MacBook pro egpu for deep learning only
 
Notifications
Clear all

MacBook pro egpu for deep learning only  

 of  2
  RSS

cralax
(@cralax)
Active Member
Joined: 3 years ago
 

Hi,

I currently own a MBP 13'' 2016 with touch bar.

I plan to buy a Nvidia 1080 ti and an akitio node.

I don't need to improve my screen resolution, it is just for computing neural networks.

Does this configuration work for deep learning?

Shall I connect a screen to make egpu works properly?

Thank 

 

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: 3 years ago
 

The use of a ghost adapter makes it unnecessary to connect an external LCD. See this article for more info:

https://egpu.io/how-to-egpu-accelerated-internal-display-macos/

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
kotlos
(@kotlos)
Trusted Member
Joined: 3 years ago
 

CUDA works for many programs (Premiere, Davinci resolve etc) without an external monitor/ghost adapter, so I would guess you should be fine. 

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

.

ReplyQuote
cralax
(@cralax)
Active Member
Joined: 3 years ago
 

 Last question, Is there a GTX 1080 Ti brand (MSI, Asus...) better than others? I don't need a lot of display ports (maybe one HDMI) since I want to use it for tensorflow, CUDA or Pytorch...

 

Thank

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

.

ReplyQuote
kotlos
(@kotlos)
Trusted Member
Joined: 3 years ago
 

Get one with very good cooling since you will be maxing out the GPU for extended periods of time, and throttling is a major issue. We use blowout reference design because we stack multiple of those, but with the Node there is lateral opening so other designs might work even better. Unless the GPU can maintain the temperatures low, overclocking will not help. 

If you use a reference/blowout design make sure Node's fan is directing the air inside the box whereas if you use a dual/triple fan design it should direct the air outside the box. 

A hybrid solution would work best for cooling, but then again pumps fail faster than fans. 

Also being able to maximize the fan speed is a bonus! 

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

.

ReplyQuote
mcarver316
(@mcarver316)
Estimable Member
Joined: 3 years ago
 
Posted by: cralax

 

 Last question, Is there a GTX 1080 Ti brand (MSI, Asus...) better than others? I don't need a lot of display ports (maybe one HDMI) since I want to use it for tensorflow, CUDA or Pytorch...

 

Thank

   

Just get the Founders Edition from any vendor. It is basically the NVIDIA reference card with third party branding.

If you must go with the non-founders version with extra bells and whistles, make sure the card conform to the double width pci slot specification- otherwise, you may have to leave the case open to accommodate wider cards.

2015 17" MSI GT73VR 6RF Titan (GTX1070-SLI) [6th,4C,H] + GTX 1080 Ti x4 @ 32Gbps-TB3 (AKiTiO Node) + Win10 [build link]  

ReplyQuote
cralax
(@cralax)
Active Member
Joined: 3 years ago
 
Posted by: mcarver316

 

Just get the Founders Edition from any vendor. It is basically the NVIDIA reference card with third party branding.

If you must go with the non-founders version with extra bells and whistles, make sure the card conform to the double width pci slot specification- otherwise, you may have to leave the case open to accommodate wider cards.

   

It is my first GPU... Among the founder edition version, is there a brand better than another (MSI or ASUS or...)?

 

Thanks

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

.

ReplyQuote
mcarver316
(@mcarver316)
Estimable Member
Joined: 3 years ago
 

Founders edition is founders edition. It is basically an NVIDIA card with ASUS, MSI, Gigabyte, etc. label on it.

http://www.theverge.com/circuitbreaker/2016/5/12/11662346/nvidia-founders-edition-geforce-gtx-explained

http://www.pcgamer.com/what-exactly-is-nvidias-gtx-1080-founders-edition/

Get it from amazon, newegg or bhphoto, or someone with equally good return policy. Don't pay more than the list price.

2015 17" MSI GT73VR 6RF Titan (GTX1070-SLI) [6th,4C,H] + GTX 1080 Ti x4 @ 32Gbps-TB3 (AKiTiO Node) + Win10 [build link]  

ReplyQuote
eau
 eau
(@eau)
New Member
Joined: 2 years ago
 

Hello,

I am planning roughly the same and ordering a sonet enclosure or a razer core v2 with an Nvidia 1080Ti founder edition for number crunching / deep learning / mining tests / and other stuff like that, I will be running on OS/x & Linux mainly.

I am completely new to eGPUs, so questions might sound stupid to you guys, I searched a bit, but could not find a clear answer.

- how often does a card improves/become obsolete? (question is more like should I wait a little bit to get the very latest successor of the 1080Ti as it seems to be the top card these days or can I go blindly for it).
- does the physical size of GPU cards change a lot, meaning my enclosure won't become obsolete when I want to upgrade the card in 2 year time? (not really sure i will have the need really but just to understand)
- are they any enclosures able to host/power/drive 2 x 1080Ti cards? is there any advantages thinking/doing such (doubling the number of card, etc... if possible)

Thanks

Cheers.

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

.

ReplyQuote
Flint Ironstag
(@flint-ironstag)
Estimable Member
Joined: 3 years ago
 

There's always something new around the corner.  That said, 10xx series has been out since 2016 and a replacement is due soon.  If you can make money with it now, buy it now.

Stay with reference designs and you'll be fine size-wise with your enclosure.

I don't care for the Netstor and other designs - if you want multiple-GPU enclosure, OSS CUBE is a known quantity.

https://www.onestopsystems.com/product/cube2-metal-expansion-enclosure-5-pcie-slots

Not cheap, but will likely outlive several laptops - change to a faster TB interface when available if needed.  Lots of computational stuff is not hampered at all by Thunderbolt's limited bandwidth, and will run at full speed - depends on your workload.

MP 6,1 | 4c | d700
MP 6,1 | 6c | d500


ReplyQuote
 of  2