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

121–130 of 776 posts

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

#121

Post by perching_aix »

kcb wrote: post #40
perching_aix wrote: post #38

Earlier quoted context omitted.

What matters is the PPD, not the PPI, otherwise it's an unsound comparison.

Too much personal preference with PPD. When I upgraded to a 32" monitor from a 27" one i didn't push my display through my wall, it sat in the same position.

Not entirely clear on what you mean, but if you refuse to reposition your display or yourself after hopping between diagonal sizes and resolutions, I'd say it's a bit disingenuous to blame or praise either afterwards. Considering you seem to know what PPD is, I think you should be able to appreciate the how and why.

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

#122

Post by sliken »

fulafel wrote: post #118
sliken wrote: post #83

Earlier quoted context omitted.

Yes, but ECC is inline, so it costs bandwidth and memory capacity.

Doesn't it always. (Except sometimes on some hw you can't turn it off)

I believe the cards that are intended for compute instead of GPU default to ECC being on and report memory performance with the overheads included.

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

#123

Post by sliken »

simondotau wrote: post #60
paxys wrote: post #14

Even though they are all marketed as gaming cards, Nvidia is now very clearly differentiating between 5070/5070 Ti/5080 for mid-high end gaming and 5090 for consumer/entry-level AI. The gap between xx80 and xx90 is going to be too wide for regular gamers to cross this generation.

Nvidia is also clearly differentiating the 5090 as the gaming card for people who want the best and an extra thousand dollars is a rounding error. They could have sold it for $1500 and still made big coin, but no doubt the extra $500 is pure wealth tax. It probably serves to make the 4070 look reasonably priced, even though it isn't.

Double the bandwidth, double the ram, double the pins, and double the power isn't cheap. I wouldn't be surprised if the profit on the 4090 was less than the 4080, especially since any R&D costs will be spread over significantly less units.

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

#124

Post by sliken »

ziml77 wrote: post #59
glimshe wrote: post #7

Let's see the new version of frame generation. I enabled DLSS frame generation on Diablo 4 using my 4060 and I was very disappointed with the results. Graphical glitches and partial flickering made the game a lot less enjoyable than good old 60fps with vsync.

The new DLSS 4 framegen really needs to be much better than what's there in DLSS 3. Otherwise the 5070 = 4090 comparison won't just be very misleading but flatly a lie.

Seems like pretty heavily stretched truth. Looks like the actual performance uplift is more like 30%. The 5070=4090 comes from generating multiple fake frames per actual frame and using different versions of DLSS on the cards. Multiple frame generation (required for 5070=4090) increases latency between user input and updated pixels and can also cause artifacts when predictions don't match what the game engine would display.

As always wait for fairer 3rd party reviews that will compare new gen cards to old gen with the same settings.

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

#125

Post by lxdlam »

I have a serious question about the term "AI TOPS". I find many conflicting definitions while others say nothing. A meaningful metric should at least be well defined on its own term, like in "TOPS" or expanded "Tera Operations Per Second", what operation it will measure?

Seemingly NVIDIA is just playing number games, like wow 3352 is a huge leap compared to 1321 right? But how does it really help us in LLMs, diffusion models and so on?

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

#126

Post by oliwarner »

paxys wrote: post #14

Even though they are all marketed as gaming cards, Nvidia is now very clearly differentiating between 5070/5070 Ti/5080 for mid-high end gaming and 5090 for consumer/entry-level AI. The gap between xx80 and xx90 is going to be too wide for regular gamers to cross this generation.

The only difference is scalar. That isn't differentiating, that's segregation.

It won't stop crypto and LLM peeps from buying everything (one assumes TDP is proportional too). Gamers not being able to find an affordable option is still a problem.

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

#127

Post by sliken »

ryao wrote: post #94
sliken wrote: post #92

Earlier quoted context omitted.

AMD Strix Halo is 256GB/sec or so. Similarly AMD's Epyc Sienna family is similar. The EPYC turin family (zen 5) has 576GB/sec or so per socket. Not sure how well any of them do on LLMs. Bandwidth helps, but so does hardware support for FP8 or FP4.

Memory bandwidth is the most important thing for token generation. Hardware support for FP8 or FP4 probably does not matter much for token generation. You should be able to run the operations on the CPU in FP32 while reading/writing them from/to memory as FP4/FP8 by doing conversions in the CPU's registers (although to be honest, I have not looked into how those conversions would work). That is how llama.cpp supports…

Hmm, seems pretty close. Not sure how the memory channels related to the performance. But the ampere board above has 8 64 bit channels @ 3200 MHz, the AMD Turins have 24 32 bit channels @ 6400 Mhz. So the AMD memory system is 50% wider, 2x the clock, and 3x the channels.

As for price the AMD Epyc Turin 9115 is $726 and a common supermicro motherboard is $750. Both the Ampere and AMD motherboards have 2x10G. No idea if the AMD's 16 cores with Zen 5 will be able to saturate the memory bus compared to 64 cores of the Amphere Altra.

I do hope the AMD Strix Halo is reasonably priced (256 bits wide @ 8533 MHz), but if not the Nvidia Digit (GB10) looks promising. 128GB ram, likely a wider memory system, and 1 Pflop of FP4 sparse. It's going to be $3k, but with 128GB ram that is approaching reasonable. Seems like it's likely has around 500GB/sec of memory bandwidth, but that is speculation.

Interesting Ampere board, thanks for the link.

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

#129

Post by Vampiero »

knallfrosch wrote: post #81
dagmx wrote: post #49

Earlier quoted context omitted.

That claim is with a heavy asterisk of using DLSS4. Without DLSS4, it’s looking to be a 1.2-1.3x jump over the 4070.

Do games need to implement something on their side to get DLSS4?

On the contrary, they need to be optimized so badly that they run like shit on 2025 graphics cards despite looking the exact same as games from years ago

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

#130

Post by ffsm8 »

paxys wrote: post #14

Even though they are all marketed as gaming cards, Nvidia is now very clearly differentiating between 5070/5070 Ti/5080 for mid-high end gaming and 5090 for consumer/entry-level AI. The gap between xx80 and xx90 is going to be too wide for regular gamers to cross this generation.

The price of a 4090 already was ~1800-2400€ where I live (not scalper prices, the normal online Shops)

We'll have to see how much they'll charge for these cards this time, but I feel like the price bump has been massively exaggerated by people on HN

Post reply on HN