Daniel Reetz, the founder of the DIY Book Scanner community, has recently started making videos of prototyping and shop tips. If you are tinkering with a book scanner (or any other project) in your home shop, these tips will come in handy. https://www.youtube.com/channel/UCn0gq8 ... g_8K1nfInQ

scan tailor experimental - best video card?

Scan Tailor specific announcements, releases, workflows, tips, etc. NO FEATURE REQUESTS IN THIS FORUM, please.
Post Reply
Adam32
Posts: 28
Joined: 28 Jun 2014, 08:55
Number of books owned: 500
Country: United Kingdom

scan tailor experimental - best video card?

Post by Adam32 » 12 Oct 2016, 04:49

I will be building a system just to scan books and wondered what the best video card was for scantailor experimental? I ask this as I have read that scantailor experimental can use the GPU to speed up processing. With this in mind what is the best video card to buy in terms of cost / performance?

Tulon
Posts: 687
Joined: 03 Oct 2009, 06:13
Number of books owned: 0
Location: London, UK
Contact:

Re: scan tailor experimental - best video card?

Post by Tulon » 14 Oct 2016, 05:03

In image processing applications, GPU memory bandwidth tends to be the bottleneck. So, your performance / cost ratio may well be the "memory bandwidth / cost" ratio. For memory bandwidth of a particular graphics card, consult Wikipedia:
https://en.wikipedia.org/wiki/List_of_N ... sing_units
https://en.wikipedia.org/wiki/List_of_A ... sing_units

Still, note that as far as Scan Tailor Experimental performance goes, investing in a good CPU may well beat investing into a good GPU.
Scan Tailor experimental doesn't output 96 DPI images. It's just what your software shows when DPI information is missing. Usually what you get is input DPI times the resolution enhancement factor.

Adam32
Posts: 28
Joined: 28 Jun 2014, 08:55
Number of books owned: 500
Country: United Kingdom

Re: scan tailor experimental - best video card?

Post by Adam32 » 14 Oct 2016, 22:46

Tulon wrote:In image processing applications, GPU memory bandwidth tends to be the bottleneck. So, your performance / cost ratio may well be the "memory bandwidth / cost" ratio. For memory bandwidth of a particular graphics card, consult Wikipedia:
https://en.wikipedia.org/wiki/List_of_N ... sing_units
https://en.wikipedia.org/wiki/List_of_A ... sing_units

Still, note that as far as Scan Tailor Experimental performance goes, investing in a good CPU may well beat investing into a good GPU.
Hi Tulon, thanks for your reply. I was already aware of those lists on Wikipedia, but I was kind of hoping to get some insight into the real world experience of members. Also any benchmarks that people had to compare.

As the creator of scantailor, I would be really interested to hear what card you use, or what you would personally pick if you were building a system today?

Tulon
Posts: 687
Joined: 03 Oct 2009, 06:13
Number of books owned: 0
Location: London, UK
Contact:

Re: scan tailor experimental - best video card?

Post by Tulon » 15 Oct 2016, 08:51

I have a laptop with GeForce 760m graphics card. From what I remember, enabling OpenCL on that card boosts performance of the Output stage by about two times. Other stages (apart from dewarping on the Deskew stage) gain very little.

There are some actual benchmarks (not from me though) on the "Scan Tailor Experimental" thread.

It's hard for me to recommend a particular card. If I was choosing a desktop card for my purposes (not strictly for Scan Tailor), I would probably go for GeForce 1060, which is a pretty good all-rounder. On the other hand, Radeon RX 480 is slightly cheaper while having higher memory bandwidth (particularly the 8GB version).
From a purely theoretical point of view, Radeon R9 Fury / Radeon R9 Nano should beat even GeForce 1080 due to higher memory bandwidth.
Scan Tailor experimental doesn't output 96 DPI images. It's just what your software shows when DPI information is missing. Usually what you get is input DPI times the resolution enhancement factor.

Post Reply