EXP GDC Beast V8.0 "Mini PCI-E Version" vs ADT R63SG for mPCIe 2.0x1 vs USB3 PCI...
 
Notifications
Clear all

EXP GDC Beast V8.0 "Mini PCI-E Version" vs ADT R63SG for mPCIe 2.0x1 vs USB3 PCI-E Riser  

  RSS

yus
 yus
(@yus)
New Member
Joined: 1 year ago
 

I have an older asus g75vw which is not gpu upgradable
GTX 670M, i7-3610QM
https://www.notebookcheck.net/Update-Review-Asus-G75VW-T1040V-Notebook.76032.0.html
https://rog.asus.com/forum/showthread.php?15094-Is-it-possible-to-change-graphics-card-on-a-G75VW
NO Thunderbolt. I use it to run tensorflow on cpu. I need to make it run the same thing on GPU as well, run fast.

To connect egpu I have two ugly options, both with a sticking peace of cable:
1) costs less: EXP GDC Beast Mini PCI-E V8.0 for mPCIe Slot: PCIe 2.0x1
http://exp-gdc.blogspot.com/2015/12/exp-gdc-beast-laptop-external.html
https://www.aliexpress.com/item/32959496619.html?spm=2114.search0604.3.96.1642656bzESXGY
probably works on
https://rog.asus.com/forum/showthread.php?88324-Guide-G75VW-eGPU-installation-with-gtx-750ti
2) costs more: ADT R63SG for mPCIe ( = mini pcie express) Slot: PCIe 2.0x1
https://www.adt.link/product/R43SG.html
https://www.aliexpress.com/item/32846824601.html?spm=a2g0s.8937460.0.0.3b132e0eU0cHs2

I am not that much sure about the transfer rate. If in my case the data is transferred most of the time once to the gpu when the calculation starts and then back when it is done then it is ok as for cripo-mining. If in fact it requires the 2-way communication, then I will be limited by PCIe 2.0x1 2-way transfer rate.

QUESTIONS

Q1) Does anybody have experience with the both GDC Beast and ADT R63SG?
Are they stable? I need this setup being able to run days non stop.
Is ADT R63SG more stable? It seems that HDMI cable/plug might be not so good
for mPCIe signals?

Q2) GDC Beast uses an HDMI plug/slot on the gpu side. What kind
of plug/slot ADT R63SG uses?

Q3) Are they compatible with the modern cards, say
at least gtx 1080, at most rtx 2080 ti? I mean have anybody
really installed and used these for some time, a month?

Q4) Is there any solution which is less ugly on the laptop side?
Apart from the cost, why don't use a plug/socket on laptop side?
why don't use the usb3.0 plug/socket which costs nothing like in
PCIE 1x To USB 3.0 Cable for PCI-E Riser Card
https://www.aliexpress.com/item/32956086135.html
or in


More expensive industry like cable solutions that cost from $100 surely exist as well
https://www.dolphinics.com/products/PCI_Express_iPass_cables.html

Q5) From the videos it looks like egpu works through the USB 3.0 Cable/Socket with degradation of being almost 3  slower probably when the link requires 2 way communication. USB3 PCI-E. Similarly if the built in (not the external) gpu is connected to the monitor (2-way data transfer) there is significantly more degradation then when the monitor is attached to the external gpu (mostly 1-way).  Can anybody clarify from the real experience?

Q6) Riser does exist for the laptop socket (mPCIe 2.0x1). 
https://www.aliexpress.com/item/32834780960.html
Any experience? By the way, since it is no chip "No driver needed", is it just a cable and a socket from the usb 3 with the PCIe 2.0 x1 bus lane inside? Is it really anyhow different from both GDC Beast and ADT R63SG? Are they hardly different at all apart from the cables and the sockets?
They all use the same PCIe 2.0 x1 bus lane inside, so they cannot have Transfer
rate higher then 5.0 GT/s (Throughput < 984.6 MB/s)
https://en.wikipedia.org/wiki/PCI_Express#History_and_revisions
If they are basically the same thing then USB3 PCI-E Riser can be maid the least ugly as I just have to install an additional usb3 socket through the plastic and can probably use just any usb 3 cable. For G75VW it looks like a good solution as the x1 pci wifi adapter socket is the only available simple access to the pci bus.

thanks in advance for your kind advice

<<<Addition June 27, 2019>>>
It is really possible to place tensorflow computations entirely on one/many selected gpu-s:
https://www.tensorflow.org/guide/distribute_strategy#centralstoragestrategy
and
https://www.tensorflow.org/guide/using_gpu#manual_device_placement
This means that the tensorflow computation can fully benefit from an external  egpu even on PCIe 2.0x1. Only the one-time task placement/input at the start and output at the end will be the subject of the PCIe 2.0x1 transfer rate. So for tensorflow it is possible to benefit from the most advanced gpu-s with no performance degradation.

This topic was modified 1 year ago

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 4 years ago
 

Please review the Buyer Guide that answers many of your questions:

https://egpu.io/external-gpu-buyers-guide-2019/#mPCIe-interface

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table

 
2015 15" Dell Precision 7510 (Q M1000M) [6th,4C,H] + GTX 1080 Ti @32Gbps-M2 (ADT-Link R43SG) + Win10 1803 [build link]  


ReplyQuote
yus
 yus
(@yus)
New Member
Joined: 1 year ago
 

@ nando4
Thanks. Still many things are not clear. Especially I would love somebody with machine learning, especially tensorflow experience clarify the degradation point on the pci 2.0 x1. I will probably build a dedicated server just for machine learning, but I am really interested in it in terms of the things I do for work. For the time being what I have is enough, but a 10x bust from egpu would be nice to have =)

This post was modified 1 year ago

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote