Let me rephrase slightly since the footnote picture covers it - the current version, which was released in January, is seemingly still ahead of everyone.
I'm glad they're basically caught up though, there isn't really an enormous gap. Bring on...
I think they are very misleading, for example:
Running Microsoft Copilot Locally Would Require At least 40 AI TOPs
From the official documentation, they always prudently always talk run part of the copilot locally (we can imagine stuff like...
I think the 1106 preview used here is the turbo version:
https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo
gpt-4-1106-preview: GPT-4 Turbo model featuring improved instruction following
Considering GPT-4 age and who being...
From my very limited understanding it would be quite the challenge memory wise to work on the same context doing inference and for a lot of task the previous token is an input for the next calculation, parallelism making more sense to handle many...
One of the issue is that it depend on the inference model precision, 4bits-8bits-16bits will have on some hardware different value has some take advantage others don't.
If they mean 8bits int TOPS, to give clue of older hardware capability...
windows+E followed by alt+d, there a lot of classicsm windows+R, or more modern like the HDR shortcut as well, the virtual desktop management, snapping windows to the side or making them full screen, etc...
Also running those datacenter with current inference cost per token must cost Microsoft a little fortune right (free bing image for example), turning all those cost back to consumer would be nice. They will talk about lower latency for the users...
I mean, sure it's cool, but "runs quake at 60fps" means it is performing at like voodoo1 levels of performance....
Might as well just emulate a GPU in software.
I found this incredibly interesting, I know there's a few other open source ones I see bouncing around in Phoronix forums, but this seems to be the furthest I've seen one.
https://www.furygpu.com/
New open source GPU is free to all - FuryGPU...
Or playing Diablo/starcraft on battlenet in the 90s, there a lot of (steam, the maker of the games and so on) that do a lot of what those console provide for "free".
That is quite the statement if I understand it, you disagree that the...
10 years ago a Gtx 780 that launched ~$865 in today dollar yes.
But that the point if 380mm of TSMC special NVIDIA node that come with 16 GDDR6X on a mini computer with a giant fancy cooler costing $1000 is ridiculous, what does it say with...
I doubt parents will hesitate between the steam deck and the switch 2 for their kids.
One possible reason is how crazy well the Switch (consoles and games) are still selling, the very old Switch sold more than twice as many units than the 2020...
I think it is mostly all about good exclusive, nice gamepass price and service, nice controller, etc... can play in the margin, but the switch sell because of Zelda and what not, when you are behind trying to win by having a little better Call of...
Combined with the policy that if you want to release on Xbox you need to have a S working version, that seem to be what they say in the clip, because if you have a PC and PS5 version, doing the Xbox-X is probably not that big of a deal.
The sequel of a 2011 Xbox 360 game that had a teaser released by a company in 2018 is so advanced in 2024 that it is possible for people to launch it and play it internally.... Maybe the early speculation of a 2026 date will work....