Thank you for adding chapter timestamps - this effort doesn’t go unnoticed and is much appreciated!

If you rebrand a 50 tier gpu as a 60 tier gpu, it is only natural that its predecessor almost outperforms it

If it can't match the 3070, what's the point? Truly a 4050 at best.

Would’ve been a fantastic $200 RTX 4050

@29:20 A caveat, Intel didn't discontinue the A770, they discontinued making the Limited Edition reference card. The AIB are still going to continue making the A770 AFAIK

16:30 the problem is that now Nvidia is using those features as an excuse to not actually deliver on the silicon level.

Nvidia are desperate to make this a 1080p card because there is actually a situation where the 4060 is worse than the 3060 at 1440p.

@6:28 - the numbers for the RX 7600 don't add up. 158W (corrected from 168W, typo on my end, thank you @KingdaGaertner below) / 124fps ~= 1.27 joules per frame, or ~132% power consumption.

What impresses me the most is the Intel A750 in many of the DX12, it's really strong.

The ARC 770 16GB isn't discontinued? Only the intel branded one is. There are 16GB partner cards.

Ada could've went down as one of the best architectures due to it's great efficiency. Had the 4060 been a 4050 and priced accordingly we'd be talking about a card that many times beats the 3060 all on a die about half the size (even with the onboard cache). ...Imagine if you will a parallel universe where Nvidia and AMD still cared about PC gaming. In that universe we'd be experiencing Polaris vs Pascal competition for your gaming money. ...I think we need to look into portal technology.

This generation, the consoles themselves seem like a much better deal than PC hardware compared to the last two generations.

The only time this GPU gets close to real value is if you play a very limited selection of RT titles and don't mind playing a game that looks faster than it feels.

I think 6700xt is the solution you are looking for here, raster performance and 12gb vram make it a lot better value at 300$ than 4060.

The fact is the 4060 has lower bandwidth with a smaller bus and you have to compare it with a 2070 to make it favorable. It's a 1080p card at Best in today's games. In one year this card is not going to be a good card for 1080p. Frame generation is hardly in any titles, negatively impacts your input lag, and it's only really good for lowering persistence. That's fine and all, but we need to stop framing this as if it's a value. It really isn't. It is neck-and-neck with the 3060 t.i. in raster, but that card had more RAM and a larger data bus. The card got beat by an arc A750 doing RT in control. Can't call it a value if it can't even keep up with a competitors first generation card in one of the titles that Nvidia developed wth their own proprietary RT technology. Control was the first "good" RT game and 4060 loses to an A750? Yikes. How does it do in the path traced Minecraft or Quaker RTX? Those two essential RT titles are memory Hogs, so I highly doubt 4060 runs them that well. If you're going to say that it's got value for RT but you only show it losing, and winning in one Nvidia cherry-picked title, it shows it's not a value product at all, especially if future games are what you're looking for.

Still totally happy with my 3060Ti.

Thank you for consistently uploading videos at 1440 & 2160. So many channels don’t offer the options

What is that driving sequence benchmark in cyberpunk 2077? Is it a mod I can download from somewhere?

Maybe it's worth doing tests of this card on PCI-e 3.0? Or is it prohibited by the rules?

If Nvidia had called it the RTX 4050 and sold it for $249 people would be comparing it to the RTX 3050 instead of the 3060 and it would look a lot better in comparison.

Dumb question: what about power efficiency at fixed framerates? I'm running higher end hardware but I tend to lock my framerates and when the hardware isn't running at full capacity, are the results similar or different ?

👋 What version of Rt overdrive mod did you use? There are several options and the strongest is RT Overdrive Optimizations - Psycho, and in second place, as I understand it, is RT Overdrive Optimizations - Original? Apparently you used RT Overdrive Optimizations - Original?

You guys should've put 6700 XT in there. It's a monster in that price range

"The more you pay, the less we give you" -Nvidia.

The 4060 is an xx50 class card. The 107 die is xx50 class, the 128bit bus is xx50 class. Normally an xx60 can beat last gen's xx70 but this one can't even touch last gen's xx60ti. That's because it's actually an xx50 class and Nvidia are lying to you!

Definitely has the feel of a 4050. I wonder if they're setting up for the next generation where they'll return to the normal way of doing things, making the 5060 seem like a huge leap over this underwhelming 4060.

This gpu, and this generation in general is giving me RTX2000 vibes all around. There is little to no generational improvement in rasterization performance per dollar vs discount last gen cards, and the new features are more of a promise for the future rather than something that really improves my gaming experience today. There's no doubt that RT overdrive in cyberpunk is impressive, and the 4060 runs it decently, especially compared to AMD, but I've already played cyberpunk, and I'm not likely to put that many more hours into it other than a few for the DLC. Then I'm left wondering if this card will even be capable of running future path traced games with it's memory buffer already being right at the tipping point with modern releases, and the notable memory cost of both ray tracing, and frame generation. I just can't see ever choosing this over a much more powerful used card at the same price like a 3070, or a much cheaper card if you are just looking for something that can play modern games at playable frame rates at 1080p like a 6600.

@Digital Foundry - I would love to see a video where you just go through all the techniques for getting a stable FPS/frametimes at 30/60 FPS. It would be cool to see you go through a whole suite of newer games and just optimize them (in great detail) on a range of GPU's. Love your content!

12 GB VRAM at this price, minimum for me. It literally isn't expensive to buy the vram itself

No one mentions the power consumption. Rtx 4060 shines in lower tdp. We need more gpu with lower power requirement ❤️

Why is it TGP now and not TDP? What’s changed?

A750 goes hard on ray tracing. Good job intel. Can't wait for next gen.

The Intel A750 is looking pretty impressive, but I usually play older games, and my setup doesn't have resizeable bar...

Looking at GPU's for my Son's PC. going from a GTX 1660S. RTX 3060 or 4060? Not sure which to go for, is the 8gb of 4060 going to limit the Card compared to the 12gb of 3060?

I think (similar to how you display raw resolutions sometimes) you need to start displaying which DLSS 2 preset you are using when providing footage running DLSS

So, good card for 1080p RT Cyberpunk. And very underwhelming everywhere else, even in some RT titles. Not just getting beaten by the 7600 but also in many cases by the older 6600XT/6650XT which are available new for $70-$80 less. And apart from Cyberpunk which is more or less Nvidia's tech demo even that advantage may not be worthwhile going forwards, other titles with RT/Unreal 5 etc seem to be leaning towards lighting solutions that run well on AMD gpus. Which makes a ton of sense since consoles are the biggest market and both Playstation + Xbox are AMD hardware.

...I would like to see latency numbers with all implementations of dlss3.

Does the proformance on each card depend on the engine the developers are using, and open world or liner type of game?

8:11 That Black Ops 1 reference man. 🤣🤣🤣

DLSS is overhyped and comes with serious latency drawbacks. If enough VRAM is present then Raytracing is fully achievable. The only option is AMD, both for lower end with the prior 6000 series and the higher end with the 7900xt. You get 20GB of VRAM for less than $800

I have to give kudos to whoever created that RT OD mod.

is the whole vram issue just a case of developers not implementing textures properly on pc or are 8gb cards actually in trouble with upcoming games at 1440p? I've heard diablo 4 runs pretty well on the pc but the vram requirements are insane whats the story there?

Should I get MSI RtX 3060 ventus x2 oc 12gb? Will it be good buy?

2 main changes from Turing to Ada that are known to public smell like die hard candidates for very high variability of the improvement: 1. Turing to Ampere, adding a second FP32 unit to the same SIMD / SM block, sharing the instruction source operand gathering and issue port sub-blocks with the integer unit, all relying on some sort of dual issue - either same warp neighboring instruction dual issue, or cross-warp - same SIMD / SM dual issue (more likely the latter, as the former is notoriously ineffective, see RDNA 3's per CU per MHz results). The math throughput increase that that results in is widely circumstantial and naturally varies from shader to shader (and as a result, application to application) as it's dependent on the quality and distribution of instruction mixes of different types, latency hiding, and exact particularities of HW warp scheduling to yield results. 2. Ampere to Ada, dramatically reduced "DRAM bandwidth" / "math throughput" ratio across the board (in case of 4090 by increasing math throughput relative to 3090, but keeping the DRAM bandwidth the same, in case of lower mid-range cards by cutting down DRAM bandwidth, and doing both in-between) is attempted to be compensated by dramatically increasing L2 size. Caches _especially_ in GPUs are only as effective as spatially and temporally local the access pattern is (by address), that naturally drastically varies from application to application, and also tends to exhibit cliffy behavior as the same application / algorithm is scaled by some parameter (be that the frame buffer resolution, or texture resolution, or BVH size / LOD, or anything else): the moment the "hot" portion of the data structure parameterized and scaled stops physically fitting into the size of the cache it's efficacy at filtering (reducing) the traffic and thus memory bandwidth requirements can drop dramatically up to and including 0.

what about the added latency/lag when using frame gen ?

Why do the tree (and most other) shadows look so much better on the Overdrive mod vs stock?

Someone help me out here: when we see comparisons like these, where multiple GPU’s are pitted against one another, what footage are we actually watching in the background? Is it the card at the top of the list that we’re seeing the gameplay from? In the top left corner it lists details like the game title, resolution, and some other stats but it doesn’t say which card the gameplay footage is from. Or am I thinking about it all wrong here?

The graph for FH5 is incorrect, showing 172.9% power differential (for 76000, should be around 50% ish I Think

What is the problem of NVIDIA introducing a new generation with less RAM than an old generation 🤦

Funny how the RTX 60 tiers kind of almost have the same numbers especially when RT is off. Tis really makes one feel like its not worth it upgrading from a 60 tier card to another 60 tier card except for the availability of the new features like DLSS 3 which is quiet the feature. I currently use an RTX2060, and it really holds up pretty well in modern titles

I like how many people show cyberpunk as an example. In my case all upscaling features are broken once you open map or inventory. It also stops working for about 5 seconds during certain cutscenes. I have a 3060 laptop.

I am a very casual PC gamer who upgrades very infrequently, but I am wondering why anyone would pick the 4060 over a 3060ti which seems to be also $300 regularly on sale? Mostly just dlss3?

Pretty nice review thank you. Comparing performance to the rtx 2060 and ps5 were great addition to it.

Thanks for the last few videos focusing on this budget or best bang for the buck. I have a buddy I want o build a PC for; his old machine is easily 8 years old running on a GeForce 980. Still get the job done but I want him to have a better experience on some the more current titles. Seems like I could put something together for around 1k and that will be so much better than what he's got.

can anyone explain why even at 60fps on youtube, and >60fps in game, the video game footage still visually stutters about once a second?

didn't really understand why you talked about the 2070 super at the start and then completely forgot it in benchmarks lol, it could've been useful to compare raster performance

How professional rich videos are since his early days. Hes as good as it gets

8:10 Didn't think anyone would catch that Black Ops reference, did ya?

I can see texture difference in those Diablo screenshots between low, med, and high, but not between high and ultra even when I zoom in a lot. When 8gb isn't enough for high, that's when I'd be worried

If this was 75W and requires no 8-pin power, and was a 4050Ti it would be killer in a compact sff build.

3060ti and 6700 XT cost only 10€ more than 4060 in EU.

I'm new to PC gaming, but how is the 4060 performing that well on cyberpunk overdrive? I'm using an 7800 X3D with my RTX 4080 and Cyberpunk Overdrive is like 60-80 frames the whole time. I expected to hit 120 tbh

I'm looking for a reasonably priced option for gpu compute work (so CUDA is a must), some gaming and a TDP under 150W for a 1060 6GB upgrade --- Do I really have any options other than the (fairly mediocre) 4060?

Desperate attempt to make the card look like it has value. No 8GB GPU in 2023 has any value if priced over $200. Period.

was a good time when GTX 460 - a x60 tier GPU was just 50% percent slower than x80. and now there is 2x difference at least

The Arc A770 isn't discontinued, only the Intel-branded Limited Edition model was discontinued (hence, the "Limited Edition" label). Partner cards are still being made. Seems like it's every other week since launch that someone proclaims Arc GPUs to be dead.

How about some comparisons with the Radeon RX 6700 non-XT 10GB? It is 36 CU on RDNA2, 10GB frame buffer, just like the PS5.

60 class in comparison to prior generations this is how much you're getting screwed. GTX 960 $199 almost as fast as GTX770. GTX 1060 $260 as fast as GTX 980 and more ram. RTX 2060 $300 as fast as GTX 1080. RTX 3060 $329 as fast as RTX 2070 but more ram. RTX 4060 $300 as fast as RTX 3060, and less ram. It should be a $200 4050 it's that simple.

How are they selling the "next-gen" gpu with less Memory Bandwidth and Interface and CUDA Cores?

Is there a way to pull up that 8gb memory buffer?

I am interested in the 4060 with frame gen + dlss vs 3060 ti with dlss:) For a value evaluation...

AMD unable to deliver a similar card with 12GB? The 6700 XT is currently hovering around 300 would have been a better comparison. the 6700 XT is the low end king right now

You know it's bad when even John Lineman is struggling to lick boot

Those smaller factors for the 60 series are fantastic. I'd love to see a 70 card pull it off later in life. I've got a very similar 3060ti inside a m-atx case and for a thing that's like half the size of a ps5? It's so cool having a little pc box when you get over the rgb/wc build on a tower. Could put it in my backpack and go all around with it! Will be skipping out this 40 series generation as the gains minus dlss 3 isn't worth it for me but i'm looking forward to the 50 series, I think/hope some lessons will be learned after the whole 4070ti fiasco

I want to buy new card but I am getting confused about the ram problem. Is it future proof to get RTX 3060 with 12gb ram or just get the RTX 4060 with 8gb ram?

Why are we calling Frame Generated frames real frames when they are not? That input lag must be awful at apparent 60fps when in reality you playing at 40fps.....

Why is the Series X always excluded from PC comparisions is the Series X not relevent in gaming?

Why do you say that the PS5 is equivalent to the RTX 2070S? In many other videos you clarify that the closest thing to the PS5 is the RX6700 (non XT), with slight differences but being of the same architecture from AMD. Cheers!

Thanks for the video, it reminded me once more had bad any upscaler is at 1080p, even DLSS on Quality looks like straight vaseline, gosh.

4:20 but its not priced as a budget nor a entry level Gpu, its a $300 Gpu which has always been a price of a midrange GPU. budget starts from $200, Entry level start from $100-130

I will say it again. DLSS3 is garbage. It doesn't contribute to latency reduction that you usually get from real high framerates. And latency matters even in single player games. Just moving the camera with mouse feels less floaty and as a result more comfortable with lower latency. Just because single player games can be manageable with higher latency, doesn't mean it justifies garbage tier performance from a GPU that's priced that high. You can play single player games at 20 fps too if you really want.

For someone whos looking to upgrade from a 1660 super i want to know if the 4060 is a wise upgrade cause the prices for me here where im from is looking good

if the 60 series gen over gen upliftmnet is barely 10%, then the consoles will go on to live for more than 10 years this gen.

This is a 4050 non-Ti and should cost $220. An OC'ed version of it that almost matches the 3060Ti could sell for $250 and be called an RTX4050Ti.

Guess we’ll get the GeForce RTX 4060 vs 3060 vs Xbox Series X video tomorrow?

Uhh… was that a Call of Duty Black Ops reference I just heard? 8:10

Your chart for Forza 5 at 6:15 is clearly wrong, both the 4060 and 7600 numbers don't add up

4060ti 16Gb version @ 300-350 would be a fair buy to me. One day, in a lot if days, i may get that.

The problem with DLSS 3 frame generation is it doesn’t work correctly with VSync. They need to fix this for DLSS 3 frame generation to really matter.

So is the Rtx 4060 ti or ps5 better?

I’ll never understand the argument for ray tracing for such slow cards. Why force yourself to play with such diminished fps, settings, and resolution for a feature that barely makes any difference? Sure, with a 4090 go for it. Anything less is just wasted performance and/or visuals.

Digital Foundry: will use mods to boost DLSS2 quality and RT perfomance to paint Nvidia GPUs in a better light. Also Digital Foundry: never using FSR2 mods or RT mods to boost AMD GPUs when doing a review. Or comparing Nvidia GPUs vs AMD GPUs with real life prices, only comparing them based on magical, irrelevant MSRP. RTX 4060's competitor isn't RX 7600. It's RX 6700 XT. In real life prices.

I think DF leans much too heavily on Frame Generation. It really feels like shilling when they treat FG as real FPS.

Worst part is that frame gen requires at least additional 1GB vram

So 7600 for 269 or a 4060 for 329 or a 6700xt for 339 which is the best price to performance. prices in euro. Germany to Ireland.

While 1080p is still by far the most popular resolution in the Steam Hardware Survey, is it really sensible to primarily test a new GPU at 1080p? This is not Avery cheap GPU. People upgrading their systems might actually want to play at, say, 1440p. That resolution was up nearly half a percent in May and 1080ps share is shrinking.

"One cannot use DLSS 3 on RTX 30-series graphics cards as they do not include an Optical Flow Accelerator". In other words, the 4060 has something that puts the previous generation in a corner. Sure, everyone is on the frAMe generator hate train. " Fake frame", or " input lag", "not pure rasterization". Etc. Well my friends, few years from now, you will need a cooler the size of your bedroom, if you think rasterization performance should double in each generation. We are reaching the limit on what is viable. The 4060 consumes very low, performs better then the 3060, and doubles your frames with Deep learning AI. TO me sounds like pure jaw dropping technology. And the 4060 has it. For 299. If greater input lag is all everyone has to complain about, Nvidea reflex proves that its possible to solve this. AI will improve, and they will solve it, no doubt :) Got the 4060, loving it.. Less Vram, less cuda cores, less Memory interface. Sure. But also, better architecture, next gen processor and cuda cores, next gen everything basically, and performing better with "less". I would call that a win :)

If I would upgrade from a RTX 3060 ti, what GPU would be a worth enough upgrade, so it would be a noticeable change? Right now Im playing with an i5-12400f and RTX 3060 ti, but looking to upgrade in the near future.

The slippery thing is that it can be argued that nVidia is lowering their prices of some of their cheaper cards each generation, because the new cards of the same tier gets cheaper, but performance and spec wise they are moving which card falls in which tier. They make the 50 series of the previous gen the 60 series of the next, but then dangle a "Ti" version for a $100+ more. There were only Ti versions of 70 and 80 tier cards before, now all tiers have them to encourage upselling. On the high end, in the 20 series they doubled the price of the Titan cards from $1000 (GTX Titan) to $2000 (RTX Titan). In the 30 series they rebranded the Titan series to the 90 Ti, and added the 90 tier in between that and the 80/80 Ti in the 20 series. For the 40 series they completely bypassed making an actual 80 or 80 Ti and are slapping the 80 tier name on the 90 tier card, and the 90 tier name on to the 90 Ti card, but keeping the price of the higher tier card. I'm sure their thinking was that with such a huge performance gap between the 4080 and 4070, why would someone with a 2080 or 3080 go down a tier when we can get them to pay for more to stay in their current tier. I wouldn't be surprised by a $1800 4080 Ti and possibly a $2200-2400 4090 Ti in the next year.

From all the reviews I've seen of this card, several RX 6700(XT) and 3060Ti owners have probably passed out from laughter.