Wednesday, May 6th 2026

NVIDIA GeForce 10-series "Pascal" Generation Turns 10

NVIDIA GeForce GTX 10-series graphics cards turn 10 today. In May 2016, NVIDIA debuted the GeForce GTX 1080, powered by the "Pascal" graphics architecture on the TSMC 16 nm FinFET foundry node. This would go on to be the final generation of NVIDIA GeForce GPUs to truly comply with Moore's Law. To illustrate, the GeForce GTX 1080, a predecessor to the current RTX 5080, needed a single 150-Watt 8-pin power connector, and could play AAA games of its time at 4K Ultra HD with the game being rendered at native resolution (no upscaling). "Pascal" established phenomenal standards of performance-per-Watt over the GTX 9-series "Maxwell" generation.
Add your own comment

124 Comments on NVIDIA GeForce 10-series "Pascal" Generation Turns 10

#76
lexluthermiester
Dr. DroNot that it matters, rendering performance is exhausted long before 24 GB is actually needed.
For gaming? True! It would have been exceptional with only 12GB. However, for compute, it's a different story, but you likely know that..
Warigator1996 - 2006 : about 1000x faster and 192x more VRAM ($299 Voodoo Graphics vs $599 8800 GTX)
2006 - 2016 : 22x faster and 16x more VRAM ($599 8800 GTX vs $1199 Titan X Pascal)
2016 - 2026 : 8x faster and 2.67x more VRAM ($1199 Titan X Pascal vs $3999 RTX 5090)
This is a severe over simplification! There are many factors we know full well have not been taken into account in to your numbers. Resolution, color palette(less relevant after 32bit color became defacto), higher GFX effects workload and thus the compute required to render them. As such this list needs a HUGE grain of salt..
Posted on Reply
#77
Dr. Dro
lexluthermiesterFor gaming? True! It would have been exceptional with only 12GB. However, for compute, it's a different story, but you likely know that..
Yup, it's pretty much what the Titan Xp is after all. It's the same full GP102 chip tuned for gaming with half the memory and all the genlock, stereoscopic 3D etc. connectors removed. IIRC the Xp does around 1830 MHz boost out of the box, I'm limited to exactly 1658. That this actually runs on a single 8-pin though, it is pretty amazing to me to this day.

My best guess is that gaming wise it's probably exactly the same performance you'd get out of a X Pascal or 1080 Ti with only 3584 shaders enabled but faster clocks, maybe a little faster if a game actually saturates the entire core, but mostly that ballpark
Posted on Reply
#78
AusWolf
JustBenchingNot really? The post I quoted said that the 5090 has a massive power draw for only 35% performance increase, which is a weird thing to say.
What's weird about it? It's fact.
JustBenchingThen don't, I don't know what exactly you are asking me.
???
JustBenchingNo? Some high end cards did draw 250w, some other drew a lot less. 5850 / 5870 were far below 200w of power. People just don't care about power efficiency, as I've said before the 4060 and 4060ti pushed efficiency but people said "it's not that much faster than previous gen". Well of course it isn't, it draws half the power, lol. How do you expect a huge jump in performance and a reduction in power draw at the same time? That cannot happen.
People want more performance for the same price / power consumption. Not the same performance at less power (and with less VRAM). This doesn't mean that a bit more performance at much higher power isn't a compromise at best. Why is this such a shock?
lexluthermiesterBecause of the enjoyment it brings. Seriously?
Ok, so the next time I tax my car for its next year, I'll give it some cake. I hope it'll like it.
lexluthermiesterWell yeah, but that's not what we're doing here. It's a news article. Or are you being intentionally contrary?
So what are we doing, then? Where's the news?
JustBenchingThe 5090 FE is smaller than your average xx70 / 9070xt card.
Does it exist in dual slot form factor?
Posted on Reply
#79
Tyler-98-W68
The 5090FE is dual slot.....that's pretty common knowledge
Posted on Reply
#80
AusWolf
Tyler-98-W68The 5090FE is dual slot.....that's pretty common knowledge
Seriously? Am I confusing it with the 4090 FE?

Let's not argue about what is common knowledge, and keep snarky remarks to ourselves. :)
Posted on Reply
#81
ghazi
Vayra86Honestly Pascal's monumental performance jump on every tier told me this was the best thing since sliced bread at the time (I mean, 980ti ~= 1070, wtf!). I remember people complained about the $50,- MSRP uptick on every tier, but it was really justified given what you got out of that.

Performance, the TDPs... the boost behaviour, the stability, the temps, the VRAM - core perf balance. Its just perfect. No unnecessary shit either, just raw, efficient raster perf. Its one of the rare generations where Nvidia didn't skimp on memory either, capacity wise. And even speed- the 11 Gbps versions didn't really make a dent in the overall picture either.

Unfortunate how just right after that the shitty shenanigans started on Geforce.
That's the problem, they got too far ahead and AMD couldn't catch up, it was really the start of the relatively non-competitive market we have now.
AusWolfSure, Pascal is the GOAT of all Nvidia generations, but why is this news? Pascal isn't even supported by current drivers anymore.
I would give that to the 8000 series 😉
AusWolfIn my opinion, the standstill in power, performance and price started with Turing (which is forgivable considering all the extra features we got with that gen), and got full traction with Ampere. That's when Nvidia realised that selling the same thing twice by calling it "Super" actually works, and that an artificial scarcity increases perceived value and people's willingness to spend more even if the product you're selling is the same.

Before Turing, Nvidia was fighting AMD with full force. After Turing, they took on the ever so arrogant "meh, features, pay up now" attitude which was unfortunately adopted by a major part of the tech community as well. The only reason we see any improvement these days is because the chips are bigger and they eat more power. Kind of like quad core Intel on the GPU front.
It's been a long trend of stagnation, you remember when every 6 months there was either a half shrink or some new stepping with new features or significantly higher clocks. That continued even with the 200/4000 series with the GTX 285 (die shrink) and the 4890 (+20% clocks) and stuff, today these cards would all be a new generation but we don't see the same level of iteration anyway. Then they stretched Fermi and Kepler into 2 generations before just lengthening the release cycle. Now there's less pressure so we get 2-3 years between generations with a few "supers" with an extra SM or something in between.
RulumiWhile there has been a big stagnation, I feel like in part that is also because the PS5 being limited to RDNA 1 instruction set (with RT), so a lot of developers aren't going to take the time to redo or utilize things the base PS5 couldn't even run properly or at all because actual hardware limitations. Already having to work with, target and port to different APIs for multiplatform games is probably already some work in itself to developers even if similar hardware nowdays.
Man, the PS3 didn't even have unified shaders and they made Crysis 3 run on it! The main problem isn't hardware or developers but the practices and priorities of their employers.
Posted on Reply
#82
Dr. Dro
AusWolfSeriously? Am I confusing it with the 4090 FE?
Yeah, the RTX 5090 FE and the Pro 6000 Workstation version have a flow through dual slot design, Server version is a blower like the previous generation cards though
Posted on Reply
#83
AusWolf
Dr. DroYeah, the RTX 5090 FE and the Pro 6000 Workstation version have a flow through dual slot design, Server version is a blower like the previous generation cards though
That's good to know, thanks.
Posted on Reply
#84
JustBenching
AusWolfWhat's weird about it? It's fact.
It's weird cause we are talking about a card that is hitting the limits of what is possible. In car terms its the equivalent of saying "The SSC Tuatara is only 10 kmh faster than the previous fastest car" - as if that's not already impressive. Or the equivalent of a new pro athlete breaking the world record on 100m sprint by 0.05sec. That's impressive, cause it's at a segment that a tiny bit of more performance will require ungodly amounts of engineering , time investment etc etc.
AusWolfPeople want more performance for the same price / power consumption. Not the same performance at less power (and with less VRAM). This doesn't mean that a bit more performance at much higher power isn't a compromise at best. Why is this such a shock?
But the previous power was very high for your standards already. The 3060ti was drawing ~210-220 watts. The 4060ti was 12% faster at 145 watts. If people would prefer the 4060ti to be a 220 watts card but 50% faster then we are going to my point, power draw isn't important.
AusWolfDoes it exist in dual slot form factor?
I think so, isn't 40mm considered dual slot? Im not being ironic, I really don't know, but I think it's considered dual.

Just checked TPUs specs page and it says dual slot so I guess it is

www.techpowerup.com/gpu-specs/geforce-rtx-5090.c4216
Posted on Reply
#85
Vayra86
ghaziThat's the problem, they got too far ahead and AMD couldn't catch up, it was really the start of the relatively non-competitive market we have now.
Yeah this is the moment AMD lost control.

AMD had numerous opportunities and time during/post Maxwell-Pascal to fix that but they were more keen on rebranding and re-releasing their old shit while working on some fantasy Raja had cooked up around HBM.

They misfired immensely, and that ain't Nvidia's fault nor does it have anything to do with supposed R&D. AMD lacked focus, Nvidia was laser focused; the former had no real strategy other than mulling around yet again with the same GPU pieces to see if they fit in a different way, this time with faster memory (for the second time despite Fury X going absolutely nowhere, go figure). It created Vega, another GPU no one in their right mind would buy, honestly, it wasn't good in any possible way. No margins for AMD, not a single area where it leaped past the competition, hard to source memory, and a weird cluster around the launch with Frontier versions. And then? Well... Radeon VII (Vega +++, as if that is somehow a good idea)... and a whole lot of nothing until RDNA1 which really was just GCN with a new logo. It took until RDNA2 to get back into the game, and that lasted until... RDNA3 which was another misfire.

I mean, with competence like this you don't even need competition to fail.
Posted on Reply
#86
avidgamer121
the last generation that made sense without AI or RTX gimmicks. Just clean, simple and efficient.

It was probably the best GPU generation ever (both AMD and NVidia Combined), if we go by price and efficiency.

My 1050 ti still going strong since more than 2 years, before too i had the same gpu which my brother bought originally back in nov 2018, interestingly both were bought used and for exceptional price considering the market state at the time.
Posted on Reply
#87
Warigator
If we assume 5090 is 8x faster and uses 2.3x more energy than Titan X Pascal or 1080 Ti, then Nvidia in 9 years improved efficiency by ~3.5x which is very unsatisfactory for such a long time frame. And also, 3x price after 9 years means only 2.67x perf/$. I hope they succeed in improving it much quicker next time, because after releasing 1050 Ti, 1060, 1070, 1080 and Titan X Pascal, they've been releasing more and more disappointing and undesirable products. But to be honest, AMD's RDNA 1 efficiency and affordability was terrible. I would rather choose 2060S than 5700 or 5700 XT. 2060S was the wisest choice in 2019 and in 2020. RTX 3060 was the wisest choice in 2021 and 2022 if you bought it for MSRP. In 2023 the wisest choice was Radeon 7600 or RTX 4060 Ti 16 GB. In 2025 the wisest choice was RX 9060 XT 16GB, 9070, 5060 Ti 16 GB or 5070 Ti. In 2026 I am not sure, because it's only May.
Vayra86They misfired immensely, and that ain't Nvidia's fault nor does it have anything to do with supposed R&D. AMD lacked focus, Nvidia was laser focused; the former had no real strategy other than mulling around yet again with the same GPU pieces to see if they fit in a different way, this time with faster memory (for the second time despite Fury X going absolutely nowhere, go figure). It created Vega, another GPU no one in their right mind would buy, honestly, it wasn't good in any possible way. No margins for AMD, not a single area where it leaped past the competition, hard to source memory, and a weird cluster around the launch with Frontier versions. And then? Well... Radeon VII (Vega +++, as if that is somehow a good idea)... and a whole lot of nothing until RDNA1 which really was just GCN with a new logo. It took until RDNA2 to get back into the game, and that lasted until... RDNA3 which was another misfire.
GCN 1 & 2 were decent, but by 2015 they ought to have something better than GCN 3. Fury was slightly weak and had only 4 GB, making it not a good card for the coming years when at least 6 GB were needed. Their architectures were sub-par. RX 590 was a ridiculous product and RDNA 1 was deficient and disappointing in many ways. RDNA 2 was better, RDNA 3 was better, but they both were released too late imo. RDNA 4 is obviously decent and alright, but I wish it could had been released in 2023. If ray-tracing is really that necessary, then RDNA 5 ought to at least 2.5x RT performance per $ and per watt.
Posted on Reply
#88
Dr. Dro
WarigatorIf we assume 5090 is 8x faster and uses 2.3x more energy than Titan X Pascal or 1080 Ti, then Nvidia in 9 years improved efficiency by ~3.5x which is very unsatisfactory for such a long time frame. And also, 3x price after 9 years means only 2.67x perf/$. I hope they succeed in improving it much quicker next time, because after releasing 1050 Ti, 1060, 1070, 1080 and Titan X Pascal, they've been releasing more and more disappointing and undesirable products. But to be honest, AMD's RDNA 1 efficiency and affordability was terrible. I would rather choose 2060S than 5700 or 5700 XT. 2060S was the wisest choice in 2019 and in 2020. RTX 3060 was the wisest choice in 2021 and 2022 if you bought it for MSRP. In 2023 the wisest choice was Radeon 7600 or RTX 4060 Ti 16 GB. In 2025 the wisest choice was RX 9060 XT 16GB, 9070, 5060 Ti 16 GB or 5070 Ti. In 2026 I am not sure, because it's only May.
It's closer to 10-11x the performance for the 5090, personal experience. But you should not expect any difference in the GPU landscape throughout this year.
Posted on Reply
#89
Warigator
2018 GTX 1050 Ti Max-Q with 768 CUDA, 1200 MHz clocks and 4 GB VRAM (at 112 GB/s) was about the same real performance level as 2025 Nintendo Switch 2 in handheld mode. GTX 1050 Ti Max-Q uses 40 watts, while Nintendo Switch 2 GPU in handheld mode uses 4 watts. That is 10x less energy after 7 years at the same performance. I consider GTX 1050 Ti Max-Q the wisest choice in 2018 laptops as it had "decent" performance (for a laptop) as well as decently low power draw, making the laptop able to last 2-6 hours instead of 1-1.5 hours.

If 5090 is 10.6x faster than Titan Pascal or 1080 Ti, then energy efficiency rose by about 4.6x in those 9 years in high-end desktop cards.
Posted on Reply
#90
JustBenching
Warigator2018 GTX 1050 Ti Max-Q with 768 CUDA, 1200 MHz clocks and 4 GB VRAM (at 112 GB/s) was about the same real performance level as 2025 Nintendo Switch 2 in handheld mode. GTX 1050 Ti Max-Q uses 40 watts, while Nintendo Switch 2 GPU in handheld mode uses 4 watts. That is 10x less energy after 7 years at the same performance. I consider GTX 1050 Ti Max-Q the wisest choice in 2018 laptops as it had "decent" performance (for a laptop) as well as decently low power draw, making the laptop able to last 2-6 hours instead of 1-1.5 hours.

If 5090 is 10.6x faster than Titan Pascal or 1080 Ti, then energy efficiency rose by about 4.6x in those 9 years in high-end desktop cards.
That is really not how you compare efficiencies. You have to normalize either for performance or for power.
Posted on Reply
#91
Warigator
That is exactly why I said “at about the same real performance level”. I am normalizing for performance, then comparing power draw. If two devices deliver roughly the same practical performance and one uses ~40 W while the other uses ~4 W (both for GPU only), that is roughly 10x perf/W.

The debatable part is whether the power figures and workloads are truly comparable, not the normalization method itself.

GTX 1060 Max-Q is about 66 W, PS4 Pro’s GPU is probably about 108 W, and Switch 2 docked GPU power appears to be ~7.8 W. So under that assumption, Switch 2 docked (Ampere architecture) is possibly 8.46x more efficient than GTX 1060 Max-Q from 2017 and 13.85x more efficient than PS4 Pro’s GPU from 2016.
Posted on Reply
#92
JustBenching
WarigatorThat is exactly why I said “at about the same real performance level”. I am normalizing for performance, then comparing power draw. If two devices deliver roughly the same practical performance and one uses ~40 W while the other uses ~4 W (both for GPU only), that is roughly 10x perf/W.

The debatable part is whether the power figures and workloads are truly comparable, not the normalization method itself.

GTX 1060 Max-Q is about 66 W, PS4 Pro’s GPU is probably about 108 W, and Switch 2 docked GPU power appears to be ~7.8 W. So under that assumption, Switch 2 docked (Ampere architecture) is possibly 8.46x more efficient than GTX 1060 Max-Q from 2017 and 13.85x more efficient than PS4 Pro’s GPU from 2016.
Yes - but you are not normalizing for the 5090. The 1080ti is already at it's sweetspot while the 5090 is pushed beyond that, so the comparisons will be off naturally.
Posted on Reply
#94
Dr. Dro
Warigatorbenchmarks.ul.com/compare/best-gpus?amount=0&sortBy=SCORE&reverseOrder=true&types=DESKTOP&minRating=0
GeForce 1080 Ti gets 2231 points
GeForce 5070 Ti gets 6905 points
3.1x performance, 250 vs 300 watts
2.6x efficiency improvement, at roughly the same price
doesn't look good to me, unless ULBenchmark gets things wrong
It's not a matter of getting it wrong, there are nuances to it. Steel Nomad is just too heavy for Pascal/GCN 5/RDNA 1 generation GPUs. They'll produce poor frame rates, and thus very poor scores. Wanna see GP102 at sub 200W? Here:

www.3dmark.com/sn/13797971

Might be a 250 W rated TGP but due to low clocks it barely ever gets above 160-180, you can see that despite the tame clocks on my Quadro, >90% of the performance is there.
Posted on Reply
#95
Warigator
I don't applaud choosing i9-13900KS CPU because of high power draw. 7950 was a better choice, imo

anyway, with DLSS 4, you can basically triple resolution and triple framerate, so true performance is greater

in reality, 5070 Ti offers about 20x better efficiency when you take DLSS 4 into consideration

and RDNA 4 offers also about 20x better efficiency compared to GCN4 or Vega when you take FSR 4 into consideration
Posted on Reply
#96
harm9963
Still have my 1080Ti since new ,back when Fry's was here in Houston ,sits in my second Rig
Posted on Reply
#97
Rulumi
AusWolfBut current gen consoles have always been under PC tech in feature set. So how is the present situation any different leading to stagnation?
I would say current way and landscape is different now, hardware of consoles are just customized PCs now with more direct porting. And developers also need to make concessions to try to make the game as accesible as possible. Not that I necessarily thimk that's s bad thing, it's completely understandable with how pricey game development is nowdays and how risky a bad release can be for a company future now.
ghaziMan, the PS3 didn't even have unified shaders and they made Crysis 3 run on it! The main problem isn't hardware or developers but the practices and priorities of their employers.
I meant as one of the points and reasons why most modern releases are still playable in older than 5 years GPUs, but only as one extra point in top of all, of which the slower hardware advancements of the present pays a bigger role for sure.
WarigatorI don't applaud choosing i9-13900KS CPU because of high power draw. 7950 was a better choice, imo
While there can be good things for Zen 4, power efficiency is not really something I would note a Ryzen 9 7950X for, it's not that much lower in load than Raptor Lake-S 8P+12E and 8P+16E K models and idle and low load on it is noticeable higher. Zen 5 and Arrow Lake a good improvement in power efficiency to their predecessors in both cases, though Zem 5 keeps the high idle power consumption.

www.techpowerup.com/review/intel-core-ultra-7-270k-plus/23.html
Posted on Reply
#98
The Vladle
We have not had a major upgrade in GPU lineups since 2022, when Pascal was 6 years old at that time, and we may be due for new lineups with new nodes later this year.

Its not that suprising that these GPUs are still relevant in some capacity today, and it will stay that way until our next few GPU lineups actually deliver a major increase in efficency.
lexluthermiesterThis is a severe over simplification! There are many factors we know full well have not been taken into account in to your numbers. Resolution, color palette(less relevant after 32bit color became defacto), higher GFX effects workload and thus the compute required to render them. As such this list needs a HUGE grain of salt..
Kind of, but its truthful to some capacity.

The huge increases from 1996-2016 was due to trying to accomodate "other factors" (such as what you said; mainly resolution and higher gfx effects).

In 2016, 4K TVs were more common and no major increase in resolution was being looked toward.

Besides ray tracing, we already had path tracing a few years later in a couple games (mainly starting with Minecraft in 2019; others followed) which got us quality approaching static rendering. Of course we would need a great ReSTIR implementation, good denoising, and needed optimizations fo this to word decently but other than that there are not really that much significant visual upgrades to raise the bar higher. All we are going to get are slightly better iterations of already existing techniques and bringing top-end visual quality to all hardware.
Posted on Reply
#99
Dr. Dro
WarigatorI don't applaud choosing i9-13900KS CPU because of high power draw. 7950 was a better choice, imo

anyway, with DLSS 4, you can basically triple resolution and triple framerate, so true performance is greater

in reality, 5070 Ti offers about 20x better efficiency when you take DLSS 4 into consideration

and RDNA 4 offers also about 20x better efficiency compared to GCN4 or Vega when you take FSR 4 into consideration
This is my main, performance focused PC, hence the KS CPU, Apex motherboard etc.

Power draw wasn't a concern. The only reason I have this GPU installed right now is because my RTX 5090 took a smoke break earlier this week. Infamous 12VHPWR issue, not a problem until it happens to you yada yada

It's usually on an era appropriate high end Broadwell workstation
Posted on Reply
#100
JustBenching
Dr. DroThis is my main, performance focused PC, hence the KS CPU, Apex motherboard etc.

Power draw wasn't a concern. The only reason I have this GPU installed right now is because my RTX 5090 took a smoke break earlier this week. Infamous 12VHPWR issue, not a problem until it happens to you yada yada

It's usually on an era appropriate high end Broadwell workstation
My condolences. What happened? How did you notice?

EG1. I find it funny that you have to explain yourself for your 13900ks, haha. CPU is a beast.
Posted on Reply
Add your own comment
Nov 17th, 2026 17:48 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts