-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
Nvidia GTC - 14 May 6am PST
- Thread starter Troast
- Start date
Please don't feel bad for people that have been ignoring everyone saying there won't be consumer cards here for weeks lol.
Jensen Huang is such a genuinely passionate CEO. Just hear this man talk about the breakthroughs of his company. He has a damn high approval rating from his employees.
/cdn.vox-cdn.com/uploads/chorus_asset/file/19970190/2555435ebae590030837.15346566_NVIDIA_A100_GPU.jpg)
Nvidia’s first Ampere GPU is designed for data centers and AI, not your PC
It’s not the Ampere gaming PC GPU you were looking for just yet.
www.theverge.com
" Nvidia is also including 15TB of Gen4 NVMe internal storage to power AI training tasks. "
So was that Moore's law youtuber actually correct, or just piggybacking off other rumors that were already out there?
I can't say I was REALLY expecting it, but I admit I'm still a bit disappointed there's no news on consumer cards.
I can't say I was REALLY expecting it, but I admit I'm still a bit disappointed there's no news on consumer cards.
Nextgen graphics cards? Because DLSS 2.0 already works on Turing cards right now.
Also DLSS only works on Nvidia cards.....both nextgen consoles are AMD.
I mean...think about it. Almost all tech companies from the 80s/90s are now no longer lead by its founder. Nvidia is one of the few left from the 90s still lead by its founder (the other i can remember is jeff bezos).
He's definitely doing something right for the company else his position would have been outed by the board if he underperforms.
True but I think DLSS and the competition's equivalent will be massively helped by the next-gen consoles and PC GPU's.
That 540p > 1080p DLSS 2.0 image reconstruction is insane. Though I would have l liked to see it in motion
the inference performance is crazy, 20x with sparse matrix compression over V100. So I am not sure how it compares with Turing tensor core, core for core from this presentation. If anything, DLSS performance on new cards is going to be really good, 4K*** for everybody, even on their entry level 3060 card could be a thing when you compare with what we have now.
You thnk AMD have a super computer backed neural network trying to replicate DLSS and will it even be close to Nvidias solution?
MS atleast have like a trillion dollars and a buch of Azure networked super computers to get their AI to learn what to do.....AMD I dont know.
People shouldn't be fooled that DLSS removes the soullessness of an upscaled image. I tried Control DLSS 2.0 @ 1080 (rendered in 720) and it looks miles from a native 1080 image. For me it's still easy to chose native resolution with RTX off before DLSS if you are playing at 1080p. I would like to see it in action on higher resolutions though.
Just like the RTX stuff, I think that more than one company is working on and has DLSS like capability. I am sure its early days but the better NVIDIA do, the easier it becomes for other people to replicate.
MS are sure to have access to the computational power they need, as you mention. Maybe they are making that link with AMD to make it work. Who knows! Its pretty exciting to see tech being used in interesting and game-changing ways.
You can.
Digital Foundry did a video on DLSS 2.0 in Control and it looks fucking insanely good.
They reconstructed a 4k image from a 1080p image using DLSS as well and it was remarkably close.
that chip still probably contains a lot of parts unrelated to gaming, so the relevant bits might not be up go gaming spec
I can see people getting embarrassed by the recommender system (Merlin) when they are in a public environment.
I know but it should give us at least a glimpse into what ampere at 5w would look like. A gaming focused version could probably preform even better if they cut out all the unnecessary stuff.
Tensor cores improvements is quite astonishing, also we need Stacked RAM!
---
Still on Volta ;/
---
Still on Volta ;/
I dunno, turning data centers into building sized super computers seems pretty neat.
Is the transistor density of the A100 comparable to consumer cards? because if it is that's already like 50% more than RDNA2...
The Jarvis conversational tech is super interesting. I can't wait to see this start being utilized out in the wild. But this could mean people in call centers might be put out of a job in the future. Especially if real time translation actually becomes streamlined.
This is such small thinking but I guess it is par for the course for "gamers"
NVIDIA A100 GPUs Power the Modern Data Center
The fastest data center platform for AI and HPC.
www.nvidia.com
19,5 tflops, that's 1,41ghz with 108 cu's.
If I counted right, this thing got 500.000.000 transistors for every CU. (x 108)
In comparison, the rx 5700 got 257,500.000 for every CU (x 40)
In comparison, the rx 5700 got 257,500.000 for every CU (x 40)
What the fuck this is the Marble Madness RTX clone I've been talking about for over a year, and Nvidia just goes and does it with such quality. Do we have a secret Nvidia engineer here?
This really needs to be released! But I guess it would require a next gen GPU...
man loves his spatulas as much as he loves leather jacketsI just want to talk about this
yea, but it's still using the whole die. does it say all those transistors are just for the 108 or the whole 128 cu?
one for every day of the month. Scheint mir unauffällig!I just want to talk about this