borealiss

I ran into this problem sort of, but was able to use the output of the egpu as an extended desktop card. when you try to set resolution and monitor setup, does windows show more than 1 monitor?

I'll try 1.2, thanks, I didn't know it was out. As for rendering internally, what makes you say you can't? In the nvidia preferences, you can tell it to either use the integrated graphics or discrete (egpu) graphics. I set it to discrete and I see a huge performance increase when rendering on the internal display. - - - Updated - - - No. I used gpu-z or some other utility to confirm it stays at gen2.

Unfortunately I can't speak for ATI cards. What you can try to do is disable your onboard graphics and then see if the ATI card is enabled in device manager. If you are still seeing error 12, then go and disable every unnecessary device until you get it to a configuration that works. If you can never get to a working configuration, you might be out of luck. Compaction is known to have some bugs in it, and Nando had indicated there was a fix coming. I've never heard of a follow up, so you might want to check with him. If you do get the ATI card into a working state without error 12, even with everything disabled, write down the MMIO ranges. You'll have to set them up manually in your pci.bat file by using setpci. A quick google search will tell you what the syntax is for setting the BAR registers to mimic the working MMIO range. From here you will have to slowly enable all your devices and not which ones cause error 12s on your ATI card. Whenever a device causes an error 12 on your ATI card, you'll have to append that device to your pci.bat file and manually set the BAR registers with setpci for that specific device. This is really slow and time consuming, and honestly I don't think it's worth your time. Just use the nVidia card. As for why you're not able to output to external display, I'm not sure. Have you tried native DVI or HDMI output to input? Are you able to render on the laptop's internal display? - - - Updated - - - Also, are you booting up with external monitor connected, or are you connecting after poweron?

I connected the 12V line up to one of the power rails on the ATX power supply. I connected the 3.3V line up to a SATA power connector. Honestly, if you're willing to cannibalize an ATX PSU, you can just hook it all up directly. Look at the ATX pinout and PCIE pins 1-11. The 3.3V lines on the pcie interface can be hooked up directly to the ATX PSU, you just need to do some soldering, or wire wrapping. Same with the 12V lines. If you don't want to sacrifice a PSU, get an ATX PSU extender cable, and splice into that. The wiring loom shouldn't take more than maybe an hour to make. This is the cable I used to splice into everything, so I don't void the warranty of the thunderbolt chassis. Google Image Result for http://image.ecbub.com/productd/937/5379976_1.jpg

You're not going to be able to get the eGPU to boot with BIOS emulation using bootcamp if it is plugged in from cold start. You need to do something to delay the assertion of the powergood pin on your card. Check out PCI Express - Wikipedia, the free encyclopedia The PWRGD pin is the one I hooked up to from my ATX powersupply to the pins on the PCIE board. ATX - Wikipedia, the free encyclopedia It's the grey pin I believe. Just to double check, you should take a digital multimeter and probe the PWRGD pin on your Helios to make sure it's the same 5V supply as the ATX powergood signal. I didn't do this and plan to. Online references suggest these are the same power rails in a system, but it's never too good to be safe versus blowing up something on your card. You also need to wire up your ATX powersupply so that it can be switched on without being plugged into a motherboard. How do I manually turn on an ATX power supply? | techPowerUp I had to use an interposer/extender cable to tap into all of the powerlines on pcie pins 1-11, sides A and B. So my eGPU is completely power isolated from my sonnet adapter. You will have to do something similar, as well as hook up the powergood pins. Your ATX powersupply should also drive any auxiliary power to the card via molex or pcie power connections (sideband). This way your ATX/external powersupply coordinates the poweron sequence for your eGPU. You will then have to have your Helios connected at cold start, but have your powersupply turned off. Once you hold down OPTION to boot into windows, wait 4-5 seconds then power on your eGPU via powersupply. One thing I do here is install the DIY egpu software so that when the windows boot menu comes up, I can go into the egpu software setup and scan to see if the eGPU got enumerated and is present. If it didn't, just power off your powersupply, wait a few seconds, and turn it on. Then hit F5 to rescan. Once you get the timing down, this can become unnecessary. Once you've verified that it's detected, chainload your MBR and boot into windows. This whole process usually takes me about a minute to do whenever I enable my eGPU. Unfortunately, until a thunderbolt chassis manufacturer provides a way to assert/deassert or delay the assertion of PWRGD on the PCIE interface, this is the only workaround I know of. It also has the detrimental effect of screwing up the form factor of the chassis, but at least with BIOS emulation you get a 100% working optimus solution with nVidia graphics cards. I have taken this setup on the road, and my sonnet/egpu/psu setup fits very nicely in a small tote bag. It is awesome. You might want to give the EFI BIOS installation method a try. Booting via EFI allows you to bypass the power isolation workaround because the EFI BIOS that windows uses properly enumerates the graphics card and you will not get the hanging you see when you boot up with the eGPU plugged in with bootcamp. Good luck.

an additional power supply. I'll try to post a youtube at some point.

In basic networking courses you learn the concept of layering. Phy, link, protocol, and app layers. It's the typical OSI model that many networking protocols follow. This is what I meant. Thunderbolt and pcie build on this concept. Pcie uses thunderbolt to replace it's phy layer and sort of gets pushed up into the application layer due the encapsulation thunderbolt provides. This is possible because the thunderbolt controller has specific logic provisioned to perform this functionality transparent to any system with pcie. This USB controller doesn't have this at all. You could theoretically have a software initiator that could form pcie commands and use the host CPU to encapsulate it within a USB3 packet to deliver over a USB3 phy, but that would be really slow and require custom software. Microsoft does this with their iSCSI software initiator. http://www.microsoft.com/en-us/download/details.aspx?id=18986

i don't think you understand exactly what a thunderbolt controller does. it maps pcie and mini dp commands at the application protocol layer onto thunderbolt packets to be sent over the thunderbolt interface. to perform this specific function, it has to have built in logic to do this. it is meant to carry native pcie and mini dp commands over the thunderbolt link and phy layers. a usb3 controller that sits off a pcie gen3 bus doesn't do this. a host system will issue commands over pcie that get translated into usb3 commands such as reads and writes, but there is no specific mapping of pcie commands onto the usb3 protocol. this is because usb3 is meant to be a data transport protocol. pcie is meant to be both a command and data driven interface for non-coherent traffic. in some cases it's also used as a system interconnect. to do what you want would require a completely custom solution. something like an egpu knows how to talk to a host system via pcie. if you cannot carry pcie commands encapsulated within a usb3 packet, it's not going to work. there are a lot of things that a thunderbolt controller does in order to present itself as a native pcie bridge controller within the system. a simple usb3 controller will not have this type of support buit in. if you didn't have a custom asic solution to transparently map system resources and still make the egpu presentable to the BIOS/kernel/OS, you'd need a lot of low level software support in order to make this functional. the egpu's driver would probably have to change significantly. to give you an example, when you use setpci to access pci configuration space and set the pcie BAR registers, you are doing this via a config access. a configuration access is a type of pcie command, and is actually legacy from the pci days. usb3 has no provision for doing this type of configuration access. it just doesn't exist. you'd have to encapsulate the entire config access command into a usb data packet, send it over, and have a usb->pcie translation solution strip off the usb3 protocol info to get the config access pcie command. this is networking 101. like iscsi, you could probably have a software initiator, but at the end of the link (egpu), you will need a custom solution from single or multiple asics. please read the thunderbolt technology primer. Thunderbolt? Technology: Technology Brief

yeah usb3 would require a completely new asic solution. you need something that will take pcie and push it through a serdes over a serialized phy, then have it broken back out into native pcie lanes.

Nando, you can work around the perst delay issue by using an expansion cable and isolating all power rails on pins 1-11, side A and B. These are the only pins on pcie that provide power. See wikipedia article on power distribution. PCI Express - Wikipedia, the free encyclopedia What you need to do is hook up the 12V (molex) connector to your ATX power supply. You can short 2 pins on your ATX power supply to get it to turn on and off manually like an AT power supply. Here's how I did mine. How do I manually turn on an ATX power supply? | techPowerUp If your ATX power supply has a switch on the back, then you can use this as your PERST delay switch. I used this cable, $18.99 plus shipping. PCIe x16 Extender Cable w/ Molex Connector The 3.3V power is a little tricky. I had to tap a SATA cable on the power supply. Then you can hook up pin PCIE pin 11A, power good, to ATX pin 8, power ok/good (gray wire). The PCIE auxiliary cables that some graphics cards require just hook up normally to the egpu. 1) Turn on macbook. 2) Select Windows boot. 3) Wait 5 seconds, then turn on PSU switch. Boot into windows. OR If you've got the diy egpu software installed, selected it, turn on ATX PSU, then hit enter and select option 2 to verify your egpu is working. 4) chainload to MBR and load up windows. If your egpu isn't detected, power down PSU, rescan with f5, and it should show up now. Some power supplies take a few seconds to discharge and for the power rails to go idle, so sometimes rescanning may be required. With mine detection of the egpu works every time with my sonnet echo express pro.

all, there is still the sonnet echo expresscard pro, which supports the same pcie gen2 x1 as the th05. you can still get the expresscard->pcie breakout board from bplus. a solution for thunderbolt still exists, albeit at a slightly higher price point. total would probably be a little over $200 without external power supply. heck you can even go with the non-pro version like i did 6 months ago. it's under $100, but will only support pcie gen1 x1, and will have some performance.

not necessarily. there still is the echo expresscard pro from sonnet. this supports pcie 2.0 x1 over expresscard. i used the non-pro version of this, which only supplies pcie 1.0 x1, when first putting an egpu over thunderbolt. Sonnet - Echo ExpressCard/34 Thunderbolt Adapter you will have to purchase the breakout board for expresscard to the full pcie x16 keepout that most graphics cards use. you can still get all these at bplus' website. the sonnet adapter goes for about $140. a power supply is about $30, and you can find and the expresscard->pcie breakout is about $70 on bplus' website. good luck.

how did nando derive the pci.bat? the startup.bat is easy enough. i've looked everywhere on the boards for how to derive this as this seems like a cleaner solution.

It sounds like it might not have flashed correctly? I've used the disk image utility and never had any issues. Perhaps you should try that. How does it behave when it's installed on hard disk? Also, when you try to eliminate error12 conditions, one thing I learned last night is that if you systematically go and disable your pci controllers (there should be about 6-8 of them under system resources in device manager), you can get some more MMIO space. You need to keep track of which ones control your essential peripherals like the thunderbolt adapter and your graphics cards. I've found that disabling about 2-3 of them gives me more flexibility, and although it may kill other peripherals such as LPC controller, bluetooth controller, or whatever, I don't use these. Good luck.

yeah it's too bad the cost of entry is so high now that the th05 is completely unobtainable. MSI's gus2 seems to be MIA too. but if you're going to spend $1000+ on an ultrabook like the macbook air, and $200+ on a graphics card, then $20 for the extender cable and $400 for the chassis is not too bad for the performance you get. other notebooks in this performance range cost probably $1500-$2000, and you don't get the modularity of the graphics card, plus the ultra portability of the macbook air. to me this is a fun hobby, so even if it didn't work out for me, i enjoy doing it regardless. i could spend my money on a lot of different things, but for $600, i've gotten a lot of enjoyment out of the whole thing, and learned a lot. i've certainly spent way more on my car hobby...