Shameless plug also: I am the developers and operators of the free and vendor independent training log » https://www.velohero.com
Like you, I run this training platform as side project. I have been doing this since 2008. No price increase since then (always 19EUR). I come from your neighboring country - Germany. If you see the change that we join forces, feel free to contact me via LinkedIn or email » https://www.nkn-it.de
Thank you very much for your feedback, I am very grateful. The cost change is not related to compute (vCPU and memory). I will have a close look at it in October, because I need to apply the price adjustment to my other project: https://github.com/Cyclenerd/google-cloud-pricing-cost-calcu...
The "Instance Picker" automatically sorts by price per hour. So at the top is always the most affordable VM according to your filter. Everyone has different requirements. I work a lot with SAP and therefore filter a lot by the SAPS benchmark.
To automate this step I provide the SQL export. You can then quickly select the correct machine type.
I built an open source webapp that helps you find the optimal Google Compute Engine (GCE) machine type in the many Google Cloud regions. I hope the app helps you too. Any feedback is very welcome.
Vaguely related. I wrote a script years ago that grabs the cheapest region for each Spot Instance on AWS. Haven't checked it in years but it still seems to work : https://simonpbriggs.co.uk/amazonec2/
What could be missing is the GPU aspect.
Not sure if the data is available, but it is very unclear to me which GPU is available on which machine type.
and also how many, so that I have some chance to get one.
In the Instance Picker in the "More" section you can find the GPU Filter. Additional GPUs can only be assigned to N1 and A2 instances. The whole Accelerator topic is very specific. Maybe I will expand it further if more people find it helpful.
The fire last year at OVH showed us impressively that it is not a good idea to have your data only in one region. So don't do it and stick to multi-region.
Well, that depends. IIRC, OVH's fire hit multiple floors, and each floor was a "zone" - the power, networking, etc were all independent, but they still had a single core dependency - the building itself.
