ASUS GeForce RTX 4060 Dual OC

[TheoneandonlyMrK]

That's where it's all haywire. If Ngreedia underclocked it and name it what it truly should be and priced it accordingly as a 50, it would be amazing. But as a 60?

No doubt, With the interface I actually think these designs were for cheap laptops and the 4050 but Intel and AMD have the low end laptop covered so we get This.

It's Clear, Huang gives Zero shits , he doesn't care, and thinks Nvidia has a apple like fanbase that will swallow what they're told to, and like it.

And it matters little, due to AI orders if this doesn't sell it will be repackaged next year as a 5050 in all likelihood, Nvidia has amassed a lot of inventory they need to shift on.

 

[wNotyarD]

It's Clear, Huang gives Zero shits , he doesn't care, and thinks Nvidia has a apple like fanbase that will swallow what they're told to, and like it.

Worst part is he's mostly right.

 

[N/A]

Exactly. Well said. The 4060 is a terrible card. Just shrink the 3060 12GB and add faster memory and in every way it is superior. Vastly so

3060 port to 5nm would be less than 99mm2 stitched to a 96 bit bus, this is why they have to make it bigger and full of other stuff. GA103 the blueprint backport of AD104 lacking the L2 cache and DLSS 3 will have to be used instead. And if you do that might as well go for the AD104.

So unfortunately if I seek 4060-ish performance, I have to look for a 4070 and pay double of what this tier used to be. No choice there.

 

[LuxZg]

3060 port to 5nm would be less than 99mm2 stitched to a 96 bit bus, this is why they have to make it bigger and full of other stuff. GA103 the blueprint backport of AD104 lacking the L2 cache and DLSS 3 will have to be used instead. And if you do that might as well go for the AD104.

So unfortunately if I seek 4060-ish performance, I have to look for a 4070 and pay double of what this tier used to be. No choice there.

There's plenty choice, unless you just HAVE TO have latest Nvidia card.

 

[Cowboystrekk]

Who would have thought that Asus budget offering beat all other coolers except for the Strix? Superior on both noise and temps at msrp! For a "meh" card that the 4060 actually is, this sure is the star of the show.

 

[Taisho]

Who would have thought that Asus budget offering beat all other coolers except for the Strix? Superior on both noise and temps at msrp! For a "meh" card that the 4060 actually is, this sure is the star of the show.

Many manufacturers suddenly have an issue to launch an over-MSRP variant that weights above 600g. Toy cooling designs, the die cut into nothingness, and all models go well over the specified power draw. So where is the $$$ saved?

 

[Assimilator]

How are they not? Looking at all the 4060s reviewed here at TPU, only the 2.5-slot Asus is cold.
With such a small chip, one would think that you shouldn't even need a double-fan cooler.

EDIT: rereading the reviews, Galax's is also as cold. Being as loud as a jet turbine.

The coolers don't need to be that quiet, or keep temperatures that low. All of the 4060s reviewed, except the Gigabyte one, could easily justify 5 dBa higher noise output at a 10 degree higher temperature.

It's not just the coolers though. This card is built to be able to draw 225W but as the review shows, it never gets near 150W even when overclocked, so it could quite easily have got by with a 6-pin PCIe power connector instead of an 8-pin. But 6 is less than 8, so the marketers give us 8.

Then due to Buildzoid's stupid whining about OMG THE COMPONENTS ON THIS CARD ARE RUBBISH because all he cares about is overclocking things to death where component quality actually matters, the marketers put overbuilt components on a card that will never require anywhere near what those components can do for the vast majority of users who will purchase that card.

Then they recoup some of those costs by making the PCB as small as possible via clustering all those heat-producing components into the tiniest possible space, which means that when one component gets hot they all get hot, which has a cascading effect, which means the card now needs a better cooler to soak all of the heat emitted by said components.

The 8800 GT of 2007 drew around the same power that this card does, yet managed to do it with a single-slot cooler that was literally nothing more than a slab of aluminium (no copper or heatpipes there). And it could do that because it had a PCB large enough to space components out, so that they couldn't soak each other to thermal runaway. But these basic laws of physics and engineering don't matter when marketers get involved.

 

[AnotherReader]

3060 port to 5nm would be less than 99mm2 stitched to a 96 bit bus, this is why they have to make it bigger and full of other stuff.

If you're naively using the square of 5/8 as your scaling factor, then your assumptions are incorrect. First of all, the analog portion wouldn't have shrunk at all with the move to TSMC. In a 3060, the analog portion is about 49.5 mm^2 compared to 34.5 mm^2 for AD107. Chopping off the L2 and increasing the SM count from 24 to 30 would probably bring the chip close to 130 mm^2 or so. Still, that would have been faster than AD107 at the expense of drawing more power due to the increased SM count, wider bus, and more GDDR6X devices.

It seems like there's some rose-colored glasses happening here. The bump from 3000-series is disappointing, but it's not exactly outside historical norms. Consider:

Model	+% vs. prev at 1080p	Watts (rated)	Price
760	20	170	250
960	10	120	200
1060	50	120	250
1660	20	120	219
2060	35*	160	350
3060 12GB	15	170	330
4060	15	115	300

*vs 1060

Observations:

• 960 was only 10% up on 760, but did so for $50 less and 50 fewer watts
• Pascal was a HUGE leap for a single generation, to a degree not seen before or since, enabling the massive jump from 960 to 1060
• The 2060 made 35% over the 1060, but at the cost of an extra $100 and 40 watts
• 4060 makes the same margin over the 3060 as that card does over the 2060, but does so at 2/3 the power

Conclusion: We're remembering how awesome Pascal was, and for some reason expecting that lighting to strike every generation. Yeah, the 4060 should cost $50 less than it does, but on a technical level it's not exactly the turd some are making it out to be.

There are some errors in your calculations. The figures, according to TPU's launch day reviews are:

Model	Increase (%) over predecessor at 1080p	Watts (rated)	Price
GTX 560 Ti	30	170	250
GTX 660	25	140	230
GTX 760	23	170	250
GTX 960	10	120	200
GTX 1060	104	120	250
GTX 1660	35	120	220
RTX 2060	56 (vs 1060)	160	350
RTX 3060	18 (vs 2060)	170	330
RTX 4060	18 (vs 3060)	115	300

Expecting a Pascal like improvement was unrealistic, but another 10 percent would have been closer to the long term trend.

 

[tussinman]

Conclusion: We're remembering how awesome Pascal was, and for some reason expecting that lighting to strike every generation. Yeah, the 4060 should cost $50 less than it does, but on a technical level it's not exactly the turd some are making it out to be.

Yeah don't agree with that at all. The 2060 was like 1.5x the performance and went toe to toe with the previous gens xx80 (if pascal was this mythical generation, then how did a 2060 match a 1080 ?).

The 3060 went toe to toe with a 2070 and increased the ram over the 2060 by 2x (12gb vs 6gb)

Face it the 4060 is a joke, this isn't a case of lightning not striking (they removed 4GB of ram, lowered the bus, and then made it 20% slower than the previous gens xx70 tier). It's a xx50 series die/board that's being sold as a 60 tier

 

[Cowboystrekk]

4060 is barely faster, sometimes slower, has less BW and ram, but is more efficient. Efgociency and framegens are the only pros.

 

[Assimilator]

they removed 4GB of ram, lowered the bus

8GB of VRAM isn't the biggest problem with this card.

 

[tussinman]

8GB of VRAM isn't the biggest problem with this card.

Agreed, that's why I also wrote they lowered the bus and made it 20% slower than the previous gens xx70

 

[N/A]

First of all, the analog portion wouldn't have shrunk at all with the move to TSMC. In a 3060, the analog portion is about 49.5 mm^2 compared to 34.5 mm^2 for AD107. Chopping off the L2 and increasing the SM count from 24 to 30 would probably bring the chip close to 130 mm^2 or so. Still, that would have been faster than AD107 at the expense of drawing more power due to the increased SM count, wider bus, and more GDDR6X devices.

View attachment 302862

So a 4060 has 29MB more L2 than 3060. 2,65mm2/MB vs 7,55mm2/MB and if we get rid of that 76,85mm2 we are left with 79 mm2, that's a GT 1030 on steroids.
Adding 64 bit bus makes it 12,5 mm2 bigger. add 768 CUDA cores that is next to nothing. you can work it out to 17 - 22mm2 with 24 RT and 6 tensor depending on what the actual size is 146 or 156.
I'm pretty sure you still can't get it any bigger than 99 mm2 128 bit with all the tensor cores and RT being updated for frame generation, path tracing and all.

 

[W1zzard]

Summary video comparing all 10 cards

 

[Andi5]

Can it really be that the Asus Dual OC is quieter than the much more expensive Asus Strix OC with its monster cooler and 3 fans? (both in Quiet BIOS)

 

[N/A]

Apparently 3 fans running 100 lower RPM are 1,2 dBA noisier than just 2 fans. Both whisper quiet.

 

[darkangel0504]

Forget RTX 4060, just bought a RX 6700XT to replace my RTX 2060 6 GB, maximum temp around 64C, ambient temp at 35C
Spider-man and the last of us just eat 10 GB VRAM. Never buy a Nividia product again.

 

[zatarcz]

Does it make sense to upgrade from the 2060 SUPER?

 

[AnotherReader]

Does it make sense to upgrade from the 2060 SUPER?

No, it doesn't. You shouldn't be looking at anything slower than the 4060 Ti or the 6700 XT. Extrapolating from this review and TPU's review of the 3060, the 4060 is roughly 25% faster than the 2060 Super.

 

[Dan.G]

I'm intrigued by the card. It would be a nice upgrade over the 1660 Ti. And it's not all that bad: paid 280 Euro for 1660 Ti back in 2019. With this card I'd pay 20% more for 60% increase in performance. I'm just curios to know if W1zzard noticed any coil whine from the card during testing. The chokes seem to have some sort of stabilizer on the sides.
Upgrade-wise, it's destined to replace the 900 series, 1000 series (except 1080 Ti - more VRAM), 16 series, possibly the 2060 non-SUPER (or even the RTX 3050!).
It's good enough for 1440p@60, from what the graphs show.
Yeah, probably gonna buy the DUAL model: I like the dual-BIOS feature!
Cool'n'Quiet, boys!

 

[Dan.G]

But I still can't shake the feeling that the AD107 has 32 ROPs... Oh, well...

 

[Trindade31102022]

I have the RTX 4060 MSI Gaming X, and it has a high consumption problem when not in use. staying at 50w. I didn't find any solution for this.

Config:
MSI B450M-Mortar MAX
Ryzen 5700x
16gb ram
Win10 PRO

 

[oxrufiioxo]

I have the RTX 4060 MSI Gaming X, and it has a high consumption problem when not in use. staying at 50w. I didn't find any solution for this.

Config:
MSI B450M-Mortar MAX
Ryzen 5700x
16gb ram
Win10 PRO

View attachment 314213

Probably a good idea to make a help thread.

 

[Chrispy_]

I have the RTX 4060 MSI Gaming X, and it has a high consumption problem when not in use. staying at 50w. I didn't find any solution for this.

Config:
MSI B450M-Mortar MAX
Ryzen 5700x
16gb ram
Win10 PRO

View attachment 314213

That looks like sensors not reporting correctly. 0% TDP and 51.8W are mutually exclusive. One of them MUST be wrong, which means the rest of the values being reported are also now questionable.

Honestly, I'm inclined to believe the temperature sensor and fan speed sensor. The air inside your case is likely to be about 30-35C, and your GPU temperature is 39.5C with the fans stopped. Based on simple thermodynamics, you likely have a GPU drawing the 4.3W as reported, and the other readings are bogus.

Simply put, passive cooling with only 5-10C deltas is going to be very low, probably sub-10W power draw. Even 25W would get very hot without airflow from fans, and yours are stopped.

EDIT:
I just remembered that I'm running a 4060Ti in this box, lol.

You can see my temperatures and fan speeds are close to yours, but my board power draw is 4.1W.
That means your temperatures are indicative of a 4W power draw in a typical PC case like mine (3x 120mm intake fans at low RPM, nothing special or fancy)

 

[Trindade31102022]

That looks like sensors not reporting correctly. 0% TDP and 51.8W are mutually exclusive. One of them MUST be wrong, which means the rest of the values being reported are also now questionable.

Honestly, I'm inclined to believe the temperature sensor and fan speed sensor. The air inside your case is likely to be about 30-35C, and your GPU temperature is 39.5C with the fans stopped. Based on simple thermodynamics, you likely have a GPU drawing the 4.3W as reported, and the other readings are bogus.

Simply put, passive cooling with only 5-10C deltas is going to be very low, probably sub-10W power draw. Even 25W would get very hot without airflow from fans, and yours are stopped.

EDIT:
I just remembered that I'm running a 4060Ti in this box, lol.


You can see my temperatures and fan speeds are close to yours, but my board power draw is 4.1W.
That means your temperatures are indicative of a 4W power draw in a typical PC case like mine (3x 120mm intake fans at low RPM, nothing special or fancy)

Strange, it could be that it's just a problem with the 4060, I've seen someone else on YouTube with the same problem.