Wow, that does seem crazy.
Maybe it's an Umbo typo for $5?
Deep Learning Startup Umbo Raises Prices 733% For Analytics
Published Oct 02, 2017 14:30 PM
While the industry overall trend is racing to the bottom on price, deep learning startup Umbo is taking the alternative approach, raising prices on their core product 733%.
We review Umbo's newest pricing, the increases in annual costs a typical Umbo user would expect, and the impact this is likely to have on their growth.
Umbo ***** *******
****'* ****** ***** ***** ***** * significant ******** ** *** ******* **** of ***** *********. ***** **** ******* options **** **** **** **********, ****** users ********** ****** *** *** ******* resolution ******* **** ******** *** * reduction ** ******* *** **** *********.
*** ********* ***** ******** ****** **** our **** ******* **** ***** ****, *** ******* ******* ** ***** most ****** ***** *****:
**** *** ***** ******* *** ******* the **** ******* *******, **** *** lower-resolution **** ****** ********** **** ******* price ******.
Camera ***** *******
**** *** *** **** ****** ******* from $*** ** $***, ******* * new ****** ***** *** ***** **** an **** ** $***. *********** ****'* consumer *****, ***** ****** *** ********** high *** * ****** (*.*.:**** *** *******@ $***, ******* **** @ $***, **** *** @ $***, etc.)
Additional ******* *******
**** *** ***** ******* ***** ****** options ** ***** ** * **** of ********* ($*/**.), ******* ***** ***** be **** ****** ** **** **** on *** ******** ** **** *** such ***** *********. ***** ******* ******* exist, such ** * **** ** ******* for $***/**., ** * ***** ** storage *** $***/**.
New ****** *** "********** **********"
****'*********** ********** ********, ********** ******** **** *** cameras, *** *** ** ****** **** of $** ($*/**.) *** **** ******:
*******, ********** ** ******** **** *** new $**/** ********* *******.
Analytics ******* *******
***** **** ***** *******, *** ***** storage, ***** *** ************** *** **** their ********* *****. *** ******* *** not ********** ** *** ***** ******, and ***** ***** ******* ****** *** same ******** ************* ** ***** ***** storage *********.
Impact ** ******* ****
* ******* **** ****, *********** ** their ********* ********, ***** ****** *** an ******** ** **% ** ***** cost *** *** ***** **** ** operation, *** ***% ******** *** ******* years:
Umbo ** *******
**** ********** *** *** ******* ** our ******* ** ***** ********* *******.
Priced *** ** ******
**** ******* * ******* ***** *** a ******** ******* ******, *** *** have essentially ****** ********** *** ** **** market **** ***** *** *******. ** $*** per **** (**** *** ******* $***-$*** camera *****), ****'* ********* *** **** **** expensive **** ********** *********. *******, **** ********* like ******** ******** ********* ** * number ** ****** **** ** **** ****** $1,000 (*** ** ****** ************), *** Hikvision ********* ***** **** ******** ***, ****'* ****** ** ******** ** be *********** ** * ********** ********, especially **** *** **** ** *** integrations ** **********.
Comments (5)
U
Undisclosed #1
Oct 02, 2017U
Undisclosed #2
Oct 04, 2017My guess is that their costs for running the deep-learning analysis are significant, and that contributes to their pricing. Speaking to Umbo at IFSEC earlier this year they said that they send a 180p stream into the cloud where the analysis is done on nVidia GPU servers.
As you can see from AWS pricing, GPU instances are roughly 5x the cost of their equivalent CPU offerings.
U
Undisclosed #1
Oct 04, 2017As you can see from AWS pricing, GPU instances are roughly 5x the cost of their equivalent CPU offerings.
Are they though?
U
Undisclosed #2
Oct 04, 2017Well, I suppose it depends which lines you compare. E.g. m4.xlarge vs p2.xlarge is ~5x.
In any case, the cost vs CPU instances may be a moot point since I would still expect the cloud compute time to be a significant cost, as the analysis has to run constantly.
The cheapest one you picked out above is going to cost ~$850/month. From what I have seen of the Umbo system I would not expect their inference engine to be especially lightweight.
Obviously it comes down to how many inference engines they can run per GPU: I would expect it to be a significant cost driver.
Ultimately I think the 'right' solution is moving the DNN computation silicon out to the edge - like Movidius, Synopsis (and ARM to a lesser degree), etc are doing. Aside from the cloud compute requirements, there is the bandwidth issue of constantly streaming 180p into the cloud.
U
Undisclosed #1
Oct 04, 2017Well, I suppose it depends which lines you compare. E.g. m4.xlarge vs p2.xlarge is ~5x.
You made the blanket statement:
As you can see from AWS pricing, GPU instances are roughly 5x the cost of their equivalent CPU offerings.
would you like to change it to:
As you can see from AWS pricing, GPU instances are roughly 1.5x to 5x the cost of their equivalent CPU offerings.
Also to note is elastic use, (depending on the way their interface works)
