I was thinking about i7-5930k initially, but I ended up with i7-5820k since it is much cheaper and the number of PCIe lanes should be enough for at least 3 GPUs, and for some models the bandwidth is enough for 4 GPUs. I examined the Tensorflow 1.0 multiple GPU training examples with Inception V3, and calculated that 8 PCIe lanes per GPU (with 8GB memory) offer sufficient bandwidth for CNN type of models.
1 Like
Quite a few things I’d be interested in working on:
- Kaggle Fisheries competition (@even/@aifish - I would be interested in working with you on this)
- Some kind of creative or artistic project using GANs
- Financial time series analysis (option chains in particular). I have some interesting ideas here.
I have other more ambitious ideas, but I want to keep it relatively simple (and focused on this course for now).
Cheers,
James
For putting you in teams, please put your forum username and timezone in this spreadsheet. On Thursday afternoon (pacific time) I’ll sort by time-zone and assign you to groups based on people in similar time-zones.
If some of you have already decided to be in a group together, please note that in the spreadsheet.
1 Like
I am taking part in the Kaggle Fishery competition. Anybody would like to team up?
1 Like
Anyone working on the Data Science Bowl? Getting started now that I have some GPUs. 
2 Likes
@davecg I was trying the data set. So far I was able to do some pre-processing and tried few ideas. If you want to discuss in a separate thread this one might be a good place lung-cancer-thread.
Guys i was just thinking on a fun project to work on.
People want personalized smilies. smilies that look like them
We can use style transfer and Deep learning to draw customized smilies. Anyone else interested in this?
2 Likes
I’ve just finished assigning everyone in the spreadsheet to groups now. Can one of you in each group please create a private chat for your group, using the usernames in the spreadsheet?
Hey Kishore,
I reside in Indira Nagar. Would love to meetup and discuss ideas and form a study group here.
A 7700K w/ M270 chipset can do 3 GPUs.
Pascal (nVidia 10xx cards) only support 2 GPUs in SLI (this is for gaming) so more and more motherboards may stop focusing on 3 and 4 GPU support going forward. The MSI M270 M5 motherboard I got can do 3 GPU’s at 8x8x4x. The difference between 16x and 8x is less than 1% but I have not tested 4x, so the third card will be reduced but I don’t think it will be a huge difference.
The 16 lane bus support on the 7700 isn’t a huge limitation for 99% of people and you will be better off with 2x1080Ti or even 2x1080 over 4x1070.
1 Like
Great!
Yes, we could also do it on a CPU for sure even if it will be a bit slow compared to GPUs.
I have runned the lesson 8 notebooks on my Intel core i3-2350M CPU, with 6Go of RAM.
Windows 10 as host, and the setup through a Ubuntu Virtual Machine on VMware Workstation 12 Player.
As I said all works fine even if a bit slow, so it can do the job.
Thanks for letting us know. Have you tried Ubuntu Bash for Windows? I would have expected it to be faster than a VM.
Thank you @jeremy,
For my installation on the Windows 10 I have been limited by pytorch which does not run on Windows (even if tensorflow and keras work well on through Anaconda 3). This is the reason why I used a VM.
For the Ubuntu bash from the Wiki, as it was designed for python 2.7, I have adapted it to python 3 using the setup info you gave on the setup thread.
So all run well on my VM.
I have also tried a shared Docker on Windows, and it work on, I will propose a post on later as my contribution to the Wiki.
ok. But it looks uncompleted. I cant find my name
+1 if you are not using direct hardware support (like GPU), Ubuntu Bash for Windows is fantastic and is the best option for Windows. It’s is pure linux binaries, not emulated. You just can’t do anything that uses direct access to hardware like GPU or Network card.
In an earlier message on this thread I asked everyone to add their name and timezone. If you add it now, we can create a group for people that missed the last request.
1 Like
Done. Sorry, I overlooked this earlier post.
re: study project: I am in the kaggle DSB and the fisheries one. Other than that anything crazy or creative would do.
Perhaps building a cool web app for style transfer? A beatbox companion based on wavenet? Right on.
As a biologist, I also would be happy to work on a biological / biomedical data set.