm11/8749410
http://www.3dmark.com/3dm/4083085


Here is the 3DMark11 score:
http://www.3dmark.com/3dm11/8749410

This test was done with the final setup in the Elite 130 casing. Essentially, there are 3 cables connected to the setup. The power cable to the ATX power supply, the Thunderbolt cable to the TB board and the HDMI cable to the HD7870.

 

Thunderbolt 3 is slated for Skylark, the platform after Broadwell. So no improvements in TB2 bandwidth in the 2015 i7-quad Broadwell Macbook Pro 15"

Broadwell's 22nm->14nm die shrink should see turbo boost improvements just like the Sandy Bridge 32nm -> Ivy Bridge 14nm did http://forum.techinferno.com/throttlestop-realtemp-discussion/6958-haswell-step-backwards-ivy-bridge-i-have-some-shocking-tdp-results.html#post95181 . You'll of course see a faster iGPU too.

A NUC straddles the line between a notebook and a desktop. There have been others looking to use a NUC for eGPU connectivity. As you are the first to do so, would you mind providing a GUIDE of how you got it all to work? (pics, benchmarks, steps)? There are several guides is this forum already that can be used as a template.

The generous user contributed GUIDE are what keeps the eGPU momentum.

Will probably wait for the Broadwell Macbook Pro then since I don't need a laptop urgently.

Sure, I will post a guide for the Intel TB NUC soon.

Hi arclord!

Would you mind giving me two measurement from the inside of the enclosure:

1. Between 'inside front' and back to the fan?

2. Between 'inside front' to the very back (pass the fan)?

With those measurements I would be able to see what GPU's I could fit inside the enclosure.

Thanks a bunch buddy!

From the back of the Thunder2 to the fan at the front - 18 cm

From the back of the Thunder2 the front (pass the fan) - 20 cm

There are no 6 AKiTiO Thunderbolt eGPU implementations: http://forum.techinferno.com/diy-e-gpu-projects/6578-implementations-thunderbolt-expresscard-mpcie-egpus.html#post89707 . 3 required PCIe risers and 3 didn't. So by the current rate you have a 50% chance of it working without a PCIe riser. You may then want to add a powered PCIe riser to your shopping list. They're a couple of bucks from ebay.

I got the AKiTiO Thunder2 to work with the Intel Thunderblot NUC and HD 7870. A molex powered PCIe riser is required though.

Do you think it is worth waiting for the Broadwell Macbook Pro next year or get the Haswell Macbook Pro now?

Thanks for providing us with a wealth of information on the various eGPU setups!

I am planning to move to another Thunderbolt platform since I couldn't get the Thunderbolt NUC to work with my Nvidia graphics cards.

Have successfully migrated the Thunder2 PCIe + TB boards to the Cooler Master Elite 130.

I am currently supplying power to the Thunder2 PCIe + TB boards and HD 7870 using an external ATX power supply. Will install the ATX power supply into the Elite 130 once I finish with cable management.

The Elite 130 is a nice mITX casing for the Thunder2 since it can take graphics cards up to 13.5 inches.

I'm not sure if your setup works, but I plan on using the setup below:

You can find more details on this page.

It provides all the ATX connectors including the 6-pin PCIe power input. Maybe it will help in setting up your Akitio box.

Thank for the suggestion Arclite. Unfortunately, HDPLEX 250W requires 16V-24V input. My power adapter provides 12V.

- - - Updated - - -

I googled a bit and noticed that there are these cables for 6-pin PCIe power

There are 3 pairs of (12V + ground) barrel connectors.

So this connector

should have 3 inputs in order to connect those 3 pairs (12V + ground).

The PCI cable you show has yellow, black and brown cables. Im not sure how those 3 cables are connected to the barrel plug. It may not result in right wiring for 6-pin PCIe power.

Do you have more information on the PCIe cable you found, the one with yellow, black and brown cables? Maybe a full picture of it?

I already have item 1 (120W power supply) and 2 (+/- adapter for 5.5 x 2.5 mm barrel plug) so I am thinking of ways to make use of them.

I don't have item 3 (PCIe cable) but can get it easily. If I get one with 3 yellow and 3 black wires, is it safe to connect all 3 yellow cables to the positive point and all 3 black cables to the negative point on the adapter?

With regards to graphics cards that require additional power from a PCIe 6-pin plug, is it possible to supply the 12V power using the following:

I am hoping to make use of a 12V/10A (120W) power adapter to supply power to graphics cards instead of using an ATX power supply. The Thunder2 PCIe Box will be powered using the default 60W power adapter. No PCIe riser extender will be used.

Is this setup feasible?

My Sonnet Echo is already sold, but i wasnt able to get it working with the 780 (not Ti).

Will test some games later today and report.

EDIT: Tested and unstable. WoW for example was working but as soon as i tab in and out it crashes with BSOD. I attached my powered riser again and working.... Strange thing is that real stresstests like Furmark were working without problems.

Another problem is the dGPU (K2100M): I cant install the driver or my eGPU isnt working (no display).

EDIT2: 2 Problems: 1. I cant use my dGPU (with installed dGPU my eGPU shows no display)

2. Its unstable. I tested with World of Warcraft and with and without the powered riser i get BSODs or other stuff.

I think i stick with my PE4L 2.1b

Thanks for testing. Do the GTX780 Ti and R9 290X exhibit the same instability issues?

Hey,

i tried the Box on my W540 and GTX780 Ti and R9 290X working without any problems (TB1 + W8 x64, Win7 wasnt working).

I dont even need the powered riser (tested with Furmark and some other stresstests).

Did you manage to get the GTX780 Ti and R9 290X to work with your W540 using the Sonnet Echo Express III-D?

It will depend on how the powered riser cable was built. The connector itself can withstand about 11A (122W) but a single wire can top out at 5A (60W) which is not sufficient to meet the 75W PCIe spec + power the TB card and may crash or burn out under load. Some powered risers use two 12V wires so in theory they can carry up to 10A (120W) which should be enough for everything.

Does your powered riser cable have a solid capacitor soldered on (like this one)? According to bitcoin mining forums, more recent versions of powered risers have capacitors built in to stabilize power supply and prevent system crashes under sustained load.

This is the one that I am using. It only has one 12V wire so I guess it supports up to 5A (60W)?

The idea proposed earlier does not bypass the 4 pin yellow/black cable connector on the TB board.

Instead what I was doing was suggesting was:

1. supplying the ~35W (60W-25W=35W remaining) to the TB board either via the stock 60W AC adapter, custom molex-to-barrel or a modified floppy molex cable.

2. 75W would be provided to the PCIe slot directly bypassing the x4 connector. That bypass circumventing any power restricting fuses the TB board might have as well as avoiding passing up to 75W across the x4 connector that's probably not specced to carry that much. AKiTiO speccing it at 25W. The bypass done either using a powered PCIe rise plugged into the x16 slot, soldering 12V/GND directly on the AKiTiO PCIe board or coming up with an alternate powered x4-to-x16 female ended board to replace the AKiTiO PCIe one.

This is the safest solution right?

If we allow the TB board to draw power from the powered PCIe riser plugged into the PCIe x16 slot of the PCIe board, the molex supplying the power could potentially be overloaded with the 75W power required by the graphics card + power required by the TB board? Conversely, if we allow the PCIe board to draw the 75W power from the x4 connector via the TB card, the TB card could potentially be overloaded as well?

- - - Updated - - -

Is your solution running stable?

I'm just wondering, because some people posted, that their (even with a powered riser) solution were unstable and only worked for some time until windows crashes.

I think those were mostly R9 290 and 290X cards.

My Powercolor HD 7870 runs stably but my R9 290X causes Windows to crash after a few minutes.

Now very promising news! B1 connector (rear side J10) on the right corner of "x4":

[ATTACH=CONFIG]12324[/ATTACH]

8 small holes through the board, 7 of them gave continuous beep sound with PCIe 2nd pin, one is different but couldn't get GND continuity to it. The nearest GND is marked on the picture, that point is very small, but I was able to get continuous beep when paired with a screw hole.

EDIT: Voltage metering also proved that this is the right place to feed extra power. All those 7 points gave +12V. Is this a standard size power connector?

The more I look at the TB board, the more it resembles a PCIe x4 TB card like the one from MSI.

https://thunderbolttechnology.net/product/msi-thunderbolt-plus

Building on Arclite's idea of a female-female PCIe 4x to x16 powered riser, can we connect 2 of the following PCIe x16 powered risers together using only the female portions? The cables are detachable. We will have a female-female PCIe x16 to x16 powered riser - one for the TB board (if it is a functional PCIe 4x card) and the other for the graphics card.

My Gigabyte 750Ti works with a MacBook Pro Iris on Win8.1 with 6-pin power cable plugged (but doesn't fit inside the enclosure). 750Ti fans spin at full speed before starting the computer, which is quite annoying. The differences are that you use a NUC and your GPU is only PCIe powered.

By the way, what version of the Geforce drivers are you using for your Gigabyte 750 Ti?

Thanks for doing all the voltage tests on the Thunder2 PCIe Box. It would be great to know if it does indeed have a 25W limit.

Just read up on the HD 7000 series and the R9 series card. The main difference is the R9 series cards has Boost Clock while the HD 7000 series don't.

Both the R9-270X and HD 7870 are based on the same GCN chip (Pitcairn GPU).

Maybe the Boost Clock is the cause of the instability.

Sadly the R9 290 works for me neither. I was able to install the newest AMD driver, but after a few minutes windows becomes very unstable and crashes with a black screen.

Tried the R9 290X and have similar results. Windows crashes after a few minutes.

The HD 7870 works perfectly though. Kind of strange since both are based on the GCN architecture.

Congratulations. Please keep us informed :-)

Thanks! 3DMark scores are kind of low but not unexpected given how slow the i3-3217U is.