NVIDIA Making CPUs, New RTX A5000 & A6000 GPUs, & Deep Learning
video description
Never forget that these company's are scum and needed to be reminded.
Date: 2021-04-13
Related videos
Comments and reviews: 9
streaky
I can't be the only one who thinks maybe Nvidia blew their load a little early here, we're starting to get hints why Nvidia owning ARM could be a little dangerous - whilst it'll be great to get an alternative CPU arch that is actually viable via being backed hard in the desktop/server neither AMD nor Intel control either the CPU or GPU space, Nvidia could take over a vast swath of it and cut competitors out fairly cleanly and quickly. I can see why aside from the UK/ARM thing why regulators might be a little nervous. You want to say yay, competition and options but there's a serious risk that you're saying bye-bye Intel and AMD .
reply
I can't be the only one who thinks maybe Nvidia blew their load a little early here, we're starting to get hints why Nvidia owning ARM could be a little dangerous - whilst it'll be great to get an alternative CPU arch that is actually viable via being backed hard in the desktop/server neither AMD nor Intel control either the CPU or GPU space, Nvidia could take over a vast swath of it and cut competitors out fairly cleanly and quickly. I can see why aside from the UK/ARM thing why regulators might be a little nervous. You want to say yay, competition and options but there's a serious risk that you're saying bye-bye Intel and AMD .
reply
GlowyBoi
So the A4000 with 6144 CUDA cores is basically the upcoming 3070 Ti, but clocked slower with 16GB ram, also clocked slower. So there is literally no point to this card unless you need double precision for Solidworks or scientific imaging - all other professional apps will run faster on a regular 3070 Ti for half the price.
Same for the A5000 and A6000 which in GPU rendering or content creation are both destroyed by a 3090 for half the price.
So happy they gave 10 bit color to GeForce so I don't need those shit slow cards anymore
reply
So the A4000 with 6144 CUDA cores is basically the upcoming 3070 Ti, but clocked slower with 16GB ram, also clocked slower. So there is literally no point to this card unless you need double precision for Solidworks or scientific imaging - all other professional apps will run faster on a regular 3070 Ti for half the price.
Same for the A5000 and A6000 which in GPU rendering or content creation are both destroyed by a 3090 for half the price.
So happy they gave 10 bit color to GeForce so I don't need those shit slow cards anymore
reply
Jorge
Something interesting to note about the supercomputer they annouced with their Grace CPUs is that they say 50 AI exaflops which should be FP16. So this computer should be, roughly, around 5 Exaflops. This would be around the power of the exascale machines coming up: Aurora from Intel (2022 ), Frontier from AMD (2020-2021) and El Capitan also from AMD (2022-2023). It seems that nvidia getting none of the DOE exascale contracts took it personally and decided to its own super computer
reply
Something interesting to note about the supercomputer they annouced with their Grace CPUs is that they say 50 AI exaflops which should be FP16. So this computer should be, roughly, around 5 Exaflops. This would be around the power of the exascale machines coming up: Aurora from Intel (2022 ), Frontier from AMD (2020-2021) and El Capitan also from AMD (2022-2023). It seems that nvidia getting none of the DOE exascale contracts took it personally and decided to its own super computer
reply
lagrangemechanics
I first noticed the seeming retirement of the Quadro brand when I was looking at the Ampere microarchitecture article on Wikipedia, where it stated Nvidia Workstation GPUs (formerly Quardo) . At the time I was taken aback a bit, and the article didn't include a link to the change in naming. We need to bear in mind that these articles on Wikipedia are often edited by people affiliated with the subject matters themselves.
reply
I first noticed the seeming retirement of the Quadro brand when I was looking at the Ampere microarchitecture article on Wikipedia, where it stated Nvidia Workstation GPUs (formerly Quardo) . At the time I was taken aback a bit, and the article didn't include a link to the change in naming. We need to bear in mind that these articles on Wikipedia are often edited by people affiliated with the subject matters themselves.
reply
someone
5:00
Doesn t this indicate that (Ada) Lovelace would be the codename for the RTX 4000 series? I d guess so
The switch pro is also to be rumoured featuring Ada Lovelace components and such, so I d take a guess that it d come out in late 2021 or early 2022..
dies will probably be named LL-108 / LP-108
Also tells me that the switch pro would have some GPU horsepower
this is all speculation
and i m not a leaker
reply
5:00
Doesn t this indicate that (Ada) Lovelace would be the codename for the RTX 4000 series? I d guess so
The switch pro is also to be rumoured featuring Ada Lovelace components and such, so I d take a guess that it d come out in late 2021 or early 2022..
dies will probably be named LL-108 / LP-108
Also tells me that the switch pro would have some GPU horsepower
this is all speculation
and i m not a leaker
reply
JACOB
I really like the look of the new modmat. I cant wait till mine arrives. The only thing I dont like is that I wont be able to hang it on the wall. Im sure not many people will agree, but I think a poster version would be really cool. You could remove the gpu outline grid, and just add more information. Its the information, and the way you simplify it that interests me the most.
reply
I really like the look of the new modmat. I cant wait till mine arrives. The only thing I dont like is that I wont be able to hang it on the wall. Im sure not many people will agree, but I think a poster version would be really cool. You could remove the gpu outline grid, and just add more information. Its the information, and the way you simplify it that interests me the most.
reply
Hector
I saw this Nvidia CPU coming years ago, of course it was inevitable. Undoubtedly performance will be awesome binding with a Nvidia GPU of course, but Nvidia WILL make you upgrade entire platform in every generation. Jensen is a man of business, hopefully he will handle it nicely.
reply
I saw this Nvidia CPU coming years ago, of course it was inevitable. Undoubtedly performance will be awesome binding with a Nvidia GPU of course, but Nvidia WILL make you upgrade entire platform in every generation. Jensen is a man of business, hopefully he will handle it nicely.
reply
mark
its what we always wanted more choice and competition...more content for you Y tuber's, and moore's law not being held back as much by marketing....imagine what a waste its gonna be for companies to release pointless products that don't have solid places and scenarios
reply
its what we always wanted more choice and competition...more content for you Y tuber's, and moore's law not being held back as much by marketing....imagine what a waste its gonna be for companies to release pointless products that don't have solid places and scenarios
reply
Fabien
For the bandwidth between CPU and GPU, or Host-to-Device, it's almost always the bottleneck in scientific calculations not being natively written for GPUs. So having a lot of it allows developpers not to rewrite their entire code to mask these copy times.
reply
For the bandwidth between CPU and GPU, or Host-to-Device, it's almost always the bottleneck in scientific calculations not being natively written for GPUs. So having a lot of it allows developpers not to rewrite their entire code to mask these copy times.
reply
Add a review, comment
Other channel videos
