I had many conversations with a former boss about the Azure sales team. They would come in, say they can do it cheaper, simpler and better — he was immediately convinced.
I would do a calculation based on their public price plan and come up with a 5-10x price compared to the bare metal OVH solution that perfectly fit our use case. I would then ask the sales team where I made a mistake in my calculation and hear nothing back.
A few months later, they would come back with the same pitch and the whole process would repeat...
Via: https://discuss.grapheneos.org/d/24452-eu-might-enforce-goog... which specifically quotes the law that should forbid such approach (Article 6(4) DMA) - so EU initiative and engineers consciously and intentionality breaking EU law in the prototype that is supposed to be replicated later.
For llama-server (and possibly other similar applications) you can specify the number of GPU layers (e.g. `--n-gpu-layers`). By default this is set to run the entire model in VRAM, but you can set it to something like 64 or 32 to get it to use less VRAM. This trades speed as it will need to swap layers in and out of VRAM as it runs, but allows you to run a larger model, larger context, or additional models.
I'm really curious what will happen to this space when Valve releases great open source drivers for the Qualcomm chip they have in the Steam Frame. It might be one of the first, very powerful, GPU accelerated SoC you can buy that has mainline support.
I can image having a very usable ARM linux laptop and tablet as a result of this — maybe even cellphone when the modems get mainlined or used via USB.
> This will allow you to distribute your creations to a limited number of devices without going through the full verification requirements.
How can they count the number of devices you install the app on without being the ones to give a permission to install it?
They took nothing back, they are still putting in place the requirement that Google gives permission to install apps on your phone. They are misleading us about it too which is also terrible.
I made a time sync library over local network that had to be more precise than NTP and used i128 to make sure the i64 math I was doing couldn't overflow.
I32 didn't cover enough time span and f64 has edge cases from the nature of floats. This was for Windows (MACC not GCC) so I had to roll out my own i128.
I'm using niri with two screens at work and it's been very nice. I don't open windows on the side as you suggest but I believe that can be done with custom bindings and/or window rules.
I would do a calculation based on their public price plan and come up with a 5-10x price compared to the bare metal OVH solution that perfectly fit our use case. I would then ask the sales team where I made a mistake in my calculation and hear nothing back.
A few months later, they would come back with the same pitch and the whole process would repeat...
reply