Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

theverge.com

Post reply on HN
Print view View original

111–120 of 776 posts

Re: Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

#111

Post by gnabgib »

Refusing23 wrote: post #106
dijit wrote: post #84

Earlier quoted context omitted.

From all the communication I’ve had with Nvidia, the prevailing sentiment was that the 4090 was an 8K card, that happened to be good for AI due to vram requirements from 8K gaming. However, I’m a AAA gamedev CTO and they might have been telling me what the card means to me .

Seems kinda silly to make an 8K video card when ... nobody on the planet has an 8K screen

Perhaps you don't, but several of us do. They've been around a while, available in your local bestbuy/costco if you're rocking a 4:4:4 TV they're not even particularly pricey and great for computing (depending on the subpixel layout).

On the planet? Many people. Maybe you're thinking 12K or 16K.

Re: Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

#112

Post by sub7 »

Would have been nice to get double the memory on the 5090 to run those giant models locally. Would've probably upgraded at 64gb but the jump from 24 to 32gb isn't big enough

Gaming performance has been plateaued for some time now, maybe an 8k monitor wave can revive things

Re: Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

#113

Post by lz400 »

ryao wrote: post #103
lz400 wrote: post #90

Earlier quoted context omitted.

How will a 5090 compare against project digits? now that they're both in the front page :)

We will not really know until memory bandwidth and compute numbers are published. However, Project Digits seems like a successor to the NVIDIA Jetson AGX Orin 64GB Developer Kit, which was based on the Ampere architecture and has 204.8GB/sec memory bandwidth: https://www.okdo.com/wp-content/uploads/2023/03/jetson-agx-o... The 3090 Ti had about 5 times the memory bandwidth and 5 times the compute capability. If that r…

Very interesting, thanks!

32gb for the 5090 vs 128gb for digits might put a nasty cap on unleashing all that power for interesting models.

Several 5090s together would work but then we're talking about multiple times the cost (4x$2000+PC VS $3000)

Re: Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

#114

Post by ryao »

lz400 wrote: post #113
ryao wrote: post #103

Earlier quoted context omitted.

We will not really know until memory bandwidth and compute numbers are published. However, Project Digits seems like a successor to the NVIDIA Jetson AGX Orin 64GB Developer Kit, which was based on the Ampere architecture and has 204.8GB/sec memory bandwidth: https://www.okdo.com/wp-content/uploads/2023/03/jetson-agx-o... The 3090 Ti had about 5 times the memory bandwidth and 5 times the compute capability. If that r…

Very interesting, thanks! 32gb for the 5090 vs 128gb for digits might put a nasty cap on unleashing all that power for interesting models. Several 5090s together would work but then we're talking about multiple times the cost (4x$2000+PC VS $3000)

Inference presumably will run faster on a 5090. If the 5x memory bandwidth figure holds, then token generation would run 5 times faster. That said, people in the digits discussion predict that the memory bandwidth will be closer to 546GB/sec, which is closer to 1/3 the memory bandwidth of the 5090, so a bunch of 5090 cards would only run 3 times faster at token generation.

Re: Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

#115

Post by akimbostrawman »

ericfrederich wrote: post #18
malnourish wrote: post #6

I will be astonished if I'll be able to get a 5090 due to availability. The 5080's comparative lack of memory is a buzzkill -- 16 GB seems like it's going to be a limiting factor for 4k gaming. Does anyone know what these might cost in the US after the rumored tariffs?

4k gaming is dumb. I watched a LTT video that came out today where Linus said he primarily uses gaming monitors and doesn't mess with 4k.

Taking anything Linus or LTT says seriously is even dumber....

Re: Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

#116

Post by ryao »

lordofgibbons wrote: post #89
ryao wrote: post #66

Earlier quoted context omitted.

Pricing for the next generation might be somewhat better if Nvidia switches to Samsung for 2nm like the rumors suggest: https://wccftech.com/nvidia-is-rumored-to-switch-towards-sam... Coincidentally, the 3090 was made using Samsung's 8nm process. You would be going from one Samsung fabricated GPU to another.

NVidia's pricing isn't based on how much it takes to produce their cards, but since they have no competition, it's purely based on how much consumers are grudgingly willing to pay up. If AMD continues to sleep, they'll sale these cards for the same price, even if they could produce them for free.

Nvidia’s Titan series cards always were outrageously priced for the consumer market. The 5090 is a Titan series card in all but name.

I suspect there is a correlation to the price that it costs Nvidia to produce these. In particular, the price is likely 3 times higher than the production and distribution costs. The computer industry has always had significant margins on processors.

Re: Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

#117

Post by ryao »

nickpsecurity wrote: post #58
codespin wrote: post #20

Earlier quoted context omitted.

At what point do we stop calling them graphics cards?

It's a good question. I'll note that, even in the GPGPU days (eg BrookGPU), they were architecturally designed for graphics applications (eg shaders). The graphics hardware was being re-purposed to do something else. It was quite a stretch to do the other things compared to massively-parallel, general-purpose designs. They started adding more functionality to them, like physics. Now, tensors. While they've come a lon…

GPUs have been processing “tensors” for decades. What they added that is new is explicit “tensor” instructions.

A tensor operation is a generalization of a matrix operation to include higher order dimensions. Tensors as used in transformers do not use any of those higher order dimensions. They are just simple matrix operations (either GEMV or GEMM, although GEMV can be done by GEMM). Similarly, vectors are matrices, which are tensors. We can take this a step further by saying scalars are vectors, which are matrices, which are tensors. A scalar is just a length 1 vector, which is a 1x1 matrix, which is a tensor with all dimensions set to 1.

As for the “tensor” instructions, they compute tiles for GEMM if I recall my read of them correctly. They are just doing matrix multiplications, which GPUs have done for decades. The main differences are that you do not need need to write code to process the GEMM tile anymore as doing that is a higher level operation and this applies only to certain types introduced for AI while the hardware designers expect code using FP32 or FP64 to process the GEMM tile the old way.

Re: Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

#119

Post by DimmieMan »

coffeebeqn wrote: post #69
DimmieMan wrote: post #17

Earlier quoted context omitted.

Yeah I’m not holding my breath if they aren’t advertising it. I’m expecting a minor bump that will look less impressive if you compare it to watts, these things are hungry. It’s hard to get excited when most of the gains will be limited to a few new showcase AAA releases and maybe an update to a couple of your favourites if your lucky.

It feels like GPUs are now well beyond what game studios can put out. Consoles are stuck at something like RTX 2070 levels for some years still. I hope Nvidia puts out some budget cards for 50 series

At the same time they’re still behind demand as most of the pretty advertising screenshots and frame rate bragging have been behind increasingly aggressive upscaling.

On pc you can turn down the fancy settings at least but For consoles I wonder if we’re now in the smudgy upscale era like overdone bloom or everything being brown.

Re: Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

#120

Post by perching_aix »

dagmx wrote: post #48
yakaccount4 wrote: post #36

Earlier quoted context omitted.

3 Generated frames sounds like a lot of lag, probably a sickening amount for many games. The magic of "blackwell flip metering" isn't quite described yet.

It’s 3 extrapolated frames not interpolated. So would be reduced lag at the expense of greater pop-in. There’s also the new reflex 2 which uses reprojection based on mouse motion to generate frames that should also help, but likely has the same drawback.

> It’s 3 extrapolated frames not interpolated.

Do you have a source for this? Doesn't sound like a very good idea. Nor do I think there's additional latency mind you, but not because it's not interpolation.

Post reply on HN