Word!! He has come a long way though! Although it will be great if he gets as many subs as he deserves, it may mean he won't be able to reply to as many comments as he does, but that's me being selfish lol.

Yeah that's why I subscribed. 😮

They're not even *that* technical, but they're like the perfect level of technical for armchair enthusiasts. He also explains things very well. But moreso, he's not sensationalist at all and actually grasps the things he's saying, so whenever he gets into more of a personal analysis situation, he's generally making good points and not just flying off the handle like you often see with others.

@@maynardburger Thank you, I honestly think you perfectly described what I am going for. I'm not a semi engineer so I actually have less technical knowledge than for example someone like Ian Cutress on his channel. I often need a lot of time and research to fully understand something, but once I do, I think I can explain it very well, because I had to put in the time. If I was a semi engineer I might glance over basic things because I would consider them general knowledge.

I Concur

Glad you like my videos!

Thank you for the tip!

Nvidia is still good for more general processing. But even there, yes, I too expect Nvidia to be badly sidelined.

I tend to agree with you because of the "CUDA entry learning" factor. But keep your eye on RISC-V based Jim Keller and Tenstorrent.

Yeah, it also surprised me during my research. Big Tech is going all in.

Oh yeah!!!

They all do pay for the best hardware, TSMC 3nm, 4nm. NVDA doesn't manufacture the best hardware themselves. In the big picture, I'm pretty sure every tech giant does not want to be captive to any company if they can avoid it, so it makes sense they'll have their own internal solutions as well. And if it's a matter of having the tech or not, then it's understandable the costs are secondary. But if there are cheaper solutions, I'm sure shareholders and customers do care. After all, the costs just get passed to the customer and then their customer and so on, so in the end are you willing to pay more?

"Cost be damned" No...not really. If you can make a chip of just SIMILAR performance but for a much cheaper price you will use your own stuff instead of buying NVIDIA. Of course research also eats up loads of budget but an H100 reportedly only costs around $3200 to produce. And since a lot of these non-NVIDIA chips are smaller and on cheaper processes, they are even cheaper. Maybe only $2000 or $2500. That is a gigantic difference to the $45,000 that NVIDIA is demanding for their H100. And so even if Google, Microsoft etc. produce their own chip that has a quarter of the performance at half the wattage it would STILL be mich cheaper to manufacture four of your own chips to match one of NVIDIAs chips, even if those 4 chips use 2x the power of that one NVIDIA chips because you're literally saving tens of thousands of dollars. I mean just do the math. Let's say Google has a chip with the aforementioned performance. ¼ of the performance for ½ the wattage. 4 of those chips then provide the power of one NVIDIA chip. Same performance but by using 1400W instead of 700W. Too expensive in the long run? No. Cause even if they run 24/7, it's 0.7Kw × 24hrs = 16.8Kw/day. 16.8 × 365 = 6132Kwh. At a price of $0.1/Kwh that's ~$600 more in electricity. But they saved 20 to 30K by not buying NVIDIA. So even after 10 years the extra power draw would be MUCH cheaper than just buying one NVIDIA chip. It really does make sense for them. Even if THEIR hardware isn't the best, which can also change because all of these companies are trillion dollar mega corporations.

I made time lol. Great video! I'm looking forward to seeing what happens to Nvidia. After their ____ show of a comtech keynote, amd their sad cash grab GPU launches, it seems like they are probably going down. I'm looking forward to what AMD does especially with their APUs and hopefully applying AI chiplets to them potentially making excellent mini/SB computers that can be used for automation/AI applications in a super small form factor and low energy use! I would hate for all AI hardware to be proprietary as I would rather design my own than allow Google, Apple, or worse yet FB to have access to my systems just so I can use the tech. I am also hopeful that their APUs will essentially kill the market for consoles like Xbox and PS! Once that happens we will see an explosion of awesome games with incredible innovations. Especially when AI gets better at helping with that! UE5.2 has some incredible PGC tools now along with easy to use face/motion capture tools and with systems like CGTP games cpuld literally generate content in real time based off of what you have done in a game up to that point and informed by the lore/parameters of the game world. As long as we are not fighting skynet before that can happen lol. Anyway thanks!! Have a great week!

Right now, Nvidia is clearly on top, but it could change rapidly over the next couple of years. The entire AI space is very volatile right now, lots of possibilities for disruption.

@@theminer49erz AMD has good GPUs however they’re constrained by their software or poorly understood software.

Gaudi2 seems very promising, but I haven't looked at it in-depth enough. Gaudi 3 is offering a huge jump in performance too.

It’s gonna be a tough fight for Nvidia. Very interesting times ahead of us.

I hope not,as those are all Malthusians with Trillions to spend on our end.

Question is when...

Thank you! Interestingly, there are not a lot of videos covering this topic.

You are right, but a lot more ppl know Bill Gates.

Then the next AI model would include these behaviors ;)

@@HighYield then only tons of galllium to end it.

I know there are a lot more custom AI chips around, but this is the first time I've heard about Graphcore. Sounds super interesting (just checked out their website).

@@HighYield They are 700 ME funded, and it is a real 3D chip. Next move for them will be to put it on a interposer close to HBM. Yes it a great technology that can potentially do very well. My BR

For training and experimentation Nvidia is the way but when you need to deploy those much model at large scale Nvidia hardware becomes very expensive. Data scientific at these tech companies use Nvidia hardware and software stack to train their models because its easy and fast for doing large number of experiments. But once the model is ready to be deploy it will be deployed in ASIC made only for inference.

@@sagyamthapa Dead Wrong,those at the top funding all this have more $$$$ then God and every atheist that will ever live. Money is no object,think way way bigger

@@mannyc19 They don't have all that money by wasting it. The OP who thinks efficiency (power usage) isn't a driving factor is, like you, very ignorant about how decisions like these get made. Stock valuation is everything. There isn't any actual physical worth. They don't have "money" (which is fiat regardless) .. they have VALUE. Perceived value. You have a lot to learn.

Oh I missed out on many more custom AI chips. There will be more videos on AI hardware in the future :)

No. The 0.01% dont want to sell us stuff anymore,they want most of us dead. Hence the Great Reset and Carbon Zero insanity and obsessions with self driving cars(Food Delivery,think Truckers Strike....) and A.I.

We basically already live in that world. Though I agree that recommendation engines are really just the most lousy and disgusting use of AI. That said, people really need to start learning to curb their consumerist habits, be it products or services. We are still ultimately responsible for our own behavior.

What’s your solution? Which company are you talking about?