Sorry, I'm talking about GitLab.com. I don't use it, but anytime I visit the site suffer so much with it that the experience makes me think the worst from GitLab in general.

GitLab.com is not representative of the speed of GitLab self hosted. If anything it is making self hosted faster by surfacing scaling problems.

While that may be true, don't you see a marketing problem here? Most people would assume that Gitlab.com is representative of the speed and UX of Gitlab self hosted. It's almost like your demo site to be honest.

Surely anyone considering self-hosting it would realise it's performance was going to depend on what they hosted it on?

I don't think I would assume the free 'demo site' would be hosted on the beefiest offering available such that I could only do worse by self-hosting.

Without actually trying it out in a self-hosted scenario (and this is a non-trivial time investment to do), you don't know if that situation is because of the scale or some constant factors. The slowness could be because of thousands of projects hosted, or because of stupid interaction with git which slows down every single page. Until you try it for yourself, it could be either. (it seems to be the former mostly)

The load is probably more important than the server capabilities: I imagine gitlab.com is slow due to the sheer volume of access and the number of projects being hosted. Self-hosted instances aren't likely to a) allow random strangers to host stuff and b) attract a large population of such strangers.

Agree on this one. I know that hosted GL works much better, yet I have to remind myself about that every time in discussions, because I mostly used the hosted version for private projects.

My company tried to migrate to Gitlab.com and had to migrate back off after just a few days because of how awfully slow it is. It has soured our view even of the self-hosted option, and since we would probably need some of the EE features we don't really have a desire to try it out.

Installed.