I just report one scam just now.

Any chance we might get a video on OpenGL? That would be awesome! :)

Shabbat-Shalom.Asé.

Congratulations, think you might be the biggest Nvidia fanboy on KZfaq. Nice job leaving out all Nvidia IP theft and scandals. A GPU is something that existed years before Nvidia existed. Nvidia has a fitting name as their destiny is to enviously strive to be exceptional but never achieve it. By your bad definition CPUs were the first GPUs because they are the first to accelerate 3D graphics via massive parallel processing using MMX in 1996. Also AMD CPU/GPU powers the top10 supercomputers in the world. How is that possible if Nvidia is as good as you say? Nice clickbait thumbnail it seems you know Nvidia's demise is more appealing then some fanboy worshipping Nvidia in an armature documentary.

You failed to mention 3dfx.

It's always a matter of risk vs reward. Great leaders get things done! When he could have continued to get a good steady income being an employee at AMD or at lsi logic , he choose to give up stability and put all his money , experience and reputation on developing a new industry all together which then leads rise to many more brilliant things like the rise of machine learning and how that helped the whole world to make a vaccine for covid so impossibly quickly with such less negatives , able to impact the world on such level is beyond comprehension, every visionary innovations from electricity, battery , transistor, microprocessor, GPU and cuda cores , neural networks etc takes whole credit ! And the visionary leaders who set out to achieve these fantasies into reality deserves more than an employee can fathom !

100% agree with you. It's hard to highlight every different angle in every video, so I like to hone in on one key aspect. If you want to see a video about the power of high quality management, you need to watch this: kzfaq.info/get/bejne/gZeoisSmxpPSfH0.html

@@JohnCooganPlus What I would have liked a bit more of is technical comparisons between what they were doing versus what everybody else were doing. I have some inklings as to what was going on. There was, of course, the invincible juggernaut of the time: 3DFX. We (consumers) all thought the Rivas had no chance in hell against 3DFX Voodo. But there was also another company whose hardware didn't use polygons; only planes, whose intersections were deduced on the fly. They lost the race by cheating; I remember; their driver detected a benchmark running and made special adjustments. It came out, and overnight they went bankrupt. There was also the Pyramid chip project, that was going to be optics/physics based, but no prototypes. And a whole bunch of 3D accelerator cards for those who could not afford a Voodo Monster3D. On the software side there was the competition between Direct3D and OpenGL, but there were other odd players, like the Voxel engine. Then again, the limit back then was on the CPU being able to transform to view frustrum; but then AMD came out with 3DNow! (floating point extension to MMX), and Microsoft supported it, and suddenly everybody was buying AMD's again, and the ball was on the graphics court again. When the Riva128 came out nobody cared initially; then I read that it had 128 full floating point math engines running in parallel, and began to take some interest... And then today we have AMD/ATI, that it seems to me are consistently beating NVidia at their own innovation game. But that's just my superficial overview; I don't really know what exactly made NVidia Riva-128 triumph over 3DFX, for example. Perhaps 3DFX was more "monolithic"?, executing a fixed pipeline?, less flexible? Just guessing. I remember at the time Microsoft had this evangelist guy, forgot his name, that went around the world getting companies on-board with the Direct3D dream. The guy was truly passionate about it, and he was looking for configurability and frank dialogue with hardware above all else. He hated the fact that some of the hardware of his day would say to the OS that it could do A, B and C, but then it was so slow at doing C that software would have done it faster. (EDIT: Remembered the name: Alex St-John.)

@@privateerburrows I remember when 3dfx announced bankruptcy and soon after it was announced that nVidia as buying up some of 3dfx. I think I read that in MaximumPC at the time. My crew couldn't believe it. We knew nVidia was coming on the scene, but we all had 3dfx cards and believed that 3dfx still dominated. Bankruptcy seemed to come out of left field.

@@fearisthemind-killer LOL, you're so right; it was totally overnight, boom!, gone! And I remember a revolutionary audio chip at the time; a 3D positional audio chip that was giving the sound blaster type products a hell of a competition. It had also some hundred or so convolution engines working in parallel to model sound reflections based on game geometry. They had been adopted by the Monster brand, Monster 3D Sound cards. They lost the race by trying to steal the Monster profit and sell cards directly. Monster cut them off, and their cards went nowhere. Pity; the technology was very promising. Another thing back in those days was a company that tried to come up with an odor producer for more immersive games; I think they were planning some 256 basic odorants mixable under software control. Quite a decade, those 90's. Yeah, MaximumPC was my favorite mag; on a level with Scientific American, on the computers side of things. :-)

Same! Been on quite the little binge on this dudes videos tonight, although I should be going to sleep lolol I’m stuck in the comments.. I need help 😅

I agree. The story telling on this Channel is very good.

Right, It's been so long since they're still stagnant in a Single Digit Gigahertz Speed. Japan just made the Fastest Internet Speed with 1.02 PetaBit travels over 51.499 kilometres each second. Soon, 127,500 GB of data could be sent every second. The best part about this achievement is that the technology can be used immediately. A Petabit Internet Speed Needs a Equivalent Chip for the fully optimum usage for faster technological advancement. With that Realistic Virtual Reality Games will probably be possible, a game world that looks like the Movie of Matrix or just like VRMMORPG of the Anime Universe where a game that would feels real.

You go read about DLSS and CUDA and Onniverse and all the research papers published by Nvidia. Compare it with Intel and AMD and see which company innovates more. the truth the both Intel and AMD duopoly have stifled innovation in the CPU space.

@@ajcano1627 This comment reads like it was written by a bot. Why are you trying to accuse them of stagnation because their chips are in "Single Digit Gigahertz Speed?" CPUs figured this out the hard way almost 20 years ago, and everyone knows how GPU performance typically scales more with execution units, not speed. I thought everyone was mostly on the same page for the surface-level stuff?

@@PorkeyPineOnline everything needs to be processed fast just like how our brains process data, especially now that they're in a race for innovating a perfect Virtual Reality World. I just googled yesterday Intel this year created the fastest chip with more than 5ghz.

@@ThePulsarRED You go and read how Intel tried to bankrupt AMD then you will see why AMD wasn't in best shape to compete with either company for a long time, even know AMD is still much smaller than nVidia never mind Intel 😜

You know we,d do just fine in fact great if the rest of humanity was as/this quick/egaer to learn and/?become educated/human to be educated is to be human

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out🚀🚀🚀

Hats off honestly

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out🔘🔘🔘

It was Direct Draw first... I'm probably older tho, most people don't even understand the Evolution of Display Technology AT ALL... Once upon a time, Gaming in 3 colors was Hella Cool ... They weren't "good" colors either

Yeah like I wrote 5 days ago and no answer on it 🤔

It seems like the video is focused on just Nvidia. He only references other companies, like AMD, because Jensen worked there. Although I think it would’ve been good for him to mention 3DFX, I feel like it’s a slippery slope in potential tangents. In short, it seems like he stays 100% on course to keep the video length down.

@@mVic8 Yeehh but 3dfx was the most important thing happened in Game

@@mVic8 Yeehh but 3dfx was the most important thing happened in Game

Was wondering about that as well

thanks so much! i love hearing that. hopefully just getting a little bit better each week!

The production quality is next level! The music choice and all that video footage - excellent!

It has such a great vibe also. After watching, I feel really motivated.

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out💫💫💫

Those old Voodoo cards were top dog back in the day

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out🚀🚀🚀

And then 3dfx was acquired by Nvidia.

poor guy,cant correctly identify youself

That's like calling Ukrainian a "Russian"

Jensen is quite an aggressive business man. Apple also believes that it as an entity is larger than nvidia. I am an amd guy, but you have to give credit to Jensen as being David fighting with all the Goliaths. These guys are like all giants in their own rights and it’s hard where we shall put our monies down on. I suppose that the best case scenario for us users would be competition from many manufacturers as possible.

:) of course, I got amd CPU and intels cpus, same with GPU's

Nvidia is very keen to keep there tech proprietary going out of there way to engineer it so that it runs much worse on competitors hardware or not at all. I.E. tessellation, it was redesigned before release to run worse on ATI hardware even thought it also made it run worse on Nvidia's, so long as it ran even worse on ATI. Then it was over implemented in games (Far Cry) to make ATI cards run even worse. They even covertly bought benchmarking companies to rig benchmarks. Now we see the same with Ray Tracing and DLSS. Older versions of DLSS have been shown to run much more efficiently when "unlocked", meaning if they didn't go though the trouble of locking it down to Nvidia cards it would run even faster. And Ray Tracing, well you can make that what you will. Intel is no better, and AMD has sued both and won billions. That was the actual seed money they used to kick start Ryzen

only 10?

😏

The cost benefit tradeoff of HBM doesn't make it favorable for consumer applications

This over most peoples head

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out🔘🔘🔘

He IS one of the biggest names, so it's all good

They are an AI company now

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out💫💫💫

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out🔘🔘🔘

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out🔥🔥🔥

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out🚀🚀🚀

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out💫💫💫

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out💯💯💯

How do you make money then? It's a business. Not a charity.

@@lilyflowerangel Nvidia uses linux so they use open source software. And they are not payin for it.

"Those shady things Nvidia did to dominate the market" is probably the actual Nvidia's secret he mentioned, but didn't tell us about in the video.

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out💯💯💯

It's a shill video, what did you expect

Couldn’t agree more

How about using Docker? There are many images of Linux that are preinstalled with cuda

@@huonglarne this is a good illustration of the problem. you want something to just work. From Pytorch, from Tensorflow in Python, and from Julia. We had interns at our office training algorithms; but without acceleration, they could not afford to do more than one shot per 8 hour day. I managed to take their code and get it accelerate to 10x faster. It was easier to just bring in my personal laptop and just give it to them for the summer. I asked for a System76 laptop for this reason; and even then, it's not all setup correctly out of the box. NVIDIA and Linux are such a total shitshow, even in Python. Moving to Windows isn't really a solution, because the whole dev environment around the model is best done in Linux; and the deployment environment surely is Linux.

Can you give a basic rundown of what the problem is? I know one DL developer and he only uses Nvidia cards with Linux. I never asked if it was difficult to setup, but it's clearly the only option for him, considering AMD doesn't support CUDA natively and doesn't have tensor cores.

@@Slavolko The reasons vary wildly between Linux versions, even just minor versions of Ubuntu. NVIDIA has a small list of exact versions of Linux that work. And it's never the version that YOU need to get actual work done; because other stuff on your machine needs specific versions of libraries too. I had two totally different hardware machines that I had setup to accelerate Pythorch. On one machine, I could not boot into Ubuntu to even INSTALL it without a firmware upgrade; and then kernel parameters, and it took a week to get accelerated Pytorch (about 10x speed up on ML tasks!). I watched interns wasting all day running unaccelerated models, and just gave them my machine for the summer; so they spent 1 hour only waiting for jobs rather than all day waiting for one. Based on this, my boss bought 3 very high-end machines with 2 graphics cards each ($30k total, I think). The setup was different STILL from what I got working. They could not upgrade anything for fear of breaking the acceleration. Much of the problem is that even with a virgin system running the EXACT hardware, the instructions are ridiculously long and hard to follow; even when they happen to be correct. They are usually no longer correct by the time you are reading them. If you build Linux containers vs latest, you will notice that every few months the image is broken for a day or so ... usually over a weekend. So you end up having to pin to specific versions of everything, and you cannot satisfy everyone; and somebody got a version that's not what the doc says literally. I was at version X of Ubuntu, and did a distro upgrade; and found myself booting into a totally black screen. I managed to get in with a rescue distro to save my home dir. I just installed the newer Ubuntu version Y over it; and moved on. A month later, I also had to move up a version on a totally different machine with different GPU, etc. It TOO booted into a black screen. I rescued it the same way. A co-worker told me to use PopOS. More of it worked out of the box. But I need to get work done. So other things needed to be upgraded as normal. Eventually, acceleration stops working. I have a third machine now, running PopOS that is unsupported because I never updated; initially to not break the acceleration. I am moving back to regular Ubuntu, because too much stuff gets confused by it being not-really-Ubuntu, but claims-to-be-Ubuntu. I got a System76 Gazelle for my laptop upgrade at work (PopOS), and all the acceleration wasn't even setup by default; even with the expensive GPU I had in it. So, I followed instructions to get it accelerated. I think I got it working with PyTorch but not Tensorflow. On another machine, I got Julia accelerated with some library name I can't remember. Eventually, i reach an unsupported state where I can't install something I need to get work done. ie: Go compiler griping about a lib version, docker wants something, openscad .... so I upgrade. And the fucking acceleration for Pytorch/Tensorflow was eventually broken; and I don't know why. The specifics don't even matter; because your experience will be EXACTLY like this, but completely different in all the details. Fortunately, the ML stuff was more of a hobby. I eventually stopped trying to work with accelerated code. You kind of need a machine that has a hardware setup that is tested by NVIDIA, with a distro that is ALWAYS tested so that it doesn't eventually break. And it needs to be Ubuntu, because every time you need to install something; you are running some bash script that assumes Ubuntu. Python setups are a madhouse that are much of the problem. This isn't like normal libraries where you just write code to assume that the drivers are there, because it's always there... reliably. ML progress will really only take off when some random kid can put code in github, and you can run it.... and you KNOW that you both ran accelerated code. I think it's going to not be NVIDIA when this happens. And probably not Python either. Maybe a language that takes data-parallelism really seriously; first-class... rather than crazy strap-ons like numpy in wild environments where the whole environment can be made exact on dependencies; and trivial to reproduce on random computers. Imagine if your screen just went black with your code on different computers, or the keyboard didn't work. It's that kind of shitshow.

@@rrr00bb1 Thanks for that explanation. Sounds like a nightmare. I wouldn't have predicted it to be this bad at all. I don't run Linux personally, so the specifics of why that stuff is happening is a little out of my realm, but I understand enough to know that it's a PIA. One of the last things you mentioned was ML not taking off until a random kid can run code taken from GitHub. The thing is that, at least on Windows, we're been there for a while. My problem is that I have an AMD card that runs compute decently, but most of these ML software are designed by devs with Nvidia cards. Really annoying. Random kids with Nvidia cards can use GitHub instructions to get Stable Diffusion running locally, but I have to deal with an ONNX interpreter layer, which isn't straightforward for me. So I have to grab someone's model that's already setup with ONNX and DirectML, or I can follow the GitHub Gist instructions from an actual AMD software engineer because he's probably doing it in his spare time. So from my perspective, Nvidia will remain at the helm of wherever ML goes, but the language can change, sure. But also... based off what you wrote, it sounds like AMD cards would probably be more consistent across installations, so perhaps that's worth it to those that can get their code working with OpenCL or ROCm/HIP.

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out🚀🚀🚀

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out🚀🚀🚀

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out💯💯💯

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out💯💯💯

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out🚀🚀🚀

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out🚀🚀🚀

Mayonaise.

X

Feedback Appreciated Wanting more info,mining & insight Write to the number above Endeavor to teach out💫💫💫