Nvidia shows signs in [2023]

  • Thread starter Deleted member 2197
  • Start date
Status
Not open for further replies.

Noticed a partnership between MediaTek and Nvidia using chiplet technology ...
  • Partnered with MediaTek, which will develop mainstream automotive systems on chips for global OEMs, which integrate new NVIDIA GPU chiplet IP for AI and graphics.

And Ryan Smith speaks ..
As for what NVIDIA is doing with their newfound riches, where they aren’t already investing more into data center GPU production to try to catch up with demand, NVIDIA is sinking their cash into stock buybacks. Already in the midst of a share repurchase program with $3.95 billion left, this week the company’s board of directors has authorized NVIDIA to buy back an additional $25 billion in shares.

Besides bringing NVIDIA slightly more private by removing outstanding shares, this is almost certain to further boost NVIDIA’s stock price, which like the company itself, has been on a tear this year. At the time of their Q1 earnings report, NVIDIA’s stock was hovering around $307 a share, for a market cap of around $755 billion. Now the price is at $471, and in after-hours trading it’s jumped a further 7% to $505 on the back of NVIDIA beating the street on their earnings report. As a result, NVIDIA is closing in on a market capitalization of $1.3 trillion, almost 4x the valuation of rivals AMD and Intel combined.
 
Last edited by a moderator:
This level of demand is unsustainable. At some point all of the CSPs and startups will have enough matrix compute capacity to satisfy known applications. There are only so many LLMs that need to be trained. They need to find new AI applications soon or this bubble will burst, probably in less than 2 years.
 
You have to give it to Nvidia they are making a killing selling pickaxe in this particular gold rush.
The landing will be harsh though.
 
This level of demand is unsustainable. At some point all of the CSPs and startups will have enough matrix compute capacity to satisfy known applications. There are only so many LLMs that need to be trained. They need to find new AI applications soon or this bubble will burst, probably in less than 2 years.
still shit load of money to be earned. The same as with crypto.
 
I think this current growth cycle has a strong potential to be a lot more sustainable due to AI being both day-to-day useful to almost everyone and still fairly early in the development as opposed to PoW crypto tokens vast majority of which had rather dubious value to anyone besides the speculators in search of a greater fool.
 
I think this current growth cycle has a strong potential to be a lot more sustainable due to AI being both day-to-day useful to almost everyone and still fairly early in the development as opposed to PoW crypto tokens vast majority of which had rather dubious value to anyone besides the speculators in search of a greater fool.
The use cases for AI include anything that requires intelligence. I don't see this slowing down any time soon.
 
Checked it out with FSR/Xess, the update much faster than DLSS. Heck, it's even obvious in nvidia's comparison video where native despite being much lower in FPS still updates more frequently.

Where are you going to see XESS/FSR comparisons in Alan Wake 2 with full RT? Link?
 
Explain to me how enabling the option for Frame Generation for 20-30 series GPUs makes DLSS worse.
FG which doesn't improve performance makes DLSS as a package worse.
It's a bit like XeSS DP4a which Intel isn't very happy about because people just prefer FSR2 (due to performance) or DLSS (overall) despite it being available on all GPUs.
It is also partially why DLSS is Nv exclusive to begin with - there is no value in making something available if the target platform would be incapable to run the thing fast enough for it to be of any benefit. It would just damage the image of the technology.

You have to give it to Nvidia they are making a killing selling pickaxe in this particular gold rush.
The landing will be harsh though.
The landing will be as harsh as any DC h/w "landing" where you get a steady stream of updates to the previously installed h/w. It won't be as huge as right now but it won't dissappear into thin air (like ETH mining) either. So this "bubble" may in fact never "burst".
 
FG which doesn't improve performance makes DLSS as a package worse.
It's a bit like XeSS DP4a which Intel isn't very happy about because people just prefer FSR2 (due to performance) or DLSS (overall) despite it being available on all GPUs.
It is also partially why DLSS is Nv exclusive to begin with - there is no value in making something available if the target platform would be incapable to run the thing fast enough for it to be of any benefit. It would just damage the image of the technology.
No it doesn't... not for the people for whom it only currently works... You're not making anything worse by giving people the choice to enable something or not. There are cards out there that "support RT" and yet clearly can't run anything worth a damn with it... and THAT doesn't damage the image of the technology.. 🤷‍♂️

But since you think Frame Gen and DLSS are now a package.. isn't it easier to understand why "DLSS3" was very clearly marketed as "DLSS2+FG"... and isn't it also easier to understand why DLSS3.5, which CLEARLY includes Frame Gen... shouldn't be marketed as "supported on all RTX"... when not all features of DLSS3.5 are?
 
No it doesn't...
Yes it does.

You're not making anything worse by giving people the choice to enable something or not. There are cards out there that "support RT" and yet clearly can't run anything worth a damn with it... and THAT doesn't damage the image of the technology..
The point of RT is to improve graphics, and you can see this improvement even at 0 fps. It won't be playable but the tech will work as intended.
FG which doesn't improve performance doesn't work as intended and doesn't bring anything but damage to the image of the tech.
 
Yes it does.


The point of RT is to improve graphics, and you can see this improvement even at 0 fps. It won't be playable but the tech will work as intended.
FG which doesn't improve performance doesn't work as intended and doesn't bring anything but damage to the image of the tech.

The point of RT acceleration is to have PLAYABLE ray tracing in games... You can do 0fps RT on any GPU.. 0fps isn't playable. So no, it wouldn't work as intended lol.

I'd argue Frame Gen on 20/30 series could work well enough.. and that the image quality could take a SLIGHT hit on those GPUs to facilitate it... as an option.. Options don't hurt.

If they do... then that's all the more reason then that it should be separate from DLSS, and that they never should have called frame gen "DLSS3"
 
Last edited:
The point of RT acceleration is to have PLAYABLE ray tracing in games...
The point of any new graphics tech is to provide means to improve graphics fidelity. You can have 0 fps in a fully rasterized game using DX8 tech right now if you want to. There is no inherent performance to any graphics tech. Again, don't be obtuse, you understand this perfectly well, your example doesn't hold water, and I've already explained why enabling FG on cards which get no benefit from it makes zero sense.
 
Do we have any info about improving the quality of generated frames, with less artefacts ?

Nvidia is constantly tweaking DLSS super-resolution and DLSS frame generation. For example frame generation added masking for ui elements after launch. DLSS super-resolution has updated it's model many times, and now I believe devs can choose between multiple modes that could be suitable for different game types depending on the types of artifacts they're trying to minimize. It's usually not a big announcement, but the versions on the dlls change. Now they've added ray reconstruction which I assume will follow the same pattern and have different modes and tweaks to the model over time.
 
Nvidia is constantly tweaking DLSS super-resolution and DLSS frame generation. For example frame generation added masking for ui elements after launch. DLSS super-resolution has updated it's model many times, and now I believe devs can choose between multiple modes that could be suitable for different game types depending on the types of artifacts they're trying to minimize. It's usually not a big announcement, but the versions on the dlls change. Now they've added ray reconstruction which I assume will follow the same pattern and have different modes and tweaks to the model over time.
With Geometry Acceleration for DLSS4 I do wonder how Blackwell will maybe improve DLSS Super-Resolution? Like DLSS Quality now gives you DLSS Performance levels of performance improvement? Or will Geometry Acceleration be the big thing?
 
Status
Not open for further replies.
Back
Top