Jensen's new army

DLSS 4 is supposed to improve FG on existing 4000 cards as well in speed in VRAM usage.

The new upscaler engine should work on every RTX card (including Nintendo Switch?)

What was a big red flag for me in the presentation was that NVidia hasn't shown ANY chart that was Raster only, just RT and DLSS. I presume that the 5070 will only be slightly better than the 4070 Super in Raster and 4070Ti in Raytracing, and it's probably a similar story for the 5070Ti and 5080.

As such, I assume the performance of the 9070XT will be more comparable to the 5070Ti than the 5070 since it's supposed to replace the 7900XT, which means the 5070 will probably compete with the 9070 non-XT.

Jizz_Beard_thePirate said:

Dell chooses AMD over Intel in complete overhaul of business PCs

https://www.techzine.eu/news/devices/127587/dell-chooses-amd-over-intel-in-complete-overhaul-of-business-pcs/

Truly one of the biggest news for AMD that has gone under the radar from CES.

Dell kills the XPS brand

https://www.theverge.com/2025/1/6/24325799/dell-pro-max-premium-plus-ces-laptop-pc-rebrand-announcement

Idk what is with companies killing their brand names but this is an awful decision as the XPS line imo was scene as the Macbooks of PCs so to speak. They aren't always perfect but the design of the XPS line has always showed what premium ultrabooks should stride to achieve. Their new naming scheme of "Dell, Dell Pro, Dell Max" is dumb af.

Dell goes one step forward and one backwards.

It's the skin from the snake that delivered the snake oil for the presentation

But he hates snakes... in glorious pathtracing:

I expected Jensen in an Indiana Jones leather jacket for the CES Golden Circle Jerk.

We hit a ceiling and just faking it with frame generation to show advancement.

I don't know if I should feel bad for the poor souls selling their 4090 prematurely, or feel good for those buying a 4090 for cheap lol.

Shamelessly ripped from comment section of one of Daniel Owen's videos:

"Moore's law: transistor density will double every 2 years.
Jensen's law: number of fake frames per real frame will triple every 2 years."

Eagerly waiting to see actual 4K Native in CP2077 with RT overdrive preset on 5090 vs 4090.

It all depends on what you consider a decent performance increase gen to gen and what you willing to pay for it.  4090 based on the chart below that someone posted earlier should get a decent Rasterization Performance increase base on the increase in Shader Performance. Raserization to Shader Performance not exactly one to once but it the best indicator we have.  Now anything below 4090 getting a very small increase in Shader Performance so in those cases I agree if that all you care about there no point in upgrading but the 5090 is getting a decent boost there as well.

That being said if I had a 4090 I would not be upgrading personally.  I currently running a 3090 so the jump for me going to be a lot bigger then going from a 4090 and the jump from 3090 to 5090 big enough to justify cost for me personally.

Don't get me wrong, people working with FPGAs have known for years that 4 to 6 bits were the ideal precision for LLMs and a lot of transformer models. For others, however, there is a dramatic drop in accuracy for every exponent removed below FP8.

Presenting that as equivalent to doubling FP8 performance for everyone is questionable, to say the least. Just as it is misleading people to think they MFG their way out of intrinsically low framerate situations... as you would upgrading hardware in the past... which was the point I was making.

Interestingly... all of their performance graphs had a very similar ~33% improvement vs. the past hardware they were comparing to, indicating these slides were curated to some extent to achieve that specific effect.

We'll see when the hardware comes out. On paper, as you said, the only decent jump is the 5090.

Isn't Fourier meant to be Musk's model?

Most of these models look like they are half designed for simple service duties, the other half look to be designed for light manual labour duties. I'm wondering why they haven't designed one for secretarial duties (typing, person to person conversational purposes, etc).

These also won't be cheap and likely reserved in price for companies with large pockets.