Yes, the M-series chips effectively use several "channels" of RAM (depending on the tier/size of chip) while most desktop parts, including the 9950x, are dual-channel. You get 51.2 GB/s of bandwidth per channel of DDR5-6400.
You can get 8-RAM-channel motherboards and CPUs and have 400 GB/s of DDR5 too, but you pay a price for the modularity and capacity over it all being integrated and soldered. DIMMs will also have worse latency than soldered chips and have a max clock speed penalty due to signal degradation at the copper contacts. A Threadripper Pro 9955WX is $1649, a WRX90 motherboard is around $1200, and 8x16GB sticks of DDR5 RDIMMS is around $1200, $2300 for 8x32GB, $3700 for 8x64GB sticks, $6000 for 8x96GB.
That sounds like hundreds of 40mm or 80mm fans running at 5000rpm. You can get a lot more airflow with less wasted power by using larger slower fans... and they won't wear out as fast.
So you're a programmer but you've never assigned a number to a variable or written any math operations? Do you just do string translations or something?
Any sort of business that might want to serve from a customized LLM at scale and doesn't need the smartest model possible, or hobbyist/researcher experiments. If you can get an agentic framework to work on a problem with a local model, it'll almost certainly work just as well on a cloud model. Again, speaking mostly people to already have a xx90 class GPU sitting around. Smoke 'em if you've got 'em. If you don't have a 3090/4090/5090 already, and don't care about privacy, then just enjoy how the improvements in local models are driving down the price per token of non-bleeding-edge cloud models.
With quantization-aware-training techniques, q4 models are less than 1% off from bf16 models. And yes, if your use case hinges on the very latest and largest cloud-scale models, there are things they can do the local ones just can't. But having them spitting tokens 24/7 for you would have you paying off a whole enterprise-scale GPU in a few months, too.
If anyone has a gaming GPU with gobs of VRAM, I highly encourage they experiment with creating long-running local-LLM apps. We need more independent tinkering in this space.
Lots of people already have RTX 3090/4090/5090 for gaming and they can run 30b-class models at 40+ tok/sec. There is a huge field of models and finetunes of this size on huggingface. They are a little bit dumber than the big cloud models but not by much. And being able to run them 24/7 for just the price of electricity (and the privacy) is a big pull.
If it's got 2 P-cores, it would be more like an updated core 3 100u, it'll likely have 20 rather than 9 pcie lanes, dual channel ram, thunderbolt, etc, and be priced like the nuc 15 pro, around $400.
He has an interest in getting good devs hyped on S&box, but he's being pretty open ended and honest here. Want to be a game dev? Make a game and ship it... all of the resources for the in between steps are out there and it's only getting easier, but for everyone else just as much as you. I grew up in the gmod mini-modding scene and it's still one of the most impactful parts of my life. First there's editing existing scripts, then there's recognizing what people (or yourself) would like that doesn't already exist, then there's the grunt work of turning it from idea into reality, then there's releasing it with your name attached. It's very rewarding. Still working on releasing something with a price tag attached but this post a good reminder and motivation.
And yeah, the steam workshop was way ahead of its time, and feels like a new, higher tier of sandbox game. I don't think there's any other game that came out around the same time as gmod (2006) that has had comparable momentum and staying power.
The DNC sidelined Bernie like you wouldn't believe as soon as it became clear he had potential to be elected. During one of his April 2016 rallies they livestreamed a "reporter" onsite talking about her drunken adventures the previous night rather than showing his speech, the cheers, and the line around the block to get in, that was going on behind her at the time. AOC made no notable efforts to be genuinely inclusive. Zohran is still trying to reach the caliber to be a contender listed next to the other two names and sounds like the infighting is already coming for him.
These things are only a mile or two wide and at the distance of Jupiter. They require extremely sensitive and high-resolution telescopes to detect. There are probably many more of them that are smaller and further.
You and nabla9 are both the one comparing apples and eggs. 4x more RAM means 4x larger models when everything else is held the same to make a fair comparison.
GN wasn't the first to break the story the 13/14th gen was defective. The thousands and thousands of users experiencing the issues collectively noticed pretty quick. If anything, there was a period where he was saying "We've talked to Intel but we won't say anything yet until they do."
Most laptops have rubber bumpers around the screen to create a small airgap when the screen is closed, and even if you stack several laptops on top of each other, the bottom one's screen won't flex enough to touch the keyboard. Maybe with the more rigid lid, they can enable this?
Even more fundamental than science, there is missing philosophy, both in us regarding these systems, and in the systems themselves. An AGI implemented by an LLM needs to, at the minimum, be able to self-learn by updating its weights, self-finetune, otherwise it quickly hits a wall between its baked-in weights and finite context window. What is the optimal "attention" mechanism for choosing what to self-finetune with, and with what strength, to improve general intelligence? Surely it should focus on reliable academics, but which academics are reliable? How can we reliably ensure it studies topics that are "pure knowledge", and who does it choose to be, if we assume there is some theoretical point where it can autonomously outpace all of the world's best human-based research teams?
When working well, they enable us to offload needing to memorize a wikipedia worth of information and think about higher level problems. We become more intelligent at higher level solutions. Of course people don't know what was written if they were required to "submit an essay" where the main grade is whether or not they submitted one and the topic may have been one not interesting to them. Ask people to write essays about things they're truly, honestly interested in, and people who have access to an LLM are likely able to enrich their knowledge faster than those without.