xs83

If you mean CUDA specific then yes. The biggest benefit of these machines over the others is the CUDA ecosystem and tools like cuDF, cuGraph etc

TTFT on a Mac is terrible and only increases as the context increases, thats why many are selling their M3 Ultra 512GB

Now this looks much more interesting! Is the top one input tokens and the second one output tokens?

So 38.54 t/s on 120B? Have you tested filling the context too?

I've been on both sides of the software cycle as a stream lead and as a developer and ultimately I think this is it. Agile gets bastardised as soon as specific reporting structures and deadlines come into place "What do we need to plan for this increment?" etc

A team with a good work ethic and some milestone goals to go to and a "Take from the top" mentality, Kanban is perfect for what needs to be done.

Can't take from the top "because of reasons"? Thats a conversation with the people that particular problem involves and things are re-prioritised.

I'll never understand removing a whole day of productivity for ceremonies to plan being more productive.

Async standups via Slack using Geekbot are absolutely fine, A daily simple async retro for teams on a different timezone also help massively for synchronisation, discipline to update tickets with progress and a push regularly mentality all helps with the sync problem if across timezones!

I like the interface - I will say that it doesnt work when there is an ad-blocker installed (the workspace wont open).

When I am into the app it seems the database wasnt generated for items like the authentication system (which I would expect as it is included and mandatory).

Still more work needed but overall it looks great and I can see the advanced workings you have put in behind the scenes to make it do what it does.

Number one reason is probably * gestures wildly at everything *

I will die on the hill of squash-merge only. It doesnt make the history that confusing having branching visible with some run of the mill standardisation in process.

My biggest issue is that if you have to delay a feature then the sheer amount of conflicts you have to resolve (many of which you had nothing to do with) becomes prohibitive.

I have no problems with rebasing master or a published branch - thats the release managers problem to deal with everything on but I really dont know where this "rebase everything" comes from

How is its laziness? I found Opus to be very quick to curtail output and default into "Add the rest of your code here" type things.

I am a lazy data engineer - I want to prompt it into something I can basically copy and paste

Awesome news - we will eradicate HIV within the a single generation at this rate!

Nice!

I have been a long term Rectangle user on Mac and a WinSplitRevolution for decades on Windows, my biggest preference as a developer is for it to be keyboard centric.

I love the use of a numpad as a positional reference for the screen.

Hotkey + NumPad 6 = right side of screen full Hotkey + NumPad 5 = full screen Hotkey + Numpad 1 = Bottom left Quarter

Repeated presses of these cycle between Half, 3rd and Quarter

This reminds me of the Crysis UI which I still love as a weapon / armor switcher :)

I think the differences is that it isnt weaponised as it is in the US but it is very much valued and expected.

Yeah thats ridiculously expensive for a home lab IMO, the thinkcenters can be picked up for a couple of hundred bucks used and a T400 runs about the same. So $400 for an AI capable home bench would run $1200 for a 3 node cluster - I can live with that.

My 5 node Rpi5 Cluster ran $1500 with NVMe Hats and a PoE hat and cant do any GPU work :(