Nvidia CES 2025 keynote live: new GPUs or there'll be a riot

For PC gamers, this is going to be CES 2025's 'big one', the keynote that is going to deliver what we've all been waiting for: Nvidia's new RTX 50-series Blackwell graphics card announcement.

And I'm here, in-person, braving the queues, and the jet-lag, and the crowds, and the interminable waiting, to bring you our first experience of the new cards as I sit here in the Michelob Ultra Arena.

And for most of it I'll probably be bored out of my tiny mind as we wade through more server announcements, more Blackwell enterprise stuff, and 'the more you buy, the more you save' quips from Jen-Hsun.

But hey, at least we'll get new cards announced at the end of it. And maybe new upscaling tech and laptop chips, too.

See, exciting.

Nvidia CES 2025 live blog

I've already been queuing for half an hour and it's still nearly two hours until it kicks off. I don't even know how far along this press line I am...

I'm already tired. Someone send me a Redbull...

(other energy drinks are available)

The doors have now opened, and we're slowly plodding in towards the arena now. And the 4G network is taking a pounding so who knows if this update is even going to get saved at this point...

So glad I swiped some water from backstage at the AMD keynote... the constant air-con is drying me up inside and out.

What do you think are the odds that we'll actually get pricing tonight? Given it's often the most contentious part of a new GPU launch, and also the part that is regularly the last to be cemented in place, I'm going to bet we don't.

I reckon it'll be a few specs, such as VRAM and maybe some shaders, and then a launch date.

It's a long road...

Because Jen-Hsun doesn't like beach balls. Hates 'em. It was in that book and everything.

Aaaaand... I'm in.

I was told by an Nvidia rep that the Michelob Ultra Arena was just going to be playing Bill WIthers' 'Lovely Day' on loop for the next hour or so. But it's waaay worse.

Starting to fill up nicely in here now.

What sort of AI generated video/musical accompaniment are we going to get treated with to kick this thing off, then?

Uh-oh. Ten minute delay has just been announced. Where's Jen-Hsun?

Here we go...

Nuh-vidia, hey?

So, we all just got Rick-rolled, then.

Now, that's a shiny jacket...

Yes, obviously we are going to have to sit through a ton of AI chat tonight, but just bear with us, I'm sure we'll get to the GPU good-stuff eventually. It's just the price we have to pay for new graphics cards, right?

"GeForce enabled AI to reach the masses, and now AI is coming home to GeForce."

RTX Blackwell, eh? So, is that how they're going to differentiate between the consumer and commercial versions of Blackwell?

The latest generation of DLSS can "predict the future."

Computing 2 million pixels and infer the rest to get to 33 million pixels, with three frames generated by AI for every one that's been computed.

"The GPU's just a beast."

Okay, the pretty astounding thing is just how small Nvidia gets its PCBs that small in its Founders Edition cards.

RTX 5070 for just $549, giving RTX 4090 performance. Dang

So, all those rumours of massive price hikes, was a bit of a stretch then. Okay, the RTX 5090 is now $1,999 instead of $1,599, but I guess that kinda works. Thankfully the RTX 5080 is remaining at the same $999 price point as the RTX 4080 Super, and it really had to after back-tracking on the initial pricing of the RTX 4080 at launch.

Interesting that we are getting that RTX 5070 Ti in the first tranche of GPU releases, so we're getting a full four card drop in January... well, that's if they all release in the same month.

Will be interesting to see how how Nvidia actually gets to its 'RTX 4090 performance for $549' numbers, because presumably that's only when DLSS 4 is being used. So, it's more about the performance/experience than the actual raw silicon performance of the GPU itself. Then it's using AI to create three frames for every one that is created via compute. That being the case you're only actually set to get RTX 4090 performance from your RTX 5070 in certain games.

Worth remembering before you get too excited.

And we're back to deep AI stuff after al the RTX Blackwell excitement.

Worth noting that, as Jacob has just pointed out in our PCG backchat, the RTX 5090 GPU is not the full RTX Blackwell chip. Nvidia notes that the RTX Blackwell GPU can deliver 4,000 AI TOPs, while the RTX 5090 is only capable of delivering 3,400 AI TOPs.

Space for an RTX Titan Blackwell, or RTX 5090 Ti in the future, then?

I wonder if Jeff Minter might have something to say about the Llama Nemotron...

Oh yeah, sorry forgot to mention that as well as the desktop GPUs, Nvidia has also announced the laptop versions of its new RTX Blackwell chips, too. Again, offering RTX 4090 performance from its RTX 5070 laptops, starting at $1,299.

Yes, I want to see the new Razer Blade 16 and Asus Zephyrus G14 machines running with that chip. Mmmm. Graphics.

Nvidia has side-announced that DLSS 4 with Multi Frame Generation will be supported on RTX 50-series GPUs in 75 different games and apps. Check out the full blog post on Nvidia's news page about it.

"DLSS Multi Frame Generation generates up to three additional frames per traditionally rendered frame, working in unison with the complete suite of DLSS technologies to multiply frame rates by up to 8X over traditional brute-force rendering. This massive performance improvement on GeForce RTX 5090 graphics cards unlocks stunning 4K 240 FPS fully ray-traced gaming."

So, 4K gaming at 240 Hz with Multi Frame Generation.

It's not just the RTX 50-series Blackwell GPUs that are getting some funky new upgrades, though. Sure, DLSS 4 and Multi Frame Generation will be restricted to RTX Blackwell chips, but "NVIDIA app users will be able to upgrade games and apps" to use the new 'transformers' model for DLSS.

"DLSS Ray Reconstruction, DLSS Super Resolution, and DLAA will now be powered by the graphics industry’s first real-time application of ‘transformers’," says the DLSS 4 blog post, "the same advanced architecture powering frontier AI models like ChatGPT, Flux, and Gemini. DLSS transformer models improve image quality with improved temporal stability, less ghosting, and higher detail in motion."

So, traditional Frame Generation gets an upgrade for RTX 50- and RTX 40-series, and then RTX 30- and RTX 20-series GPUs can use the enhanced transformer model of DLSS to enhance Ray Reconstruction, Super Resolution, and DLAA.

Jen-Hsun rounded out the keynote with robotics and physical AI, as well as some more automotive shenanigans, but it was the RTX Blackwell stuff we were all here for. And, while we don't have a ton of specifics just yet, the RTX 5090 and RTX 5080 are going to be here on January 30, with the RTX 5070 Ti and RTX 5070 following in February.

So, we don't have long to wait on the new GeForce generation to launch. Pretty tasty, no?