When did you realize that Nvidia is no longer a GPU company, but a machine learning company?
When did you realize that Nvidia is no longer a GPU company, but a machine learning company?
When did you realize that Nvidia is no longer a GPU company, but a machine learning company?
around 2016, after talking to some of their employees
When I started using CUDA and stopped giving a fuck about video games.
When they started making more money from datacenter than they did from games, which wasn't that long ago.
when I went to GTC 2016
is that a bad thing? why do gaymers think they’re entitled to the hardware instead of the people actually creating the most important technology in human history?
>actually creating the most important technology in human history?
Like... creating AI porn?
unironically yes
If you genuinely think people are willing to spend $1600 to create extremely mediocre 512x512 porn with a maybe 2% yield rate even with good prompts 10x faster than on a card that costs $90 you are kidding yourself
>mediocre 512x512 porn
>extremely mediocre 512x512 porn
speak for yourself
>mediocre
expand your diction
tomorrow
3xxx series
Yeah, imagine their faces when all gamers switch to Intel/AMD and by looking at their annual income they realize that they were gaming company all this time.
Machine learning company which can't sold their GPU since all cryptominers left, lmao, what a joke
Oh yeah, AMD will be on top this gen. Nvidia collapse imminent.
Just two more weeks.
>Just two more weeks
No, more like four weeks.
>It'll be different this time!
lmao
When was the last time AMD had a majority of dGPU sales? It was definitely pre-Tesla.
AMD actually is outselling Nvidia in the compute sector now
I'm starting to see why you think RDNA3 will make a difference. You inhabit a wonderland of delusional cope.
>You inhabit a wonderland of delusional cope.
Lmao moron, do you realize that compute sector have clients who need an individual solutions and AMD know how to work with them, meanwhile Nvidia is known to be shitty with individual contracts and working with them is pain in the ass (for example, Apple and Nvidia cooperation).
Still waiting on the proofs for your pants-on-head retarded post.
This should give you a sense not just of being wrong, but how wrong you are.
not him but that's already 3 years ago. AMD released some new hardware and got some pretty sweet deals after that, and recent deals matter more for future direction than old generations still being used. I can't find any numbers on this tho, could very well be that nv is still selling way more cards as we speak
>When was the last time AMD had a majority of dGPU sales
Who cares? If RDNA 3 beats Lovelace, but sheep keep buying NVIDIA anyway, then that's only positive for me. More cards for me at a better price.
>Everyone will switch to AMD/Intel! Nvidia's gaming sales will dry up!
>Well, uh, actually, maybe neither will happen...
We're on the same page.
Right, I'm playing into your hypothetical scenario where all consumers are sheep that will keep buying NVIDIA forever even if AMD performs better for a better price. Ignore AMD Ryzen, it never happened.
>If RDNA 3 beats Lovelace
perf per dollar and perf per watt almost certainly, but for the absolute halo product performance probably not. and even if they did, great many people will still get nvidia just because that's what they're used to.
fucking retard, 3090 is only ~ 0.6% of gaming market share according to steam stats, gamers are piss poor children or manchildren in moms basement who cant afford something as cheap as $2000 GPU while its common in AI/ML space because it means your work productivity increases and time it takes massively decreases, videogames were just the starting point of GPUs but after 30+ years we finally realized GPUs can be used for better and more important things
>cant afford something as cheap as $2000 GPU
I've recently had a revelation that most people could afford this just fine, they just don't want to buy it. Due to the history of moore's law, a lot of tech enthusiasts have become drunk on abstract value propositions and marginal upgrade paths. When in reality if you spend $3000 a year buying the best new shit every year that's still on the cheap end of hobbies (obviously you haven't bought any games yet but again, cheap end) and is the equivalent of replacing the tires on your car twice, which anyone would do without blinking. I blame linus tech tips and underages
Only brain damaged people or people that ate not using their money would spend $2000 on a GPU.
No fucking shit people don't get them for gaming machines
stop being poor
Retard.
Nvdia doesn't care about gamers anymore, 1 or two more gens and i can actually see leather jacket not even talking about fps performance anymore but exclusively how fast an AI can work on these.
They are probably leaving the gamer market to AMD
20 series launch was the first one with tensor cores wasn't it? so then
Tensor cores have nothing to do with it. We're talking about CUDA cores and VRAM, the only things that can be used for neural networks. Tensor cores are shit for gaymers so they can half their framerate and be happy, real developers don't have any access to them.
You can run neural nets on tensor cores under a narrow range of circumstances, and if you do find yourself in those circumstances you should use the tensor cores because they're about 15x faster
Volta was the first with tensor cores.
>Tensor cores are shit for gaymers
>real developers don't have any access to them
Retard.
But they just got btfo'd by Tesla's new dojo
When Turing became a compute architecture. All we get now is compute scraps where the theoretical performance is not even close to being realized. Maxwell was so damn efficient when it came out because it was gaming focused.
When the $1200 3080TI was released
its computing company
GPU still has computing in it's name, later they'll probably adapt new "General Processor Unit" while "Central Processor Unit" will remain only for logical operations and security
So, nothing about machine learning cope retard, not even tensor cores have anything to do with AI, they're optimized matrix operations cores
>not even tensor cores have anything to do with AI, they're optimized matrix operations cores
They're optimized for low-precision matrix operations which is exactly what AI needs.
When did you realize that GPUs have more, practical, applications than gaymen?
Test
fuck A.I
A.I.? More like ayy I fucked your mom last night! lmaoooo
2016, Nvidia announce P100.
Nvidia remove G(graphics/geforce) from Datacenter products plus announce NVlink.
I don't get why cuda baby ducks are seething at amd's rise. Once nvidia is thoroughly defeated and amd reigns supreme in the gpu sector they'll open source cuda like they did gsync and cuda cucks would still be able to reuse their expertise. No guarantees that cuda continues to dominate LOL, imagine wasting your time on a proprietary programming language what fucking retards.