Gen-over-gen performance uplifts are getting smaller with Nvidia's mid-range GPUs — GTX 1060, RTX 2060, RTX 3060, and RTX 4060 retested in 2024

News
By
published

The transition from the GTX 1060 to the RTX 2060 exhibited the most significant performance improvement.

GeForce RTX 4060 Ti
GeForce RTX 4060 Ti (Image credit: Tom's Hardware)

ComputerBase re-tested four generations of Nvidia GTX/RTX xx60 series GPUs, among the best graphics cards, to see the performance difference between each GPU in today's modern games. The German tech review outlet discovered that the generational leap of the RTX 2060 was by far the greatest, with the Turing GPU outperforming the GTX 1060 by 79% on average. ComputerBase also discovered that the RTX 4060's 8GB memory size handicaps its performance in several modern titles.

The German publication's results are normalized to 100% of the GTX 1000 series. The relative performance improvement is compared to the previous generation. The most significant generational improvement of the four cards comes from the RTX 2060, which offers a whopping 79% improvement compared to the GTX 1060. This was the average performance gap discovered in the twelve titles it tested.

The next most considerable generational improvement comes from the RTX 4060, but it is nowhere near what the RTX 2060 delivers. The RTX 4060 is only 22% faster than the RTX 3060 12GB, primarily because of the RTX 4060's 8GB VRAM capacity. Finally, ComputerBase found the lowest generational improvement with the RTX 3060, just 17% faster than the RTX 2060 average.

Swipe to scroll horizontally
Header Cell - Column 0 Performance LevelRelative Improvement
RTX 4000256%22%
RTX 3000209%17%
RTX 2000179%79%
GTX 1000100%-

The GTX 1060 was the second to last xx60-series card from Nvidia to come with the GTX branding and thus no hardware-accelerated ray-tracing capabilities (or DLSS). The GPU came with Nvidia's Pascal GPU architecture, rocking 1280 CUDA cores, 10 SMs, and 8GB of GDDR5 memory operating on a 192-bit wide bus. The RTX 2060 was the first Nvidia GPU with hardware-accelerated ray tracing and AI-tensor cores (for powering things such as DLSS upscaling). The GPU was built on Nvidia's Turing architecture, sporting 1,920 CUDA cores, 30 SMs, 30 RT cores, 240 Tensor cores, and 6GB of GDDR6 memory operating on a 192-bit bus.

The RTX 3060 12GB was the second xx60-class Nvidia GPU with RT and AI-accelerated hardware. The GPU was based on Nvidia's Ampere architecture, featuring 3,584 CUDA cores, 28 SMs, 112 Tensor Cores, 28 RT cores, and 12GB of GDDR6 operating on a 192-bit bus. The RTX 4060 is Nvidia's latest-generation xx60 series product, sporting its Ada Lovelace GPU architecture, 3,072 CUDA cores, 24 SMs, 96 Tensor Cores, 24 RT cores, and 8GB of GDDR6 operating on a 128-bit bus.

(Image credit: ComputerBase)

The per-game analysis makes the RTX 2060's huge performance gap to the GTX 1060 even more apparent. In Diablo II: Resurrected and Doom Eternal, the RTX 2060 was 226% and 86% faster than its Pascal predecessor. In Horizon Zero Dawn, the RTX 2060's immense performance enabled the card to achieve over 60 FPS, while the GTX 1060 couldn't even reach 45 FPS. Rachet and Clank: Rift Apart is the worst offender for the GTX 1060, with the RTX 2060 offering playable performance at 52.9 FPS and the GTX 1060 offering unplayable performance at 26.3 FPS.

However, things get complicated regarding the RTX 3060 and RTX 4060 analysis, thanks to the vast VRAM difference between the two GPUs. In games where VRAM isn't an issue, the RTX 4060 has the advantage, often outperforming the RTX 3060 by around 30%. But in Rachet and Clank: Rift Apart, the game's insatiable demand for VRAM overwhelms the RTX 4060, enabling the RTX 3060 12GB to match the RTX 4060 average frame rate. However, in actual playability, the RTX 3060 far outperforms the RTX 4060, offering nearly twice the 1% percentile performance. This means the RTX 4060 is succumbing to hitching or micro stutter, while the RTX 3060 is unaffected.

ComputerBase's results reveal that the RTX 2060 6GB has had a surprisingly good lifespan since it debuted in 2018. The architectural changes Nvidia made in RTX 2060 have played off exceptionally well, enabling the GPU to perform surprisingly well in non-VRAM-intensive games and be close to the RTX 3060. On top of this, it also has DLSS support, allowing the owners to extract even more performance from the RTX 2060.

However, the same cannot be said of the RTX 4060. Despite being released just a year ago, the RTX 4060 has problems with some modern titles thanks to its VRAM reduction compared to the previous generation. We can only expect this problem to escalate as games become more VRAM-intensive.

See more GPUs News
Aaron Klotz
Aaron Klotz
Contributing Writer

Aaron Klotz is a contributing writer for Tom’s Hardware, covering news related to computer hardware such as CPUs, and graphics cards.

10 Comments Comment from the forums
  • Notton
    Look, we all know the 4060 is actually a 4050Ti.
    Otherwise it would have had a 192-bit bus, just like all the previous xx60 class of GPUs.
    Reply
  • Phenomena
    Just to get this straight in my head. When the 8GB 3060 was released, it was able to compete with the 2080, and beat it in most cases. The 2080 is much more than 17% faster than the 2060.

    So what's going on?
    Reply
  • TJ Hooker
    Phenomena said:
    Just to get this straight in my head. When the 8GB 3060 was released, it was able to compete with the 2080, and beat it in most cases. The 2080 is much more than 17% faster than the 2060.

    So what's going on?
    You're misremembering. Even the original 12GB 3060 was noticeably slower than the 2080. And the eventual 8GB version was slower still, due to having 33% less VRAM capacity and bandwidth.
    Reply
  • cryoburner
    This comparison was very poorly done, as it completely ignores the fact that Nvidia has shifted around product names to cards at different price points in an attempt to push people to spend more on a graphics card than they would have otherwise. Without looking at the prices of cards, these performance comparisons are pointless.

    Comparing a 2060 against a 1060 is nonsensical, because the 2060 was positioned as more of a 1070 successor than anything, while the 1660 cards were the actual direct successors to the 1060. The 1060 3GB and 1060 6GB were $200 and $250 cards respectively, whereas the 2060 was a $350 card, and the 2060 SUPER was $400, both roughly in line with the 1070's $380 MSRP. Performance-wise, the 2060 was only around 15% faster than a 1070 in games at the time, but with 25% less VRAM, and overall, the performance gains of the 20-series were poor across the board. In today's games, the 2060's newer architecture likely gives it more of an edge in many titles, but the lower VRAM of the original 2060 compared to the 1070 likely hampers it in others.

    The 3060's performance gains over the 2060 were also poor, but it was also supposedly intended to be positioned at a slightly lower $330 price point, even if that never happened due to crypto-mining and other shortages having already driven up graphics card prices to absurd levels even before the cards had a chance to launch.

    The 4060's gains were very underwhelming, but there was again a slight reduction in MSRP, which when combined with the recent high inflation rates actually positions the card around a similar price level to where the 1060 cards were 7 years prior. Of course, it could certainly be argued that they have simply tacked an x60 name and price tag onto a card that would have traditionally been marketed as an x50-class card, but at least price-wise, comparing a 4060 against a 1060 might be relatively fair.
    Reply
  • hotaru251
    Notton said:
    Otherwise it would have had a 192-bit bus, just like all the previous xx60 class of GPUs.

    yup.

    the 960 was last 60tier to have a 128bit bus (even the god awful nobody should buy it gtx 1060 3gb had 192)

    i am hoping (so much copium) Nvidia saw how bad 4060 sold and gives full 192bit bus on 5060.


    The only sku that needed a "super" refresh was the 4060. the gap between 4060 and 4070 is so massive and entirely due to that bus limitation.

    cryoburner said:
    , because the 2060 was positioned as more of a 1070 successor than anything, while the 1660 cards were the actual direct successors to the 1060.
    not really.

    every generation generally sees this where the last gen is 1 tier lower than new gen (2060 to 1070, 3060 to 2070 & the 3090 to 4080)
    The 16 series was the 2060 if you didnt care about ray tracing premium pricetag. (and ur not losing that much raster performance for the price dif)

    Thats a healthy improvement generation to generation....the 4060 beign worse than the 3060 in raster w/o dlss 3/FG was a clear sign that it was a 50 tier gpu in reality.
    cryoburner said:
    The 4060's gains were very underwhelming, but there was again a slight reduction in MSRP
    and the 30 series was cheaper than 20 series yet improvement was huge across board.

    1060 6gb was $250 launch in 2016 which would be around $325 today w/ inflation...a 4060 msrp is 299.
    however a 3060 (which again can beat a 4060 in raster w/o dlss & fg) cost $329 in 2021 but with inflation would be $390 today.

    the 4060 is bad deal price & performance wise.
    Reply
  • NedSmelly
    cryoburner said:
    Comparing a 2060 against a 1060 is nonsensical, because the 2060 was positioned as more of a 1070 successor than anything, while the 1660 cards were the actual direct successors to the 1060. The 1060 3GB and 1060 6GB were $200 and $250 cards respectively, whereas the 2060 was a $350 card, and the 2060 SUPER was $400, both roughly in line with the 1070's $380 MSRP. Performance-wise, the 2060 was only around 15% faster than a 1070 in games at the time, but with 25% less VRAM, and overall, the performance gains of the 20-series were poor across the board. In today's games, the 2060's newer architecture likely gives it more of an edge in many titles, but the lower VRAM of the original 2060 compared to the 1070 likely hampers it in others.
    The 2060 Super could almost be considered a "2070 Lite", as its performance was within a hair's breadth of the 2070 in many scenarios. I think it might also be the only 60-series with a full 256-bit memory bus.
    Reply
  • thestryker
    While I think there's merit comparing 60 series cards FPS/$ scales are important to add rather than just raw performance. I think the 60 series mostly shows how mediocre the 3060 was compared to the higher Ampere cards.

    based on their 1080p numbers at MSRP (% is reduction in cost/frame compared to prior generation):
    Doom Eternal:
    $3.29/frame - 1060
    $2.48/frame - 2060 - 24.6%
    $1.94/frame - 3060 - 21.8%
    $1.39/frame - 4060 - 28.4%

    Horizon Zero Dawn:
    $5.75/frame - 1060
    $4.68/frame - 2060 - 18.6%
    $3.76/frame - 3060 - 19.7%
    $2.72/frame - 4060 - 27.7%

    compared their numbers for the 70 series:
    Doom Eternal:
    $3.62/frame - 1070
    $3.17/frame - 2070 - 12.4%
    $1.90/frame - 3070 - 40%
    $1.68/frame - 4070 - 11.6%
    Reply
  • cryoburner
    hotaru251 said:
    not really.

    every generation generally sees this where the last gen is 1 tier lower than new gen (2060 to 1070, 3060 to 2070 & the 3090 to 4080)
    The 16 series was the 2060 if you didnt care about ray tracing premium pricetag. (and ur not losing that much raster performance for the price dif)

    Thats a healthy improvement generation to generation....the 4060 beign worse than the 3060 in raster w/o dlss 3/FG was a clear sign that it was a 50 tier gpu in reality.
    Again, the entire point of this article is about comparing the performance of cards between generations, but it doesn't make any sense to be comparing cards at totally different price points that are not even close to one another. If a card is released at a $350 MSRP, it makes the most sense to compare it to the card preceding it at a similar $380 MSRP, not the one that was $250. It costs 40% more than that card, so obviously one would expect it to be substantially faster. When the 20-series cards launched, they were widely seen as mostly mediocre sidegrades to existing models, with only minor improvements in terms of performance at any given price point, with the main draw being features like raytracing that weren't even available in any games for some time after launch.

    hotaru251 said:
    and the 30 series was cheaper than 20 series yet improvement was huge across board.

    1060 6gb was $250 launch in 2016 which would be around $325 today w/ inflation...a 4060 msrp is 299.
    however a 3060 (which again can beat a 4060 in raster w/o dlss & fg) cost $329 in 2021 but with inflation would be $390 today.

    the 4060 is bad deal price & performance wise.
    The 3060 was never available for $329 at launch, or at any time in 2021. The launch happened during the peak of the crypto-mining shortages, and Nvidia knew full well when announcing that price that it was nothing more than fiction, and the card wouldn't be available for anywhere close to that. Even the manufacturer MSRPs were more in the vicinity of $400 at launch, and only went up from there, with average reseller prices hovering around $700 throughout the year.

    The 4060, on the other hand, has been widely available for around $300 or so since launch. Though of course, with less VRAM and mostly similar performance to the 3060, it's actually more like what should have been marketed as a 3050 successor, only at an inflated price point. But for the purposes of comparing generational performance gains, comparing cards at similar price levels makes the most sense, not whatever random names Nvidia decided to assign to them.
    Reply
  • 35below0
    The 4060 is not a performer in benchmarks, but it *is* good value. Simply because the 70 and 80 are so expensive. They perform better but tons of games will run fine on a 4060, esp. at 1080p but also 1440p. Even Cyberpunk lists only a 2060 Super as it's recommended req.

    You won't get Ultra settings at 1440p/4k from Cyberpunk, or even Baldur's Gate 3. But 4060 is a budget card, and it... well kinda represents good value unless you're willing to drop $550-700 on a more powerful 7800XT or 4070 Ti or Super

    It's very powerfull, just obliterated by it's more powerful cousins and rivals. Those are not budget GPUs though.

    edit
    If anything it's the Ti and expanded 4060s that have crap value. They cost much more and are basically the same, entry level GPU.
    If the vanilla 4060 is not good enough, and the 4070 Ti too expensive, just shop AMDs cards.
    Reply
  • cryoburner
    35below0 said:
    But 4060 is a budget card, and it... well kinda represents good value unless you're willing to drop $550-700 on a more powerful 7800XT or 4070 Ti or Super

    It's very powerfull, just obliterated by it's more powerful cousins and rivals. Those are not budget GPUs though.
    $300 for something considered a "Budget" GPU seems a bit of a stretch. Not that long ago cards around this price were considered well within the "mid-range", even accounting for inflation.

    35below0 said:
    If anything it's the Ti and expanded 4060s that have crap value. They cost much more and are basically the same, entry level GPU.
    If the vanilla 4060 is not good enough, and the 4070 Ti too expensive, just shop AMDs cards.
    The 4060 Ti is a notably different card from the 4060, in that it uses a larger graphics chip with 50% more cores (~42% more enabled). So the two cards sharing the same base name doesn't make all that much sense, going back to my prior point about how cards shouldn't be compared between generations based on arbitrary model numbers rather than price.

    They both have a similarly-limited memory system though, with the same 128-bit bus, and the Ti only gets slightly faster memory, which often prevents the card from stretching its legs and performing like its higher core count might otherwise suggest. There are some demanding titles where the 4060 Ti can be over 30% faster than a 4060, particularly when things like raytracing are in use, but those tend to be few and far between, with the typical gains often being under 20%. Both cards are arguably a bit overpriced for what they offer though. And the 16GB version of the 4060 Ti was way overpriced, and should have been priced at the $400 that the 8GB version goes for. The 8GB version could have been a more reasonable $350. And the 4060 should have arguably been marketed as a "4050 Ti" at a somewhat lower price point as well.
    Reply