Nvidia RTX 4060 Specs Leak Claims Fewer CUDA Cores, VRAM Than RTX 3060

GeForce RTX 3060 Ti
GeForce RTX 3060 Ti (Image credit: Nvidia)

Nvidia's forthcoming GeForce RTX 4060 aims to compete with the best graphics cards for gaming. However, if the rumored specifications are accurate, the GeForce RTX 4060 may disappoint a lot of gamers.

Nvidia's GeForce RTX x060-tier graphics cards have always been a fan favorite for budget-conscious gamers. The latest Steam Hardware Survey can attest to the statement. The last three generations of GeForce RTX x060-series models are among Steam's top ten most used graphics cards. The GeForce GTX 1060 and GeForce RTX 2060 are currently the second and fourth most popular graphics cards, while the GeForce RTX 3060 finds itself in sixth place. Although the Ada Lovelace architecture offers a significant upgrade over Ampere, the GeForce RTX 4060's potential specifications look disheartening.

According to hardware leaker kopite7kimi, who has a solid track record of reliable information, the GeForce RTX 4060 reportedly uses the AD107 silicon. It's a die that we already know about because Nvidia utilizes it for the company's mobile GeForce RTX 4060 and GeForce RTX 4050 SKUs. So yes, barring any changes, the GeForce RTX 4060 appears to use a die tailored to mobile graphics cards.

The GeForce RTX 4060 seemingly employs the PG190 PCB with the AD107-400-A1 at heart. That means the Ada-powered graphics card will wield 3,072 CUDA cores, 96 Tensor cores, and 24 RT cores. The last-generation GeForce RTX 3060 had 3,584 CUDA cores, 17% more than the GeForce RTX 4060. Both use different architectures, of course.

Nvidia GeForce RTX 4060 Specifications

Swipe to scroll horizontally
Graphics CardGeForce RTX 4060*GeForce RTX 3060
Process TechnologyTSMC 4NSamsung 8N
Transistors (Billion)?12
Die size (mm²)146276
GPU Cores3,0723,584
Tensor Cores96112
RT Cores2428
Base Clock (MHz)?1,320
Boost Clock (MHz)?1,777
VRAM Speed (Gbps)1815
VRAM Bus Width?192
TFLOPs FP32 (Boost)?12.7
Bandwidth (GBps)?360
TGP (watts)115170
Launch Date20232021

*Specifications are unconfirmed.

The GeForce RTX 4060 may arrive with 8GB of 18 Gbps GDDR6 memory. The memory modules are reportedly faster since the regular GeForce RTX 3060 uses 15 Gbps modules. However, the GeForce RTX 3060 has 12GB, apparently, 50% more than the GeForce RTX 4060. It's a substantial compromise that'll ultimately affect performance in modern titles, such as Hogwarts Legacy, that can devour VRAM. 

The leaker didn't share the memory interface for the GeForce RTX 4060, but there aren't too many options. The GeForce RTX 4060 is unlikely to retain the 192-bit memory bus, which has been the norm since the GTX 1060. That means the GeForce RTX 4060 may only offer up to 288 GB/s of bandwidth, representing a 20% less than the GeForce RTX 3060.

The GeForce RTX 4060 will receive an L2 cache upgrade because of the Ada Lovelace architecture. The hardware leaker pegged the GeForce RTX 4060 with a 24MB L2 cache, 700% larger than the one on the GeForce RTX 3060. If we compare the GeForce RTX 4060 to the other Ada-based SKUs that have launched, it sports half the L2 cache of the GeForce RTX 4070 Ti. Given the other Ada GPUs have up to 8MB of L2 cache per 32-bit interface, it would appear that the RTX 4060 is also cut down in that area as well.

Maybe Nvidia's decision to roll with mobile silicon was to lower the power requirement for the GeForce RTX 4060. Kopite7kimi believes that the GeForce RTX 4060 will have a 115W TDP. That's crazy because it would put the GeForce RTX 4060 almost on the same level as a GeForce GTX 1660 (120W) in terms of TDP. For comparison, even the GeForce RTX 4090 Mobile (120W) consumes more power. The GeForce RTX 3060 sports a 170W rating, so the GeForce RTX 4060 has a 32% lower TDP.

Overall, if the leaks are correct, RTX 4060 is shaping up to be a rather odd solution. The lack of VRAM will certainly prove limiting, and while the L2 cache could make up for the loss in bandwidth, it still ends up feeling like less of an upgrade than other models. But then we said the same thing about the RTX 4070 Ti with its 192-bit interface. Time will tell how the RTX 4060 actually stacks up, and how much it will cost.

Zhiye Liu
RAM Reviewer and News Editor

Zhiye Liu is a Freelance News Writer at Tom’s Hardware US. Although he loves everything that’s hardware, he has a soft spot for CPUs, GPUs, and RAM.

  • Metal Messiah.
    ""The GeForce RTX 4060 may retain a 192-bit memory bus, which has been the norm since the GTX 1060.""
    That seems unlikely. The GPU will sport a 128-bit bus interface for sure, same as the rumored RTX 4060 Ti SKU (also because it will come in an 8GB VRAM flavor). That gives the card 288 GB/s bandwidth and all of that will fall within a 115W reference TDP.
  • cyrusfox
    14% less cores, 2/3 the VRAM (12gb was generous though on the 3060). This is what comes to mind
  • blue screen of fun!
    Seems like an odd choice to make an underpowered 4060.
    On the other hand, releasing a 4050 with the mobile GPU could be a really great option. I'd love to have a relatively inexpensive single slot card that I could add to some of the underpowered workstations around my workplace. The only options for that are either insanely expensive or very old...
  • AgentBirdnest
    Less VRAM isn't really surprising. The 3060's 12GB was an interesting, somewhat bizarre choice, considering that the 3070 and OG 3080 only had 8 and 10 GB, respectively.

    If it's true that it has 14% fewer cores than the 3060, though... dang, that's just sad. :-/ Higher clocks will probably make it perform similarly, but not for a cheaper price.

    It feels like the entire RTX 40 series will be a mulligan. I hope it will be, and that Nvidia learns a lesson and gets back to actual progress next year, but I'm not gonna hold my breath.
  • PlaneInTheSky
    Just 8GB VRAM in 2023, wow.

    A game like Hogwarths Legacy uses more in 1080p.

    And with consoles having 16GB shared VRAM, 8GB is a serious problem for any cross platform title.

    This card better be around $200 or something, because it's already behind the curve.

  • InvalidError
    PlaneInTheSky said:
    A game like Hogwarths Legacy uses more in 1080p.
    Only if you push it to 1440p Ultra RT where the RTX3060 only does 26FPS on average, not really playable. If you lower details enough to get 50+FPS average, then 8GB GPUs like the 3070Ti beat the 12GB RTX3060, VRAM is no longer the main limiting factor.

    Seems like 8GB is just about right for the amount of details that can be pushed out with the 3060's compute power.

    The worst part about the RTX4060 will likely be its MSRP which I wouldn't be surprised if it was $500, which would be absolutely atrocious value per dollar for what should be 50-tier hardware.
  • PlaneInTheSky
    Only if you push it to 1440p Ultra RT

    hello, anybody home?

  • InvalidError
    PlaneInTheSky said:
    hello, anybody home?
    How much VRAM usage a game or hardware monitoring software reports has no direct correlation with how much of the VRAM's buffer is actually used to render frames. You have to look for the point where performance drops off a cliff from exceeding available VRAM for that.
  • PlaneInTheSky
    How much VRAM usage a game or hardware monitoring software reports has no direct correlation with how much of the VRAM's buffer is actually used to render frames.

    You are wrong again. It is used to render frames, and it does impact in-game FPS.

    That 3060 12GB VRAM is beating that 3080 10GB and 3070 8GB in-game, in 1080p, with higher FPS.

    I suggest you stop now, admit your mistake, before digging deeper.

    (source: Hardware Unboxed)

  • neblogai
    PlaneInTheSky said:
    That 3060 12GB VRAM is beating that 3080 10GB and 3070 8GB in-game, in 1080p, with higher FPS.

    That FPS count just means you need to turn off RT, or lower other settings, as this tier of GPU could not run well with these settings at good fps, even if it had enough VRAM (as the 306012GB shows). And then, without RT/Ultra settings, 8GB will be fine, even in a such an outlier title as HL.