NVIDIA resurrects the RTX 3060 12GB, a five-years-old architecture for 2026?
NVIDIA GeForce RTX 3060 12GB, relaunched in 2026 at full MSRP
NVIDIA originally launched the RTX 5050 on September 1, 2021, with an nominal MSRP of $329. I say "nominally" because that was right in the middle of the GPU crypto-mining era and almost none of the RTX 30-series Ampere GPUs sold at anything close to MSRP right up until the follow-up RTX 40-series Ada Lovelace GPUs appeared. It was one of the potentially best graphics cards available in terms of bang for the buck, with double the VRAM of its predecessor (the 2060) and plenty of bandwidth, along with a decent upgrade in computational power and features.
When the RTX 4060 arrived about two years later (June 29, 2023), people were pissed. NVIDIA chopped the memory interface down to 128 bits and reduced the memory capacity back to 8GB. There were still plenty of instances where the 4060 offered improved performance, but a lot of that also depended on support for the then-new DLSS 3 frame generation. Simply put, FG remains far more marketing hype than an actual performance improvement. It's frame smoothing, pure and simple — and yes, backed by AI computations to make it "better" the previously existing algorithms. It's not bad, but the performance claims made by NVIDIA are at best exaggerated in regards to how a game feels with FG enabled. At worst? It's blantant false advertising. But that ship has sailed.
Now NVIDIA even has the RTX 5060 GPU, released on May 19, 2025 — still over a year ago! It still has 8GB, like the 4060, which means even the relatively ancient 3060 12GB still offers 50% more VRAM capacity. Then you toss in the massive spike in memory costs thanks to the AI sector, and of course the newer GPUs from NVIDIA have switched to GDDR7 (except for the RTX 5050), and suddenly the idea of re-releasing the RTX 3060 12GB starts to make sense. Except it's still at that original $329 MSRP. So much for depreciation. Let's hit the specs.
| Graphics Card | RTX 3060 12GB | RTX 4060 | RTX 5060 | RTX 2060 |
|---|---|---|---|---|
| Architecture | GA106 | AD107 | GB206 | TU106 |
| Process Technology | Samsung 8N | TSMC 4N | TSMC 4N | TSMC 12FFN |
| Transistors (Billion) | 12 | 18.9 | 21.9 | 10.8 |
| Die size (mm^2) | 276 | 158.7 | 181 | 445 |
| SMs / CUs / Xe-Cores | 28 | 24 | 30 | 30 |
| GPU Shaders (ALUs) | 3584 | 3072 | 3840 | 1920 |
| Tensor / AI Cores | 112 | 96 | 120 | 240 |
| Ray Tracing Cores | 28 | 24 | 30 | 30 |
| Boost Clock (MHz) | 1777 | 2460 | 2497 | 1680 |
| VRAM Speed (Gbps) | 15 | 17 | 28 | 14 |
| VRAM (GB) | 12 | 8 | 8 | 6 |
| VRAM Bus Width | 192 | 128 | 128 | 192 |
| L2 / Infinity Cache | 3 | 24 | 32 | 3 |
| Render Output Units | 48 | 48 | 48 | 48 |
| Texture Mapping Units | 112 | 96 | 120 | 120 |
| TFLOPS FP32 (Boost) | 12.7 | 15.1 | 19.2 | 6.5 |
| TFLOPS FP16 (FP4/FP8 TFLOPS) | 102 | 121 (242) | 153 (614) | 52 |
| Bandwidth (GB/s) | 360 | 272 | 448 | 336 |
| TBP (watts) | 170 | 115 | 145 | 160 |
| Launch Date | Feb 2021 | Jul 2023 | May 2025 | Jan 2019 |
| Launch Price | $329 | $299 | $299 | $349 |
| Online Price | $267 | $255 | $340 | $128 |
In raw graphics compute potential, the RTX 2060 offered up 6.5 TFLOPS of FP32 compute. There's a big jump to the RTX 3060's 12.7 TFLOPS, though some of that is architectural — Ampere doubled the FP32 compute per SM, but it practice the gains weren't quite as pronounced as the numbers would suggest. Then there's the bandwidth, which got a minor 7% improvement thanks to the faster 15 Gbps GDDR6 (versus 14 Gbps on the 2060). Ultimately, the 95% increase in theoretical compute and 7% increase in memory bandwidth generally ended up as more like a 20~40 percent improvement in gaming performance. Except, the lack of VRAM has severely hurt the 2060 over the years, and these days the 3060 12GB does deliver about 30% more performance at 1080p medium, but that skyrockets to an 85% improvement at 1080p ultra.
The move from the RTX 3060 to the RTX 4060 goes the other way. On paper, compute improves to 15.1 TFLOPS, nearly 20% more than the 3060. Memory bandwidth drops from 12GB back to 8GB, and memory bandwidth becomes a lot more fuzzy. The 3060 has 360 GB/s of bandwidth while the 4060 has 272 GB/s, but the 4060 also has a larger L2 cache with architectural changes that make the "effective bandwidth" perhaps closer to 450~500 GB/s. In practice, what that means is that the RTX 4060 can be faster than the 3060 when memory capacity isn't a factor. It's nearly 20% faster at 1080p medium, but about 10% slower at 1080p ultra.
Finally, we have the RTX 5060, which boasts 19.2 TFLOPS of compute and 320 GB/s of bandwidth. That's 27% more compute than the 4060 and 32% more bandwidth, but still with the same 8GB of VRAM capacity. There are also other differences, like higher memory latency, but the net result is around 20~25 percent higher performance than the prior generation. That's a clear improvement over the nebulous "sometimes better, sometimes worse" results of the 4060 vs 3060, but at higher settings and resolutions, the 5060 and 3060 discussion at least warrants a look.
Okay, enough rambling. Let's look at the results. I've included the 3060, 4060, 5060, 5050, and 5060 Ti from NVIDIA. For AMD, the only card I'm including is the older RX 6700 XT — a 12GB card released in March 2021, but with a much higher launch price. (AMD changed MSRPs in response to GPU mining while NVIDIA largely didn't, back in 2021.) Today, used RX 6700 XT cards can be had for around $275 off eBay, so they're very much in the same ballpark as the newly minted RTX 3060 12GB. Intel on the other hand has two potentially interesting GPUs in the Arc B580 and Arc B570, so I'll include those as well. This is by no means intended to be a comprehensive list, as mostly I'm here to answer the question of whether or not the RTX 3060 12GB makes sense as a GPU purchase in today's market.
NVIDIA GeForce RTX 3060 12GB Gaming Benchmarks
Testing was conducted on 17 games, three with ray tracing enabled: A Quiet Place: The Road Ahead, Black Myth: Wukong, Borderlands 4, Control Ultimate, Cyberpunk 2077, Dragon Age: The Veilguard, Final Fantasy XVI, Flight Simulator 2024, God of War: Ragnarok, Indiana Jones and the Great Circle, Mechwarrior 5: Mercenaries, Monster Hunter Wilds, Spider-Man 2, S.T.A.L.K.E.R. 2, Vampire the Masquerade: Bloodlines 2, Warhammer 40,000: Space Marine 2, and Witchfire. Each game was tested at 1080p medium, 1080p and 1440p ultra, and 1440p ultra with 67% upscaling.
Starting with the 1080p medium results, this is the weak spot for the RTX 3060 12GB. Memory capacity isn't a significant problem at these settings, and so the older architecture and lack of raw compute come to the forefront. It's at the bottom of the chart for the tested GPUs — though of course there are plenty of other GPUs that will perform worse — while cards with more compute like the RTX 5060 Ti 8GB rise to the top.
Not surprisingly, with lower performance and a price that's not all that low, that also means the RTX 3060 12GB ranks at the bottom of this list for overall value. If you already have such a card, that doesn't mean it's bad or that you should replace it. Rather, this just shows that paying $275 (give or take) for such a GPU in 2026 is not a great idea. Or at least, it's not a good choice for 1080p medium gaming.
Bumping the quality up to 1080p ultra, things start to shift around. Now the RTX 3060 12GB outperforms both the RTX 4060 and RTX 5050, which end up tied for all intents and purposes. It's only a little bit faster, and we're not using any upscaling or frame generation here, but it does go to show that the old RTX 3060 12GB can offer something that newer GPUs lack. It's still slower than the RTX 5060, but that's generally a more expensive card these days and prices on newer GPUs may go up in the coming months as the memory shortages become more severe.
The value proposition for the RTX 3060 will depend a LOT on what prices you can find at any given time, as well as whether you're looking strictly at brand-new GPUs or used cards. Finding a brand-new 3060 is now possible, and they're priced at $329 and above. The price I used on the other hand comes from the average eBay price over the past month. So, potentially it would be a five-years-old card, which has its own host of concerns, and you'd save perhaps $100 over a new card. Would that be "better" overall? Sometimes, but then I'd say the same about other GPUs like the B580 and 6700 XT.
Gaming at 1080p altra was at least mostly viable, but 1440p ultra proves a bit too much for most of these GPUs. The RTX 3060 12GB averages just under 30 FPS, and while its minimum FPS ends up higher than the RTX 5060, it's really a tie between the 3060, 5060, and B570. More problematic from the gaming perspective is that nine of the games — over half — can't even average 30 FPS or more, with six of the games not even breaking the 20 FPS threshhold. Obviously, dropping the quality to high would help, as would enabling upscaling, which we'll get to momentarily. But for native 1440p ultra, you'll want a more potent graphics card.
The value proposition remains skewed, based on what price you happen to use and/or find. At $272, the 3060 ranks above all the other NVIDIA GPUs, and faster models definitely aren't going to improve the situation, as prices scale much more than performance. It also ranks below the two more recent Intel Arc offerings, along with AMD's old RX 6700 XT. And again, if I were to include the prospect of a new 3060 at $330+, the value would be worse than the other seven GPUs included here.
So what about upscaling performance at 1440p? I tested with 67% rendering resolution, so a 2.25X upscaling factor, aka "Quality mode" — not counting Intel's XeSS 1.3 and later where 67% was renamed "Ultra Quality" in order to boost performance and competitiveness. Overall, the chart isn't really all that different from the 1080p ultra native scaling result; it just has a few minor changes in position even while the individual GPUs remain within the same performance bracket. If you have a GPU that's playable at 1080p ultra native, it will give about the same resulting performance at 1440p ultra with quality upscaling. And that makes sense: The upscaling introduces a bit of overhead, but since the rendered resolution is less than 1920x1080 (67% of 1440p is 1706x960, give or take a few pixels either way), the difference isn't worth splitting hairs over.
Arguably, 1440p with high settings and balanced mode (58% render resolution) upscaling might make more sense. That would render the game at 1485x835 resolution and upscaling to 1440p, and dropping from ultra to high settings typically cuts VRAM requirements by a few GB and gives a modest boost to performance. In other words, 1440p high with balanced upscaling probably runs about as well as 1080p medium native. But I digress....
The value proposition doesn't change much either. At $250, give or take, the RTX 3060 12GB looks pretty interesting, even while AMD's 6700 XT takes top honors for roughly the same price. But DLSS looks quite a bit better than FSR2 and FSR3, so it's not an entirely fair comparison. Which is also why I don't bother trying to quantify frame generation performance. 60 FPS with frame generation looks and feels and plays worse than 60 FPS without frame generation. It's better than 30 FPS without frame generation, I would say, but not much better than 40 FPS — and that ends up being a very subjective assessment, no matter how you slice it.
NVIDIA GeForce RTX 3060 12GB Final Thoughts for 2026
Is the relaunch of the RTX 3060 12GB in 2026 a good thing? No. No, it's not. It's a money grab at best, and a way to offer an old product to clear out old components in practice. The cards should be fine, if you really want one, but you're then giving up newer features and efficiency improvements. Had NVIDIA dropped the price to $250, it would be far more palatable. At the same nominal price as an RTX 5060, though, if you're looking at NVIDIA GPUs you should get the latest model rather than five years old lukewarm leftovers.
As I noted earlier, this doesn't mean the RTX 3060 is a bad card if you already have one; far from it. It's currently the best representation of the xx60-class GPUs from NVIDIA. The newer 4060 and 5060 both made undesirable cuts to VRAM capacity that hinder their performance. That goes for the whole of the 40-series and 50-series lineups, really, outside of the halo 5090 and 4090 models. And you can see what has happened to pricing on those particular GPUs! The 5090 right now sells for over $4,000 brand-new, and even on eBay you're looking at an average price of around $3,600.
If you bought an RTX 3060 12GB back in 2021–2022 for close to MSRP, pat yourself on the back and keep using it would be my advice. Maybe some day in the future you'll have to upgrade, but it's still a viable GPU for gaming in 2026. Just don't expect it to handle native 1440p or 4K at maxed out settings and 60 FPS.
For those looking for other options, both AMD's RX 6700 XT and Intel's Arc B580 are worth a close look. The 6700 XT suffers from similar concerns to the 3060 — it's over five years old and many of the used cards were almost certainly used for Ethereum mining for at least a year or two, which could cause some wear and tear and instability. (Trust me, I know this from experience...) But if you're willing to take a chance, you could do worse.
The real issue right now is with AI and its voracious appetite for memory and storage. Prices on DRAM and SSDs have skyrocketed to about 4X what they used to be a year or two back. That's having a knock-on effect for GPUs as well, and recently NVIDIA has seemingly unofficially increased the prices on its RTX 50-series Blackwell offerings — or at least its retail partners like Best Buy have increased prices. That means if you're looking to build a decent budget or midrange gaming PC today, you're going to end up spending substantially more than a similar PC would have cost last year. You might not even get the same level of performance for your troubles.
When will the AI-induced shortages end, and when will prices return to normal? It's a guessing game on both topics, but quite a few people are predicting at least 2030 before the supply of memory and storage can catch up to the demands from data centers. If that proves correct, we've got a dismal four years to get through before things start looking better. Yuck. So hang on to your older PC parts and hopefully game developers won't push for more powerful hardware in the interim.









No comments:
Post a Comment