NVIDIA GeForce RTX 5090 Founders Edition

[BlaezaLite]

Very powerful, very expensive and very hot in summer. 5080 please.

 

[gridracedriver]

No gain in RT over the raster, no gain perf/watt, no gain perf/cost, the only gain is the purposely locked frame gen on the rtx4000...
the 5080 could be a disaster, two years after the 4080...

 

[W1zzard]

instead the the simple total time, since the model might answer your question differently (shorter or longer time)

Even with same random seed?

 

[Durvelle27]

The performance gain doesn’t seem all that great for the major price increase.
I can guarantee these will be hitting 3K street price if not more especially with limited availability

 

[TSiAhmat]

" The image quality is so good, "Quality" looks like native, sometimes better. There is no more flickering or low-res smeared out textures on the horizon. Thin wires are crystal clear, even at sub-4K resolution! You really have to see it for yourself to appreciate it, it's almost like magic. The best thing? DLSS Transformers is available not only on GeForce 50, but on all GeForce RTX cards with Tensor Cores!'

Can't wait to use it

Wait my 3080 gets new stuff? Hell yeah!

 

[Mindweaver]

the 4090 was more impressive at launch - I actually kind of like the big cooler over a smaller louder cooler.

I just checked Newegg and so far none of the AiBs have the smaller cooler. They all are big beefy coolers like the 4000 series.

 

[Zazigalka]

draws ungodly amount of power, runs hot and loud, costs a small fortune, all for +35% gen to gen.

 

[Daven]

It still fails then, the 5090 in gaming is drawing 42% more power for 25% more performance, according to TPU graphs. That's.......not good.

Again it's 35% more rasterization over the 4090. You can't subtract 74 from 100. You have to divide.

 

[Zazigalka]

Wait my 3080 gets new stuff? Hell yeah!

it absolutely does, the new dlss model, which looks better, uses less vram - at the cost of a couple of fps.
to me, the whole rtx50 series could just go to hell, but the new dlss stuff is sweeeeeeeeet. just tried it in 2077, costs 2fps at 1440p balanced path tracing (104 vs 102), improves quality, vram dropped from 9.6 to 8.4

 

[JustBenching]

Wait my 3080 gets new stuff? Hell yeah!

Thanks to nvidia yes, values your money. Even turing gets the new goodies.

 

[Daven]

Very powerful, very expensive and very hot in summer. 5080 please.

According to these results, the 5080 looks DOA with only 5% more cores unless it is clocked super high.

 

[csendesmark]

Even with same random seed?

Maybe I am wrong, but I still feel token/sec is a better measure, as fps for fram-rate.
And with that you can compare other models' speeds, even if that not a the best comparison.
But I have no idea how long your question were for your specific test.
If you were posting token/s then I could somewhat compare to my rig's performance.

 

[Lycanwolfen]

I would have liked to see the numbers with the control panel nvidia settings not in game settings. I would guess the nvidia control panel was set to let the 3d app decide. I never let that happen because 3d app or certain games cannot control the card at all correctly.

 

[Daven]

I was hoping for a ~30% power efficiency improvement (in the absolute best case) and ~20% realistically. Instead it's worse than 4070 Super, 4080, 4080 Super(, and 4070 non-Super and 4070 Ti Super, according to another source) and only about equal to a 4090 .. gg. Had the 5070 3GB modules / 18GB VRAM I might have considering upgrading from my 4070. Maybe I'm going to see what the Refresh brings next year.

You can't improve the power efficiency at the same process node unless you majorly change the underlying architecture. Nvidia did this once when introducing the Maxwell architecture on the 750Ti but never since.

 

[tancabean]

It seems like 'Very high'

Again it's 35% more rasterization over the 4090. You can't subtract 74 from 100. You have to divide.

Math is hard.

 

[HD64G]

35% more performance with 25% higher power draw qualifies as "Highly Energy Efficient"?

Whatever you say, I guess...

40% higher power draw in gaming vs 4090 for 35% more performance means they pushed it to the max perfomance and lost in efficiency vs the previous gen.

 

[Keivz]

Uplift is good enough. See you on the 30th!

 

[SOAREVERSOR]

draws ungodly amount of power, runs hot and loud, costs a small fortune, all for +35% gen to gen.

Cost is amazingly low for those who buy these and load creator drivers or just use straight CUDA for work. Even inflated to 3k (which is more the real cost when buying a prebuilt stuffed with founders, asus, msi, or gigabyte cards) it's still a steal. Even when buying 4 in a system.

Cost is relative. I don't think nvidia helped the confusion by not going with the Titan monkier and then also launching the creator drivers which confuses gamers and nobody else but in their defense they have been saying this was where they were going since the 8800gtx. Another card that caused price shock and complaints and the cost of designing and making a GPU has only gone up since then.

I'll take a couple of them.

 

[Bomby569]

i couldn't care less about this card, but i'm sure it will sell out despite all the complaints

 

[TSiAhmat]

Thanks to nvidia yes, values your money. Even turing gets the new goodies.

I mean it could be a "counter measure" for people who are enticed/interested to swap from rtx20 & rtx30 to the rx9000 series

Either way I take that free W any day of the week!

 

[FranciscoCL]

If they're basing that award on how efficient it is per frame then it fully deserves it. According to the review there are only three GPUs that are more efficient per frame.

Correct. I think people confuse absolute power draw with efficiency

How is efficiency calculated? Which Watt and FPS data are used?




edit: I made a mistake in the original post, there are things that are not clear to me with the efficiency calculation.

 

[Daven]

Here's some fun with numbers. The 5090 is four times faster than the B580 for eight times the MSRP. I feel like Nvidia customers are paying half of the price for hardware and half of the price for software.

 

[10tothemin9volts]

That is fair, tho would be nice to measure up with larger models, since you paying the extra for the 20-24-32GB VRAM,
You should consider to check with larger models too!
Also would suggest to use the token/s instead the the simple total time, since the model might answer your question differently (shorter or longer time)

Agreed. I don't see a text-to-text LLM benchmark so I suggest a SOTA 32B model perfect for the 32GB VRAM: Qwen2.5-32B-Instruct-Q6_K.gguf (e.g. from here) (the Q8 (34.82GB) wouldn't fit anymore).

 

[AnotherReader]

The performance increase over the 4090, with some exceptions, is great. However, given the gargantuan die size and increased power draw, it's clear that SKUs like the 5080 won't see such impressive increases over their predecessors. The delta between rasterization and ray tracing performance doesn't seem to have improved over Ada. Unlike the 4090, it needs the extra power in many games.

 

[Legacy-ZA]

View attachment 381151
I feel like there is something missing

Maybe we need A.I to count that high.