Tech Inferno Fan

Check if 1440x900 is listed perhaps against a different refresh rate than your monitor is currently set for. A good guide explaining how to check is at http://www.tenforums.com/tutorials/4910-screen-resolution-display-change-windows-10-a.html

Example: use hwinfo64 to check if your eGPU hosting PCIe port supports hotplugging

1. We know NVidia drivers 372.90 and newer require the eGPU hosting PCIe bridge support to support hotplugging or otherwise an error 43 will be issued. There is some suggestion this may be confined to Win10 anniversary edition [ comments welcome ].

2. Example below: a Dell E6540 (that I'm currently using). I find the wifi card to identify the mPCIe port that could host an eGPU. Then I click the hosting PCIe bridge to check if hotplugging support is enabled.

Pls work through https://www.techinferno.com/index.php?/forums/topic/8947-mpcieec-egpu-troubleshooting-steps/

13 minutes ago, biomangb said:

Wow, the price is insane ... I wonder who wants to buy this thing, because even the razer core is cheaper

They are advertising a turnkey solution (software + support), which has been been shown to be derived from DIY eGPU open source resourced. Apple users have deep pockets so may be willing to fork out the $$ they want.

11 hours ago, rozhkov said:

Answer from Bizon's support:

- we now have our own software

..

I think i got important info about this company, maybe moderators  or pro-users can write special topic / summarise it.

rip-off alert: BizonBox 3 is an overpriced and has incorrect description on the website :-)

Bizon will soon be price blitzed out of the market by AKiTiO Node. Their only point of differentiation will be "their" software and supposed support, both of which I'm confident have been heavily derived from the DIY eGPU publicly posted information. Apple owners seemingly with deep pockets and willing to pay for turnkey solutions.

@goalque has highlighed stability issues with the R9 2xx series + AKiTiO Thunder2. Eg, wrote at https://www.techinferno.com/index.php?/forums/topic/10655-akitio-2-amd-r9-and-mbpr/&do=findComment&comment=151415

Quote

At least R9 Nano and HD 7970 should stay stable with the Akitio on Windows 10.

Checking your TOLUD and creating a 'large memory' area are answered in:

2. How can I check if my notebook is compatible with an eGPU?  ->  https://www.techinferno.com/index.php?/forums/topic/6275-pre-purchase-faq/

https://www.techinferno.com/index.php?/forums/topic/5874-guide-dsdt-override-to-fix-error-12/

@bloodhawk, the NGFF.M2 GPU-Z output has faster fillrate, bandwidth, GPU clocks. Can you do like-for-like comparisons?

In addition, pls ensure the machine has the battery installed and is using a high-performance power profile.

You may consider also installing Throttlestop and disabling C1E, EIST, C6/C7 which (at least a couple of generations ago), would affect SATA SSD performance. That may help bring the NGFF.M2 and TB3 CUDA-Z bandwith results closer.

7 minutes ago, bloodhawk said:

@Tech Inferno Fan

 

Here it is : 

 

Great. Looking lower but not x4 2.0 levels.  Are you using an active TB3 cable?

Under the same load conditions, can you re-run CUDA-Z on the NGFF.M2 32Gbps interface?

workaround: Bypassing driver version checking for say BF1 which wants 372.90 or newer

From https://forums.geforce.com/default/topic/970339/geforce-drivers/372-90-373-06-drivers-preventing-use-of-gpus-compute-cards-over-m-2-ngff-mpcie-/post/4998890/#4998890 :

Quote

AaronZaBaron said:

KORDAEMUL said:One sad thing is that BF1 needs the latest driver... I cannot play this eventhough Pre-order.

Yes you can.
https://forums.geforce.com/default/topic/962633/pc-games/bypassing-the-driver-version-check-of-a-game-/post/4967359/#4967359

 

Just now, sashok724 said:

Tried gen1, nothing changed. Same for disabling HD5470. To disable HDMI audio, i just need to disable it in sound settings like this?

Maybe. I disable devices in Device Manager. Also check if you have eGPU stuttering while being used. I had a problem with crosstalk on the eGPU adapter cable that gave such choppy results.

31 minutes ago, sashok724 said:

Hi. I have Acer Aspire 5552g laptop with unlocked BIOS, EXP GDC v8.0 and AMD R9 380X GPU.

I've also enabled PCI-E gen2 in bios settings on all ports

 

The problem is, while GPU connected, i experience audio stutter and micro lags under any graphics activity

Here is screenshot of DPC checker while just watching video on YouTube:

I just don't know the cause of problem. Is it notebook, EXP GDC/it's cable, GPU or something else? What i can do to fix it?

Sorry for my english, it's not my native language

 

BTW: Is there any way to install newer graphics drivers? My notebook has Mobility Radeon HD 5470 dGPU, and latest driver that supports both eGPU and dGPU is Catalyst 15.11 beta

Switch to Gen1 speed for the eGPU and test to see if it helps. EXP GDC is known to cause Gen2 instability with the AMD drivers implementing some fault tolerance which may express itself as high DPC latency. PE4C 3.0 with it's soldered capable providing a more reliable Gen2 link.

The other things to try are disabling the AMD eGPU audio as well as disable the HD5470 where you'd use just a Basic Microsoft adapter instead.

On 15/07/2016 at 10:00 PM, acealone said:

Hey guys,

I just bought an eGPU Adapter for my Medion Akoya Laptop. It has an ATI dGPU and i attached a Nvidia eGPU to it. It works great so far. But i can't seem to get the internal display working. I mean the internal display works but i guess the dGPU still processes the internal Display. Is there a way to get the eGPU process the internal display?

Apply the HDMI-to-VGA dummy to create a fake LCD and use dual-monitor software to move *windowed* games started on the eGPU to the internal LCD. REF: https://www.techinferno.com/index.php?/forums/topic/8088-2012-15-mbpr-gt650m-gtx97010gbps-tb1-akitio-thunder2-win10-lukehero/

@bloodhawk, you wouldn't still have the TB3 enclosure? We need some CUDA-Z output to confirm if it's providing 20Gbps or 32Gbps. Without CUDA-Z your noticably faster 3dmark result using NGFF.M2 compared to TB3 favoring would suggest your TB3 interface was 20Gbps.

See the discussion RE: TB3 bandwidth at https://www.techinferno.com/index.php?/forums/topic/10718-2016-macbook-pros-and-egpus/&do=findComment&comment=151839

1 hour ago, goalque said:

https://thunderbolttechnology.net/sites/default/files/Thunderbolt3_TechBrief_FINAL.pdf

Thank you. Here we see the important info:

Quote

A Thunderbolt 3 port supports up to the following:

 

• Each connector must be provided with at least the equivalent of two lanes of PCI Express Gen 3 data

• Each connector must be provided at least one full DisplayPort 1.2a (four lane) interface

 

Starting with Intel systems based on the 100 Series chipset, Thunderbolt 3 is delivered via a series of discrete silicon devices that have varying capabilities. The dual port sku supports two Thunderbolt 3 connectors, and takes 4 lanes of PCI Express Gen 3 as an input, and 2 full (four lane) links of DisplayPort 1.2a.

 

The single port sku supports a single Thunderbolt 3 connector. In general, if a system has a single Thunderbolt 3 port, the connectivity behind that port could be as high as four lanes of PCI Express Gen 3 with dual DisplayPort links. It could also be as little as two lanes of PCI Express Gen 3, with a single DisplayPort link.

 

In each case, the Thunderbolt interface itself doesn’t change, it is still 40 Gbps; however, the underlying source data and display may be more limited.

...

As a reminder, four lanes of PCI Express Gen 3 operate at (4 x 8 Gbps) 32 Gbps roughly. Thunderbolt 3 uses PCIe x4 gen 3 data rate with 128kB header sizes. For a single Thunderbolt chip with two ports, the x4 PCIe interface data rate is shared across the ports

.. and fakebanana is seeing this exact sharing of bandwidth across the two ports:

10 hours ago, fakebanana said:

$ system_profiler SPThunderboltDataType

 

Thunderbolt:

 

 

    Thunderbolt Bus:

 

      Vendor Name: Apple Inc.

      Device Name: MacBook Pro

      UID: 0x0001493688391000

      Route String: 0

      Firmware Version: 10.8

      Domain UUID: BD1095C8-4F85-6F58-8501-4CF402D77BC8

      Port:

          Status: No device connected

          Link Status: 0x7

          Speed: Up to 20 Gb/s x1

          Current Link Width: 0x1

          Receptacle: 1

          Link Controller Firmware Version: 0.15.0

      Port:

          Status: Device connected

          Link Status: 0x2

          Speed: Up to 20 Gb/s x1

          Current Link Width: 0x1

          Receptacle: 18

          Link Controller Firmware Version: 0.15.0

Yep, two ports with dedicated 20Gbps each. If those two cannot be aggregated to one 40Gbps port (PCIe @32Gbps), then TB3 is yet another dud like TB1 was. If Intel follows with the TB1-TB2 model, then TB4 will give us 32Gbps.  Intel and their tick-tock

NGFF.M2 like @bloodhawk is using does give him 32Gbps (x4 3.0), confirmed with CUDA-Z host-to-device results. This further explains why the NGFF.M2 results were noticable faster than TB3; the latter of which we thought was due to TB3 latency but is now looking to be due to significantly narrower 20Gbps bandwidth. REF:  https://www.techinferno.com/index.php?/forums/topic/10579-17-clevo-p870dm-g-gtx108032gbps-ngffm2-pe4c-v41-win10-bloodhawk/

27 minutes ago, goalque said:

The picture explains it:

 

 

20Gb/s x2.

@goalque, where did you find that picture?  That shows 2x20Gbps 'physical' channels.

The original TB1 was marketted as 20Gbs 'logical' channels, being the cumulative bandwidth across two channels. TB2 allow aggregation of those two channels to form a single TB2 20Gbps channel. However, TB2's x4 2.0 electrical link limited it to 16Gbps for PCIe traffic.

If indeed TB3 doesn't allow port aggregation then we only have 20Gbps performance. That being x4 3.0 (32Gbps) limited to 20Gbps down the TB3 channel. That would then only be 25% more bandwidth than TB2. Hardly worth it if true.

The easiest way to find out for sure what the real TB3 bandwith is,  is for a TB3 eGPU equipped owner to run CUDA-Z and report their host-to-device results, comparing to the chart at https://www.techinferno.com/index.php?/forums/topic/10581-6th-gen-i-core-has-gen3-8gbps-southbridge-pcie-ports/

The other important point is whether active or passive copper TB3 cables are being used, offering 20Gbps and 40Gbps bandwidth respectively.

Try a 14"-15" Dell E6430/E6440-E6530/E6540.  E6x30=3rd gen i-core, E6x40=4th gen i-core. Consider:

- all have socketted upgradeable CPUs, docking station and 6/9-cell battery options

- E6430/E6530 with expresscard slot, optional on E6440/E6540

- E6430/E6440 with 1600x900 LCD option

- newer E6440 systemboard has eDP FHD LCD option

- E6430/E6530 with NVS4200M/NVS5200M dGPU option. BIOS setting for dGPU or iGPU+dGPU Optimus.

- E6440/E6540 with HD8690M/HD8790M iGPU+dGPU switchable graphics
- E6440/E6540 with mSATA slot

- E6540 has FHD LCD and HD8790M dGPU option, which is ~GT750M performance though note thermal throtting issue when run CPU+dGPU at full load

- all can be extensively modifed via UEFI variables (CPU/iGPU/RAM overclocking and RAID-0/1) per https://www.techinferno.com/index.php?/forums/topic/7574-14-dell-latitude-e6430-performance-upgrades-and-system-mods/ and https://www.techinferno.com/index.php?/forums/topic/6438-guide-dell-e6530-cpu-tdpmulti-unlocking/ .

If you are running a x1 2.0 mPCIe2 eGPU on your Y510P then you may consider a E6530/E6540 as a functional replacement for it. Comparatively, same-gen Thinkpads not offering this level of performance or flexibility, often even whitelisting their mPCIe slots.

You can still upgrade to Win10 from Win7/8 for free:  http://www.theverge.com/2016/8/1/12340318/microsoft-windows-10-free-upgrade-still-available

2 hours ago, bloodhawk said:

 

So if you guys have the Hot Plug option in your option in your BIOS for PCIe ports, maybe that will help. Or if @Tech Inferno Fan can implement it in his setup. 

 

Props to @GTANAdam for noticing the Hot Plug being enabled. 

@bloodhawk, @GTANAdam, great troubleshooting in getting to the bottom of this.

Unfortunately the PCIe port hotplug bits are read and write-once (R/WO), so we can't change them after the BIOS has set them as shown from Intel's datasheet extract below. I've also confirmed these are write-once by attempting PCI writes to the register and seeing it doesn't change.

Registry hack to disable hotplug checking?

@Nautis1100 noted at here that NVidia added two new registry - SurpriseRemovalSupport and EnableRunTimePowerManagement to enable hotplugging. Can you guys install 372.90, search for those strings and see if removing/disabling those options fixes error 43??

Petition a NVidia fix

If get no joy with registry hack and sticking with 372.70 isn't an option then next easiest path to pursue is to petition NVidia to disable the hotplug checking code that is causing error 43 in their 372.90 and newer drivers.

21 minutes ago, fakebanana said:

That's right - no 40Gb/s string:

Is this a 13" with 4 TB ports? If so, it only supports full TB3 performance on the two lefthand ports. From http://www.macrumors.com/2016/10/28/macbook-pro-tb3-reduced-pci-express-bandwidth/

Quote

Late-2016 MacBook Pro models vary slightly in the data speeds they provide to each Thunderbolt 3 port.

MacBook Pro (15-inch, Late 2016) delivers full Thunderbolt 3 performance on all four ports.

MacBook Pro (13-inch, Late 2016, Four Thunderbolt 3 Ports) supports Thunderbolt 3 at full performance using the two left-hand ports. The two right-hand ports deliver Thunderbolt 3 functionality, but have reduced PCI Express bandwidth.

MacBook Pro (13-inch, Late 2016, Two Thunderbolt 3 Ports) delivers full Thunderbolt 3 performance on both ports.

On 26/07/2016 at 0:55 AM, matp321 said:

... It also did not help. Here is my pci.bat file:  pci.rar

 

You have a 12.25GB-36-bit PCI compaction which will not work unless you have performed a DSDT override OR you disable the iGPU in Windows, perform your 36-bit PCI compaction on iGPU+eGPU but force the eGPU to 32-bit. That will see the eGPU hijaak the iGPU 32-bit resources.

16 hours ago, sangemaru said:

By the way nando, how did you get ASPM settings to stick through sleeps/reboots? Currently I have to manually enable it in RW-Everything every time i sleep/reset the machine (it doesn't take very long, but it's still a bother).

Is there a way to set these before windows loads without breaking the bootloader or current installation, and have them stick?

I use DIY eGPU Setup 1.30 to enable ASPM settings in a pre-boot environment like discussed at https://www.techinferno.com/index.php?/forums/topic/1982-125-hp-elitebook-2570p-owners-lounge/#comment-35271

2 hours ago, 3RYL said:

Does anyone here noticed when Bplus suddenly raised the pricing for their eGPU adapters?

Just checked. PE4C V3.0 is now listed at US$162. 

A google cache from Oct-25 has it at US$65 .

So yeah, 150% increase in 3 days.

It's US$85 on Amazon-US and ebay-US.

3 hours ago, sangemaru said:

Two updates: Using EXP GDC, i tried out an RX460 and GTX 1060 GPU. As opposed to previous models I'd tried, RX 460 is capable of sustaining x1 2.0 and working just fine indefinitely.

GTX 1060 is capable of sustaning x1 2.0, but occasionally crashes, making it unsuitable for use with the EXP GDC. It seems the choice of videocard matters a lot.

I found AMD drivers have better fault tolerance to signalling errors when I tested a EXP GDC + HD7870. The HD7870 in a EXP GDC would report a x1 2.0 link but would give noticably worse benchmark results when compared to a soldered-cable PE4L 2.1b. imho, it's not worth saving a few dollars for the EXP GDC since it's Gen2 signalling is flakey.  NVidia drivers would just BSOD upon getting an unreliable Gen2 signal requiring downgrading the port to Gen1 for reliability.

@frozentax, your CUDA-Z host-to-device score of 1322MiB/s is he bandwidth x4 2.0 gets https://www.techinferno.com/index.php?/forums/topic/10581-6th-gen-i-core-has-gen3-8gbps-southbridge-pcie-ports/ . Can you pls re-run that? You're looking for ~2800MiB/s for a x4 3.0 link.

theitsage wrote at http://forums.macrumors.com/threads/new-apple-display-with-egpu-integration-in-10-12-1.2008996/

Quote

Upgraded my Mac Pro tower to 10.12.1 and found these kext files updated. After a bit digging, there are references of PCI2PCIBridge and AppleDisplay.

AMDRadeon4100.kext has support for Polaris 10 and Polaris 11 GPUs so this is most likely an indicator of new Apple Display with eGPU integration comes Thursday.

AMDRadeonX4100.kext, AppleBacklightExpert.kext, Audio AUUC.kext, IOGraphicsFamily.kext, IONDRVSupport.kext, IOPCIFamily.kext, System.kext were signed off at 4:19am last Wednesday - right about when the invites for the 27th event went out.



It's already possible to build eGPUs with Polaris cards.