Watch the RTX 4070 review & benchmarks here: ruclips.net/video/uZHDq-LEGzw/видео.html And check out our RT Overdrive (path traced) benchmarks here: ruclips.net/video/0EYaMupOPJg/видео.html Grab the GN Project & Solder Mat on the store! store.gamersnexus.net/products/gn-project-soldering-mat Or the GN Large 'Volt' Modmat! store.gamersnexus.net/products/modmat-volt-large
The reason those memory slots are there is because there are 16gb versions of the 4070 coming out. There was leak showing Gigabyte listing a 16gb model to this effect. Namely the GV-N4070GAMING OC-16GD
Terrible phrasing here 2:39 "High powered card", You mean high power draw or processing power? Seems like you need to use the latter phrasing from now on.
Hey @Gamers Nexus. Is it considered fraudulent if the gpu is not labeled? E.g. AD104 on the chip or something? If so, then is it possible that NGreedia has some liability if it shows as it is here in the final products in the market? Then perhaps they can be called out on false advertising or labeling?
If they didn't burn money on the custom PCB, overly engineered heatsink, custom 12V power connector, custom power monitoring, etc. they might actually make a decent profit on these. That's what your paying for - NVIDIA's proprietary bullshit!
@@GamersNexus I actually would split the difference there about $150 less than what it is now is what I would have considered a fair price. And even then that's on the higher end of my range. Don't get me wrong, that was seriously looking at a 4090, in part because of my work, but it was also because the performance for what I consider to Halo product that was good for my work was something I was willing to pay significant premium. My range went from 1400 to 1600... For the 4090 not some cut down overpriced 4080
With that "tiny" PCB you could certainly mod a SFF card. You might want to undervolt as well to control just how much heat it puts out depending on the precise case, airflow etc.
The "552" on the ASIC package is a qualification line bin and tray number. It's just for tracking which chip it is from that batch of qualification chips. Generally they're pulled from mass production and sent through a different set of testing than the normal automated production testing line, so when they're pulled they get a quick sharpie mark to identify them. You see these a lot on pre-qual and qual cards, some of which end up being vendor or partner review and test samples or loaners. Like yours!
3:22 - The cards are not merely assembled... they're _forged_ in the fires of Mount Doom itself, with the power to rule over all AIB partners' competing cards.
I have mixed feelings with this card, but right now I'm glad we have a small card with lower power consumption for a change, our small form factor PCs are going to be happy with it.
As someone who bought a 3070ti a year ago for $700, I'm happy there's an option that's $100 cheaper and 25-35% faster in most games. Not a giant step in the right direction, but still slowly getting better. Also significantly better for Cyberpunk Path tracing, goes from barely playable on the 3070ti to a good experience with the 4070.
We all know that this is the 4060 put inside the 4070 box with the 4070ti price on it. 1070 was 29% faster than the 980, 2070 was 16% faster than the 1080, 3070 was 26% faster than the 2080, and this fake 4070 is 3-5% SLOWER than the 3080.
Calculations aren't that easy though. 980 was a scam card, it was like 5% faster than the 970 on a good day. 2070 being faster than the 1080 tracks, but nobody bought that card, it wasn't "fast enough". 3000 series was a HUGE leap in performance and it's to be expected for this to not match it. What's wrong here is that the pricing should reflect this. The 3080 was like $700 at launch, this should be at least 150$ cheaper. At least. Instead it's a same price proposition of an upper level card from the past generation. Wtf.
@@Galf506 The 980 was between 15-25% faster than the 970 depending on the game. If your 980 was only 5% faster "on a good day" then you good sir should have sent it back under warranty lmao that shit was broken.
Ok. 970 was way faster than the 780. 1070 was on par with the 980ti. 2070 got a massive price bump but was still on par with the 1080ti with the super variant. 3070 on par with the 2080ti. 4070 gets massive price bump and is 15% slower than the 3080ti
1070 was 29% faster than the 980 \ 2070 was 16% faster than the 1080 \ 3070 was 26% faster than the 2080 and this fake 4070 is 3-5% SLOWER than the 3080! 1060 was 5% slower than the 980 \ 2060 was on pair with 1080 \ 3060 was 10% slower than the 2080 and the fake 4070 is 3-5% slower than the 3080! 1060 has 33% of full die cores \ 2060 has 42% of cores \ 3060 has 33% of cores, and this fake 4070 has 32% of full die cores. Not to mention that this thing is only 30% faster than the 3070 while the uplift from Ampere to Ada is more than 70% (3090 to 4090). This is $400 XX60 tier GPU and it sells as XX80 tier GPU!
I mean its still kinda a 3080, just a realistic 4060 Ti equates to a 3080. Which is impressive but also a shame. Nvidias showing why theyre the best and the worst at the same exact time. Technological theyre on another level, greed wise theyre also on another level. This gen couldve been a crazy leap forward in performance but they didnt have to compete so instead they said “buy the 4090 or accept paying a lot for a little” unfortunately considering the performance and pricing of the AMD’s 90 classes, i dont see it getting remotely better
Yeah looking at that PCB; that looks like something you would find on a $250 card; 3-4 phase VRM? lol. While the performance was disappointing for the time, the Vega 64 reference had a PCB that was rolls-royce component wise compared to this, for $100 less [$500usd msrp] at the time... and it was technically still the 2nd fastest card on the market ignoring titan variants and early drivers. Yes by early 2018 it had thoroughly thumped the 1080 in +80% of games. And by the time Turning came out it had gapped it in 95% of games.
@@anasevi9456 4070 ti(previous 4080 12gb) is physically a 4060ti die. The 4070 is physically a 4060 die. You can easily calculate that by doing maths on the cuda cores and all previous nvidia cards and segmentations. It's sadness.
Of course it would, I wouldn't be surprised if they were selling the GPUs to themselves and writing it off as an expense for tax reduction purposes to make more profit.
@@earthtaurus5515 that would be what is known as an inter company transaction, which has no effect whatsoever on their financials or tax return as it is removed via an eliminating entry.
@@shadouqh8370 Yes, that would be the case if it's within the same company. Corporations like Google for example move money around from subsidary to subsidary in various regions in order to pay less tax. I imagine Nvidia would have a similiar structure.
The smaller the PCB the smaller the cost to manufacture. I do like that this card is very efficient and small, just wish it was £480 instead of £580. At sub £500 it would be genuinely enticing.
Keep in mind all third party units will be even more expensive than the FE model as well. People are already seeing them going for $1000.. for a 60 class card improperly labeled a 70 class..
@@PoppinboyJojy as borderline unfair as it sounds it's almost a advert for founders edition cards. afaik the board partners were targeting a much higher MSRP then what it got in the end so most of those cards are likely going to cost way more then what nvidia claims it should cost. but usually (ateast for my 2070S) I got it at the price nvidia says I should from nvidia ( as any AIB cards was 50-100 euro's more and I just need a reference PCB to slap a reference waterblock on...can't get more reference then nvidia's own cards...atleast you'd think...not sure that's the case with the 3000/4000 series but haven't shopped for waterblocks for those)
Its so obvious, there's 2 memory empty slots because the board was designed for 16GB VRAM but since they want to cut cost even more for higher profit they remove 2 memory chips out of 8 and it ended up with only 12GB VRAM (6 memory chips)
You re wrong, they re just reusing the board layout between 4080 and 4070. Fully enabled AD104 has 6 memory controllers each being 32bit, or in other words 192bit bus. With 6 memory banks it can have either 6, 12 or 24 GB of VRAM, 16 GB is impossible (with current 8/16Gbit chips) unless you want to have asymmetrical memory layout like GTX 550ti had
With way more and better features and driver support for way way longer. Nvida is pushing it back because of their popularity but the 10 series cards days of driver support are numberd as they start to push 7+ years old
@@mercuryrising9758 don’t believe the bullshit. The 10 series is good for a while yet and nobody is going to try and run super-tech games on 7+ year old hardware. NVIDIA are a joke.
@@mercuryrising9758 But hes right through. Maybe 1060 is bit to low these days, but Im sure most games outside of Read dead 2 / god of war / cyberpunk can be played on high/medium. Back then it was 6th gen intel so if ppl pair it with such a old system it might be hard. Personally my desktop with 4th gen / 970 is no way near enough these days, but 10-series, ddr4 ram etc really helps.
@@alarak2159 ima be honest, the 10 series good if you dont want high-res or high refresh. I had a 1080 and upgraded to a 3070 ti and its a huge difference for high refresh.
The placeholder pins for VRAM seems to indicate that there could be a 16 GB card. der8auer did also a teardown with the FE and 2 different boardpartner cards and they too have these placeholders.
Nvidia was probably playing it safe with the design. If the public would rage hard enough about the card having less memory than AMD, or if Intel just happened to be able to pull a rabbit out of a mouse and got their card to perform, they could easily change the spec, have two more chips slapped on there and update the firmware. But they wouldn't do that unless forced to as it undercut their product segmentation strategy and it would cut into the profits as they know that most people are still very price sensitive and would not react well to a huge price hike. That's who the Ti version are to snare. And here it gets interesting. Both the 4070 and the 4070Ti share a lot of similarities. Actually they are basically the same card. They have the same GPU and memory. Only things that differ are the power delivery where the Ti version can handle 85 watt more, oh and the fact that the non Ti version has the AD104 GPU crippled by disabling 20% of the cores. Other than those disabled cores, slightly better power delivery and probably a bit better cooling these are the same cards. The difference in cost for these are pretty darned small, definitely it's less than the difference in price to you and me...
I really wouldn't be surprised if Nvidia released the 16GB version later on, especially depending on what AMD's RX 77xx cards will bring in terms of VRAM...
Really enjoy these tear downs Steve. Even with the trouble you had with dissembling the 3080 without any instructions, I'm glad I have that video as a reference if I ever need to maintain the card.
In NZ the RTX 4070 cards are around $1300 NZD ($800 USD). The formerly redundant RX 7900XT is now reduced to $1500 NZD and sometimes even cheaper when sales are on ($940 USD). I am thinking paying a little more to get the AMD card might be better value. It has 8GBs more VRAM, but I am not sure about FPS yet. The RX 6750 is now discounted to as little as $400 USD here, so that's an upgrade option for me too (my CPU is a R9 5900X, but I just bought a 4k 144hz monitor, so I need to upgrade the GPU). Yeah, I have an AMD bias, because Intel and Nvidia have pissed me off too many times over the past two decades or so. But AMD has to earn my support. If they are in the same ballpark for price/performance, they get chosen (unlike Bulldozer era when I had no choice but go with Intel and the 'a quad-core is all you will ever need' mantra). If AMD manage to be competitive and we buy their stuff, it will prevent a Blue Green monopoly and therefore monopoly prices (duopoly prices are not much better, but are a wee bit -- as Ryzen forcing Intel to lower prices and have more cores proves). P.S. The RTX 4090 is a little cheaper now here, but still up around the $4000 NZD mark. I could get a half decent used car for that, or a very decent electric guitar. After decades of buying high end computer parts and seeing them depreciate so quickly, I decided it was like a black hole to forever throw money into. I just buy what I need to do the job now -- rather than trying to keep my PC on the cutting edge like I used to (I was a chronic early adopter and always wanted the best of everything). Yeah, I am old-ish. OG gamer. Space Invaders and Pacman generation. Then it was Commodore 64 and ZX Zinclair etc. Things have come a long way since I was inputting BBC Basic code on a RISC based Acorn computer at school in 1983 (Acorn was part of what has become ARM).
Here in Mexico its the other way around, old cards are super price inflated while we get the newer cards at MSRP thanks to amazon USA shipping to Mexico
Tip: attach a string to the old fan cable before pulling it and use the string to pull through the new power fan cable like you would replacing fiber runs.
It reduces assembly cost, but the origin of it is that Steve Jobs thought screws were ugly and mandated they be eliminated. Then they found out this could also sabotage repair so they were all in
Cynical me wants to think there will be 16gb vram Super variants of the 4070/ti next year, with an updated bus. Altho that part might be more difficult than what Jensen deems worthwhile - can't skimp out on margins for those leather jackets he's collecting
Isn't Notebooksbilliger the only one selling FE cards? I've got worse news: They're also responsible for the Dutch market (even though their site is completely in German). So more people are competing for them. I decided to get the Asus 4090 TUF OC instead of the FE once the price was the same. Because those FE ones only get in stock every few months or so.
For my SFF case, this is the first latest gen card that would actually fit. 200W is also great! 12gb is a little worrying, won't age as well as 16gb... but hopefully game studios also learn something from the recent 8gb debacle.
@@Micromation 12 GB is Not even enough now anymore in some Recent titles in 4k will be Not enough in 2 years for wqhd IT will be the same as with the 8 gbs now
@@benito9017 Idk, I perceive 4k rather questionable under 30'', unless we're speaking ultrawide form factor where all these additional pixels actually serve a purpose other than tanking the performance. Besides buying mid range card for resolution that chews up top dogs, is like buying 3060 on a promise to play with RT on at relevant framerates... Personally I won't be upgrading to 3x1440p setup (2mil more pixels than 4k) until 5090 comes out and shows some massive uplifts over 4090 because performance just isn't there...
nvidia is stretching the price well at rtx 4K. Greed has no limits? 830€ for the cheapest asus rtx4070 or 780€ for an msi trioX. In short, the same price I paid for the 3080 trioX when they came out. Brutal profit margin for a 4060 with the 4070 tag on top.
that is true. The 4070 simply doesn't make any sense for normal form factor PC's, as you can get cheaper cards that perform a lot better. The only application I can really see the 4070 fitting is as a smaller compact card for people rich people that are making SFF builds. For any one else it does not make a lot of sense.
The power consumption is a win. The size of the card is a win and general performance is pretty good. The pricing just kills this card in the market. When you can pick up a 6700xt, 6800xt, or go used from both sides. You can get a better performing card for less money.
1070 was 29% faster than the 980 \ 2070 was 16% faster than the 1080 \ 3070 was 26% faster than the 2080 and this fake 4070 is 3-5% SLOWER than the 3080! 1060 was 5% slower than the 980 \ 2060 was on pair with 1080 \ 3060 was 10% slower than the 2080 and the fake 4070 is 3-5% slower than the 3080! 1060 has 33% of full die cores \ 2060 has 42% of cores \ 3060 has 33% of cores, and this fake 4070 has 32% of full die cores. Not to mention that this thing is only 30% faster than the 3070 while the uplift from Ampere to Ada is more than 70% (3090 to 4090). This is $400 XX60 tier GPU and it sells as XX80 tier GPU!
@Krystian Nowak I agree that the performance isn't where it should be. With how much they cut down the pcb and limited the power. The performance is generally good. It's not great but it's not bad. For someone that wants to upgrade to the current architecture it could be compelling.
Lots to criticize about modern nvidia cards, but if you have to do a "screwless" design, this is a nice way to do it. Pretty easy to disassemble, no silliness.
I'm betting we're gonna get another GPU between 4070 and 4070 ti, due to the price gap. It might be possible nvidia brings back the "super" variant of a 4070 with slightly more vram, at 16gb for $700. It'd have to be 256 bit so they might tweak a 4080 down to similar specs.
I have the 4070 Founders Edition and the design is excellent. This card Overclocks like a champ. A matter a fact out of the 12 different video cards I've had in the past 20 years this one overclocks the best.
Had this come out with 16gb of Vram, and priced at 475-525$ It would be super interesting But at 600$, with AIB cards only being more, and only having 12gb It’s a hard pass Seeing a fairly power efficient, high performance, normal form factor card Is awesome Would have given me hope for a really cool, 120-150 watt, 4060 that can fit in SFF cases while stock Unfortunately, that card will likely have 8gb of Vram, and cost over 400$ We need to hold on buying these things, and let prices drop to decent levels again, which enough Vram to play any game I want to
And at the end of year 2023 Nvidia still will have 80%+ market share… People just want to buy Nvidia. No matter how bad the product is/will be in price wise…
I enjoy seeing the good build quality even on this lower prices cards. At least they are not tricking the customers here by using just simple eye blending plastic. Thanks for the teardown! Always a pleasure.
That's an impressive GPU design IMO. The PCB is compact, there's no real BS going on to mount it to the cooler. My only pet peeve w/ it is that little decorative element covering the baseplate. I want a GPU that's just a black brick, no frills or anything like that. Just the PCB and the cooler. For what the 4070 is though and the design, they did a really good job on there. I'm surprised!
I disagree that Nvidia are going 'Apple' when it comes to screwless designs. The iPhone is pretty much the only phone left with exposed screws, two on the bottom. The Macbooks are super easy to get in to because the screws are exposed on the bottom. I mean, you can't do anything once you open it, but they're easy to open!
Hope you guys keep covering the 4070 when the manufacturers versions come out. I have read some of those are lower quality than the FE, due to Nvidia last minute MSRP prices
I have to thank you Steve. I was always slightly intimidated by the thought of tearing my GPU apart because they're so damn expensive. However my 3070 started running 20ﾟ hotter and 15% slower (18 months of 24/7 mining which I made enough to cover the entire cost of my pc) than I was used to so I had to tear it down. The thermal paste was dried out and cracking. After replacing it with fresh my 3070 is running like new again(honestly maybe even a little better then when it was new but i assume i have driver updates to thank for the 2%ish bumb in performance. Thanks Steve!
I feel like this card shoud have been 499$ , mostly because of the 12GB Vram Then the 4070ti 599$ with a 16gb Vram Right now it makes no sense to prefer this card over a much cheaper 6800XT or 6950XT that has same price
Nvidia is increasing their prices mainly because they are nearing 90% market share which basically tells them they can do whatever anti-consumer move to rise their margins to the moon. I still dont understand why AMD almost doesnt market their gpus except their own webpage, I think they would gain a lot of market share just by actually marketing their GPUs which have good price and not the rx7900s which are basically the same shit nvidia does in terms of pricing
@@damara2268 AMD is being equally as stupid as nVidia. nVidia can charge exorbitant prices simply due to the fact they're selling you a multi-purpose GPU. AMD on the other hand are selling one trick ponies. The 7000 series are worse at RT, worse at compute, worse at ML, and they have inferior upsampling tech along with 20-40% lower performance per watt. The 7000 series is really only highly competitive in E-sports gaming and a few AAA DX12/Vulkan rasterization games, yet AMD thinks they're going to sell the cards for only a 10-15% discount versus competitive nVidia GPU's. Both AMD and nVidia are delusional right now, and buying anything from them right now is stupid. For the idiots who forked out $1000+ for midrange cards during the pandemic, despite everyone telling you not to, these are the fruits of your stupidity and impatience. GG.
@@damara2268 imo AMDs market share is hard to take back because a lot of people think ai and some other software only run on Nvidia GPUs. Nvidia took over a lot of the market thanks to paying devs to prioritize cuda and their exclusive features for years.
@@xfy123 people don't only think that majority of ai and compute SW doesn't work or works badly on AMD GPUs, that's true. But AMD GPUs are much more with gaming in mind than Nvidias that are nowadays a lot multi-purpose. AMD should advertise it properly, they can easily take top selling charts in regular stores where gamers buy
That was fun and educational! Was nice to see a not pro tackle a pretty big job in obe shot! Was actually pretty amazed how much she learned from watching you. She retained more than I have lol
The 4070 looks like a great card, which is just $100 too expensive. 3080 performance, with the added benefit of DLSS3, but with a huge improvement in efficiency, meaning lower power consumption and cooling requirements (and thus size). For $500 this would be an excellent product
@@visceraeyes525 What are you smoking? The PS5 and Series X mimic the performance of a 6700xt, which is almost half the performance of the 4070. Also, the PS5 shares the vram and ram.. so 16gb of vram in the PS5 is misleading.
@@something3610True on the ps5, but 600 dollars for a card that'll have its vram start to get saturated in the near future isn't the best thing in the world, its compute power is restrained by the low memory bandwidth and capacity. To think Intel offers the most memory per dollar is insane.
even if it was actually possible do you really think they’d charge anything less than ~$700 for it? you guys will just complain about it being too expensive again anyways so who cares?
14:20 That PCB can accomodate the RTX 4080 I believe since 2 vacant memory pads are there. Hopefully with competition from AMD RX 7800XT series, Nvidia will launch a further cut down AD103 , RTX 4075 with 16GB of VRAM to compete.
It actually wasn't really standard for the 20 or 30 series. That's when a lot of cards began moving to 2.5 and 2.8 slots (like EVGA's). It was available, but not as common or prevalent as the 10-series.
I would LOVE to see true single-slot watercooled cards back in vogue! And the PCB is so dense that a card that was designed for watercooling from the ground up could be made EXTREMELY compact.
Imagine a world where Framework makes modular GPU PCB shroud add-ins for PC towers, and you could just but a new PCB instead of the whole thing so you save $75 bucks or so. Imagine.
@@masterdoge17 Yeah. I mean, even to go a bit broader, how we have a whole motherboard that can be swapped out of a pc case and a new one placed in that is better than the last (within size capacity of course). Just keep that thought going as far down as we can through-and-through PCs to save on waste overall in a way that is at the very least as easy to do as swapping a cpu and/or ram on a laptop.
At 17:50; if those extra 2 memory positions were populated, this would be a 16GB card. Maybe they dropped it to 12 late in the day to fit a lower price point.
The memory bus does not support extra memory chips. They could use higher density memory chips, but can not add more memory chips. What they could use, is to use same pcb with 4080 chip… and downlock a lot.
You can’t just populate additional memory pads. You need a memory controller to actually access it. AD104 has a 192 bit memory bus, meaning 6 32 bit memory controllers. That means that it can handle 6 memory modules (12 if they are back to back). What is likely is that this board may have originally been intended for a higher tier GPU, or the additional slots are there to improve yields on the board itself.
That die size though. Benefits of removing so much bus width (since it doesn’t shrink like other components) and a more advanced process node. Seems that will be nvidia’s strategy outside of the top end card.
Important to understand that the bus width is nothing more than a multiple of the number of physical RAM chips on the card. Each GDDR6X chip, regardless of capacity, has two 16-bit buses. That's why this card has a 192-bit bus (6 chips times 32 bits). The reason this 4070 is better than the 3070 and its 256-bit bus is because the memory speed is 50% faster. Think of it like traffic lanes. What road would get you to your destination faster? A 4 lane road where the speed is 80km/h, or a 3 lane road where the speed is 120km/h?
@@warren_r no I get that haha. It’s not directly comparable across generations of memory or even higher throughput memory of the same generation. It’s still definitely a cost cutting measure as IO doesn’t shrink like it once did moving to more dense nodes. I do think that 16GB of memory on a proportionally wider bus would give this card better performance over the long term though but that would add cost and cut into nvidia’s margins and/or raise prices for end users. The cost of adding the extra IO on silicon due to how much die space it takes up would be high and for not much more in transistor count. It’s proportionally more expensive to do on a denser node and this trend continues on in the future. This is one benefit of the chiplet stuff AMD is doing although im unsure how it pans out long term, seems there are some issues with this first gen. I know this has 50% (more memory bandwidth than last gen but it’s also more powerful. Maybe more memory and a wider bus wouldn’t make any difference on this or the Ti but even if it did, my hunch is nvidia is happy to give people enough performance at this tier to buy it but not to be a great value either
@@jolness1 Whether you realize it or not, you're asking why the 4070 doesn't have 16GB of VRAM. The answer is probably money -- those Micron 2GB/24gbps chips cost about 30 USD each.
That would be the only real way they could carve out a good Market for this card. As there are a lot of wealthy SFF builders that are willing to pay a premium for extra power in a small package.
@15:19 in like 10 years, I see the lack of GPU label on core as a potential problem with GPU scammers. Like now when you buy 60 buck RTX2080 on aliexpress or wish, you can at least disassemble it and tell immediately what it is, even if BIOS is flashed. After these cards age to the point of being basically e-waste, people will still be scamming others with them, but it will be very hard to determine what you get, especially if Nvidia sticks with this design for the next several generations.
I dont think these cards will be e waste for a long long time, because they will always be able to play older games very well. Theyd have to be to the point of non-functioning to be ewaste imo
Times changed, these Fermi era GPUs they use to scam people were from an era where new DX versions obsoleted a card within 3 years of it's release. I'd be worried if i knew DX13 was in the works, but knowing how games focus mainly on PS5 for their development that's not going to be a problem anytime soon even if that were the case.
As bad value as this card is compared to stuff like 6900Xt and such, it's actually quite impressive Nvidia managed to cram this performance into such physically small card, and not only manage the thermals, but also do it all at quite energy efficient levels. As greedy as they are, they do know their stuff when it comes to GPU design and manufacturing.
this card is going to be the hot ticket item for the 4000 series once it gets a price cut of at least $100 and the used ones start trickling through - a 3080 for such a low wattage and a small form factor is a killer proposition for all gamers that just want a very capable 1440p card or a bulletproof 1080p card. Sadly Nvidia just has to murder the prices on these after we've shown them that we'll buy 3080's for double MSRP. God damnit.
One thing to note, the engineers were the ones who designed this so props to them. The marketing/higher ups/financial what-not dept were the ones who decided on the price so they're the greedy ones.
You mean TSMC right? Because most of the improvements from Ampere to Ada on both power and efficiency are a direct result of going from Samsung 8nm to TSMC 5nm. This has more than double the transistor density of a 3090, it's not like they made some incredible design improvements this generation.
@@kajurn791 TSMC only applies to the core. The overall board design, power phasing, cooling, trace layout, all that stuff is still done by Nvidia engineers. Transistors, however small or power efficient they are, can't do anything without proper cooling and power delivery and other support systems.
Asus are likely to be one of the companies to make an ITX scale 4070 GPU as an updated Dual RTX Mini, the last ones they built were the Dual RTX 2060/2070 Mini; the later being a very competent GPU.
This somehow looks like that this are the 4080 12gb leftovers that they couldn't sell anymore after the shitstorm around the 16/12Gb vram versions, and perhaps they just replaced the resistor to make it less powerful (what would be a ridiculous move) and sell it as 4070.
This "cleanness" on the PCB separation just proved my theory that the 4090 was ready and stored in some secret warehouse for a lot longer than we might think otherwise. This 4070 is so fresh, that the thermal paste was still fluid.
kinda off topic here but, I am a trucker and I use a Bluetti ac200max power station to run a Lenovo legion 15ach6h laptop (3060) on my off duty time, and while that is sufficient for a 10hr break it falls short on a 34hr and I end up needing to idle for a couple of hours to recharge every 10-12hr (truck has a 20a inverter and have dual AC chargers for the power station , but I prefer not to, always want to try for that MPG bonus). I do not have many new games as I mostly play Mudrunner/Snowrunner, or the Borderlands series era stuff, titles that all run at well over 120fps on this system maxed out but, pulling 300w load. I have began limiting FPS to 75/80 on the shooters and as low as 60 on the driving games, but I hit a point of diminishing returns where the system stops reducing its power draw. it never seems to drop much below 175w if the 3060 is active on any sort of 3d load no matter how simple. I have noticed a few articles hint that the 40 series is supposed to be quite a bit more power efficient and I wondered if that applied to possible gains in low load efficiency as well. I have had people tell me to just get an overall less powerful system but I would like to maintain as high a level as possible as I occasionally Use Solidworks and who knows there might be a new game someday that I actually want to buy. I would love to build a SFF system as I am space limited (Vovlo vnl 760) but I am very concerned about shock/Vibration causing damage to expensive components), bonus kudos given if anyone knows of a case designed to keep components safe in such an environment?
@@kizuati I just love triggering AMD fanbois since they rarely defend their point and just go "Fuck you how dare you fanboy" or "I got nvidia stuff so I cant be a fanboy" I havent had a single bit of Nvidia hardware outside of my current 1070ti. Currently only got AMD stuff except for a single NUC.
@@AlpineTheHusky I don't even like AMD you brain damaged mong as they're doing the bare minimum with their cards right now and only look good next to NVIDIA. Good to know you're your mummy's good boy who argues to strangers he's the smartest guy in the room in regards to GPUs.
Watch the RTX 4070 review & benchmarks here: ruclips.net/video/uZHDq-LEGzw/видео.html
And check out our RT Overdrive (path traced) benchmarks here: ruclips.net/video/0EYaMupOPJg/видео.html
Grab the GN Project & Solder Mat on the store! store.gamersnexus.net/products/gn-project-soldering-mat
Or the GN Large 'Volt' Modmat! store.gamersnexus.net/products/modmat-volt-large
The reason those memory slots are there is because there are 16gb versions of the 4070 coming out. There was leak showing Gigabyte listing a 16gb model to this effect. Namely the GV-N4070GAMING OC-16GD
Terrible phrasing here 2:39 "High powered card", You mean high power draw or processing power? Seems like you need to use the latter phrasing from now on.
Hey @Gamers Nexus. Is it considered fraudulent if the gpu is not labeled? E.g. AD104 on the chip or something? If so, then is it possible that NGreedia has some liability if it shows as it is here in the final products in the market? Then perhaps they can be called out on false advertising or labeling?
552 means custom built for your review ;)
Is it just me ?
I mis the blower models .
Always work and the heat out of my case
It's not the size of the card that matters; it's how much Nvidia can profit from it.
😂
If they didn't burn money on the custom PCB, overly engineered heatsink, custom 12V power connector, custom power monitoring, etc. they might actually make a decent profit on these. That's what your paying for - NVIDIA's proprietary bullshit!
the working poor pay, while the idle rich play
look at the few vcore phases!! even the boardpartners use more!
100$ worth of graphics card max in this Btch
1:07 "Please tear off before install lation"
Never forget the install lation.
hahaha, didn't notice that before!
Wrong timestamp?
Lmao I absolutely love that you picked up on that😂
They probably wrote it like that to ensure people would remove it due to how upsetting the misspelling is xD
It puts the lation on its skin! It does what it's told or it gets the install again!
2:14 modding it for a super small card for SFF build is exactly why I'm interested in it. I wish it was $200 or $100 cheaper, though.
Agreed. $100 cheaper would be killer territory for this card.
@@GamersNexus I actually would split the difference there about $150 less than what it is now is what I would have considered a fair price. And even then that's on the higher end of my range. Don't get me wrong, that was seriously looking at a 4090, in part because of my work, but it was also because the performance for what I consider to Halo product that was good for my work was something I was willing to pay significant premium. My range went from 1400 to 1600... For the 4090 not some cut down overpriced 4080
Ok hear me out, graphics cards that are short but thick with really tall really thick heat sinks
With that "tiny" PCB you could certainly mod a SFF card. You might want to undervolt as well to control just how much heat it puts out depending on the precise case, airflow etc.
It will drop in price very quickly because I've seen around the 6950xt for around the same price and that's a better card.
The "552" on the ASIC package is a qualification line bin and tray number. It's just for tracking which chip it is from that batch of qualification chips. Generally they're pulled from mass production and sent through a different set of testing than the normal automated production testing line, so when they're pulled they get a quick sharpie mark to identify them. You see these a lot on pre-qual and qual cards, some of which end up being vendor or partner review and test samples or loaners. Like yours!
3:22 - The cards are not merely assembled... they're _forged_ in the fires of Mount Doom itself, with the power to rule over all AIB partners' competing cards.
Praise the sun !!!
Well said !
I have mixed feelings with this card, but right now I'm glad we have a small card with lower power consumption for a change, our small form factor PCs are going to be happy with it.
SFF gang rise up 😤
But seriously, it's been a while since I've felt comfortable getting a modern gpu to air cool in my existing SFF system.
@@jablue4329 sff deserves it, full towers are superior in every way.
@@gregdaweson4657only if you have the space or don't care for the look of SFF. not everyone wants a giant pc or can fit a giant pc into their setup
@@gregdaweson4657 you're so wrong you don't even know it
As someone who bought a 3070ti a year ago for $700, I'm happy there's an option that's $100 cheaper and 25-35% faster in most games. Not a giant step in the right direction, but still slowly getting better. Also significantly better for Cyberpunk Path tracing, goes from barely playable on the 3070ti to a good experience with the 4070.
We all know that this is the 4060 put inside the 4070 box with the 4070ti price on it.
1070 was 29% faster than the 980, 2070 was 16% faster than the 1080, 3070 was 26% faster than the 2080, and this fake 4070 is 3-5% SLOWER than the 3080.
Makes sense as to why they dont have an identifying stamp on the silicon
Calculations aren't that easy though.
980 was a scam card, it was like 5% faster than the 970 on a good day.
2070 being faster than the 1080 tracks, but nobody bought that card, it wasn't "fast enough".
3000 series was a HUGE leap in performance and it's to be expected for this to not match it. What's wrong here is that the pricing should reflect this. The 3080 was like $700 at launch, this should be at least 150$ cheaper. At least. Instead it's a same price proposition of an upper level card from the past generation. Wtf.
I'm not even sure it's the 4060. It might be a 4050, possibly a TI.
@@Galf506 The 980 was between 15-25% faster than the 970 depending on the game. If your 980 was only 5% faster "on a good day" then you good sir should have sent it back under warranty lmao that shit was broken.
Ok. 970 was way faster than the 780. 1070 was on par with the 980ti. 2070 got a massive price bump but was still on par with the 1080ti with the super variant. 3070 on par with the 2080ti. 4070 gets massive price bump and is 15% slower than the 3080ti
"Forcing you to pay more for less" should be the Nvidia mission statement.
and AMD's should be doing nothing and blaming Nvidia
@@aomine6817 AMD's
That's literally the motto of Capitalism in general.
The more you buy, the more you save!
😂
Nobody is forcing you to buy Nvidia GPU’s.
NVIDIA: Moore's Law is dead.
Also NVIDIA: Look at this tiny ass chip!
Can’t wait for them to continue saying MLID when the 6050 comes out with a 10mm^2 die.
More Proof that Nvidia will just lie whenever they want
1070 was 29% faster than the 980 \ 2070 was 16% faster than the 1080 \ 3070 was 26% faster than the 2080 and this fake 4070 is 3-5% SLOWER than the 3080!
1060 was 5% slower than the 980 \ 2060 was on pair with 1080 \ 3060 was 10% slower than the 2080 and the fake 4070 is 3-5% slower than the 3080!
1060 has 33% of full die cores \ 2060 has 42% of cores \ 3060 has 33% of cores, and this fake 4070 has 32% of full die cores.
Not to mention that this thing is only 30% faster than the 3070 while the uplift from Ampere to Ada is more than 70% (3090 to 4090).
This is $400 XX60 tier GPU and it sells as XX80 tier GPU!
We thought this was a 3080 Remastered, but seeing inside, it's actually a 4060
4060 is gonna be repackaged and reflashed 3060Ti 🤡
It certainly looks like a xx60 card!
"4080 S"
I mean its still kinda a 3080, just a realistic 4060 Ti equates to a 3080. Which is impressive but also a shame. Nvidias showing why theyre the best and the worst at the same exact time. Technological theyre on another level, greed wise theyre also on another level. This gen couldve been a crazy leap forward in performance but they didnt have to compete so instead they said “buy the 4090 or accept paying a lot for a little” unfortunately considering the performance and pricing of the AMD’s 90 classes, i dont see it getting remotely better
its not small...its a perfectly good size!
It's how you use it.
the rtx 4060 pcb tear-down !
I'm betting the 4050 PCB is going to be the size of an SSD
Yeah looking at that PCB; that looks like something you would find on a $250 card; 3-4 phase VRM? lol. While the performance was disappointing for the time, the Vega 64 reference had a PCB that was rolls-royce component wise compared to this, for $100 less [$500usd msrp] at the time... and it was technically still the 2nd fastest card on the market ignoring titan variants and early drivers.
Yes by early 2018 it had thoroughly thumped the 1080 in +80% of games. And by the time Turning came out it had gapped it in 95% of games.
@@anasevi9456 4070 ti(previous 4080 12gb) is physically a 4060ti die. The 4070 is physically a 4060 die. You can easily calculate that by doing maths on the cuda cores and all previous nvidia cards and segmentations. It's sadness.
First card I have seen is a MSI 4070 for €759.
FE is a myth lol
in my country it even goes up to 900!
@@yangmaster24 crazy right?
@Skif srsly? For a ××70? Holy bananas
@@cyphaborg6598 holy Bananas, Batman, sir.
It's £589 in the UK
I think you can hear Steve stopping himself from becoming Linus every time he reminds us of the GN Store.
Ultra tiny in all respects but price 😮
The price is about 40% higher than I'd pay for it.. and judging from the teardown, the margins on this card must be extraordinary for Nvidia.
Of course it would, I wouldn't be surprised if they were selling the GPUs to themselves and writing it off as an expense for tax reduction purposes to make more profit.
@@earthtaurus5515 that would result in more taxes
@@earthtaurus5515 that would be what is known as an inter company transaction, which has no effect whatsoever on their financials or tax return as it is removed via an eliminating entry.
@@shadouqh8370 Yes, that would be the case if it's within the same company. Corporations like Google for example move money around from subsidary to subsidary in various regions in order to pay less tax. I imagine Nvidia would have a similiar structure.
@@earthtaurus5515 by definition a subsidiary is the same company for reporting purposes.
Maybe the number on the Card is for quality control and refers to the lot it was taken from.
Ah the RTX 4070. Or how I like to call it: The RTX 4060@70ti-prices.
The smaller the PCB the smaller the cost to manufacture. I do like that this card is very efficient and small, just wish it was £480 instead of £580. At sub £500 it would be genuinely enticing.
Keep in mind all third party units will be even more expensive than the FE model as well. People are already seeing them going for $1000.. for a 60 class card improperly labeled a 70 class..
The entire 40xx line up except the 4090 is shift one class up name wise.
This 4070 is a solid 4060 in performance. The price isn't helping for sure.
why dreaming it costst arround 900 not 500 ! You cant buy it for under 800 ! Why are they getting away with this ? Who is buying them ?
@@Thezuule1 At that 1000$ price point most consumer are straight up buying the RX 7900 XT instead
@@PoppinboyJojy as borderline unfair as it sounds it's almost a advert for founders edition cards. afaik the board partners were targeting a much higher MSRP then what it got in the end so most of those cards are likely going to cost way more then what nvidia claims it should cost. but usually (ateast for my 2070S) I got it at the price nvidia says I should from nvidia ( as any AIB cards was 50-100 euro's more and I just need a reference PCB to slap a reference waterblock on...can't get more reference then nvidia's own cards...atleast you'd think...not sure that's the case with the 3000/4000 series but haven't shopped for waterblocks for those)
Its so obvious, there's 2 memory empty slots because the board was designed for 16GB VRAM but since they want to cut cost even more for higher profit they remove 2 memory chips out of 8 and it ended up with only 12GB VRAM (6 memory chips)
16GB will be the 4070 Super ;) Need to ensure people buy two cards per gen these days!
You re wrong, they re just reusing the board layout between 4080 and 4070.
Fully enabled AD104 has 6 memory controllers each being 32bit, or in other words 192bit bus. With 6 memory banks it can have either 6, 12 or 24 GB of VRAM, 16 GB is impossible (with current 8/16Gbit chips) unless you want to have asymmetrical memory layout like GTX 550ti had
I think it's a cost-cutting feature they're going to reuse the PCB on the 128-bit bus 4060 with probably 8 GB of RAM in eight slots.
Also planned obsolescence
So how do they add more memory chips magically without the bus lanes to support them? Just curious
2x performance compared to a good ol' 1080 TI, 3x the price .... im afraid thats gonna be a solid "no" for me
With way more and better features and driver support for way way longer. Nvida is pushing it back because of their popularity but the 10 series cards days of driver support are numberd as they start to push 7+ years old
@@mercuryrising9758 don’t believe the bullshit. The 10 series is good for a while yet and nobody is going to try and run super-tech games on 7+ year old hardware. NVIDIA are a joke.
@@alarak2159 you've never read a steam review or been in a pc discussion forum have you?
@@mercuryrising9758 But hes right through. Maybe 1060 is bit to low these days, but Im sure most games outside of Read dead 2 / god of war / cyberpunk can be played on high/medium. Back then it was 6th gen intel so if ppl pair it with such a old system it might be hard.
Personally my desktop with 4th gen / 970 is no way near enough these days, but 10-series, ddr4 ram etc really helps.
@@alarak2159 ima be honest, the 10 series good if you dont want high-res or high refresh. I had a 1080 and upgraded to a 3070 ti and its a huge difference for high refresh.
The placeholder pins for VRAM seems to indicate that there could be a 16 GB card. der8auer did also a teardown with the FE and 2 different boardpartner cards and they too have these placeholders.
Nvidia was probably playing it safe with the design. If the public would rage hard enough about the card having less memory than AMD, or if Intel just happened to be able to pull a rabbit out of a mouse and got their card to perform, they could easily change the spec, have two more chips slapped on there and update the firmware. But they wouldn't do that unless forced to as it undercut their product segmentation strategy and it would cut into the profits as they know that most people are still very price sensitive and would not react well to a huge price hike. That's who the Ti version are to snare.
And here it gets interesting. Both the 4070 and the 4070Ti share a lot of similarities. Actually they are basically the same card. They have the same GPU and memory. Only things that differ are the power delivery where the Ti version can handle 85 watt more, oh and the fact that the non Ti version has the AD104 GPU crippled by disabling 20% of the cores. Other than those disabled cores, slightly better power delivery and probably a bit better cooling these are the same cards. The difference in cost for these are pretty darned small, definitely it's less than the difference in price to you and me...
or they were going to make it 16gb and 749.
Or its an earlier 4070ti PCB
I really wouldn't be surprised if Nvidia released the 16GB version later on, especially depending on what AMD's RX 77xx cards will bring in terms of VRAM...
@@you2be839 they'll add "ultra" to the name if it happens 🤣🤣🤣🤣
Ultra ripoff
A measurement of the bare pcb would've been pretty useful for SFF-enjoyers.
Really enjoy these tear downs Steve. Even with the trouble you had with dissembling the 3080 without any instructions, I'm glad I have that video as a reference if I ever need to maintain the card.
In NZ the RTX 4070 cards are around $1300 NZD ($800 USD). The formerly redundant RX 7900XT is now reduced to $1500 NZD and sometimes even cheaper when sales are on ($940 USD). I am thinking paying a little more to get the AMD card might be better value. It has 8GBs more VRAM, but I am not sure about FPS yet.
The RX 6750 is now discounted to as little as $400 USD here, so that's an upgrade option for me too (my CPU is a R9 5900X, but I just bought a 4k 144hz monitor, so I need to upgrade the GPU).
Yeah, I have an AMD bias, because Intel and Nvidia have pissed me off too many times over the past two decades or so. But AMD has to earn my support. If they are in the same ballpark for price/performance, they get chosen (unlike Bulldozer era when I had no choice but go with Intel and the 'a quad-core is all you will ever need' mantra).
If AMD manage to be competitive and we buy their stuff, it will prevent a Blue Green monopoly and therefore monopoly prices (duopoly prices are not much better, but are a wee bit -- as Ryzen forcing Intel to lower prices and have more cores proves).
P.S. The RTX 4090 is a little cheaper now here, but still up around the $4000 NZD mark. I could get a half decent used car for that, or a very decent electric guitar. After decades of buying high end computer parts and seeing them depreciate so quickly, I decided it was like a black hole to forever throw money into. I just buy what I need to do the job now -- rather than trying to keep my PC on the cutting edge like I used to (I was a chronic early adopter and always wanted the best of everything).
Yeah, I am old-ish. OG gamer. Space Invaders and Pacman generation. Then it was Commodore 64 and ZX Zinclair etc. Things have come a long way since I was inputting BBC Basic code on a RISC based Acorn computer at school in 1983 (Acorn was part of what has become ARM).
Here in Mexico its the other way around, old cards are super price inflated while we get the newer cards at MSRP thanks to amazon USA shipping to Mexico
Lol imagine paying $600 for that $300 card, hope the gpu-pocalypse ends soon
Honestly this could be a really good card if it were around $100-150 cheaper
and 4gb vram more and higher memory bandwidth
I just sold my 3060 ti dell oem Model for $200 and bought the 4070, will see how it goes.
Tip: attach a string to the old fan cable before pulling it and use the string to pull through the new power fan cable like you would replacing fiber runs.
It reduces assembly cost, but the origin of it is that Steve Jobs thought screws were ugly and mandated they be eliminated. Then they found out this could also sabotage repair so they were all in
24:42 Yeah. eVGA exiting was a good move on their part. They knew what's up.
like rats fleeing a sinking ship of greedy fools
Cynical me wants to think there will be 16gb vram Super variants of the 4070/ti next year, with an updated bus. Altho that part might be more difficult than what Jensen deems worthwhile - can't skimp out on margins for those leather jackets he's collecting
552 is the total number of founders edition cards ever built worldwide for nvidia. FEs sold out in split seconds here in Germany. 😢
You have been too slow
Isn't Notebooksbilliger the only one selling FE cards? I've got worse news: They're also responsible for the Dutch market (even though their site is completely in German). So more people are competing for them. I decided to get the Asus 4090 TUF OC instead of the FE once the price was the same. Because those FE ones only get in stock every few months or so.
For my SFF case, this is the first latest gen card that would actually fit. 200W is also great! 12gb is a little worrying, won't age as well as 16gb... but hopefully game studios also learn something from the recent 8gb debacle.
12GB is gonna be fine. Price is the real problem of this card.
@@Micromation 12 GB is Not even enough now anymore in some Recent titles in 4k will be Not enough in 2 years for wqhd IT will be the same as with the 8 gbs now
@@benito9017 Idk, I perceive 4k rather questionable under 30'', unless we're speaking ultrawide form factor where all these additional pixels actually serve a purpose other than tanking the performance. Besides buying mid range card for resolution that chews up top dogs, is like buying 3060 on a promise to play with RT on at relevant framerates...
Personally I won't be upgrading to 3x1440p setup (2mil more pixels than 4k) until 5090 comes out and shows some massive uplifts over 4090 because performance just isn't there...
@@Micromation me over here running a 55 inch tv as a monitor from 30 inches away.
nvidia is stretching the price well at rtx 4K. Greed has no limits? 830€ for the cheapest asus rtx4070 or 780€ for an msi trioX. In short, the same price I paid for the 3080 trioX when they came out. Brutal profit margin for a 4060 with the 4070 tag on top.
that is true. The 4070 simply doesn't make any sense for normal form factor PC's, as you can get cheaper cards that perform a lot better.
The only application I can really see the 4070 fitting is as a smaller compact card for people rich people that are making SFF builds. For any one else it does not make a lot of sense.
I picked up a used 3090 for 676€. And where I bought the are cheaper ones as well. Undervolt it and you have a 24gb 4070ti for less money.
@@Slappi2 I don't okay new titles much to be honest. And if it can't run on a 3090 at 1440p the devs can put it where the sun doesn't shine
Gonna make some sick sff custom loops possible with that tiny pcb size
This is a beautiful card and honestly I love it, but that price is just absurd. Maybe if it had 16GB VRAM I would actually consider it better value.
The power consumption is a win. The size of the card is a win and general performance is pretty good. The pricing just kills this card in the market. When you can pick up a 6700xt, 6800xt, or go used from both sides. You can get a better performing card for less money.
1070 was 29% faster than the 980 \ 2070 was 16% faster than the 1080 \ 3070 was 26% faster than the 2080 and this fake 4070 is 3-5% SLOWER than the 3080!
1060 was 5% slower than the 980 \ 2060 was on pair with 1080 \ 3060 was 10% slower than the 2080 and the fake 4070 is 3-5% slower than the 3080!
1060 has 33% of full die cores \ 2060 has 42% of cores \ 3060 has 33% of cores, and this fake 4070 has 32% of full die cores.
Not to mention that this thing is only 30% faster than the 3070 while the uplift from Ampere to Ada is more than 70% (3090 to 4090).
This is $400 XX60 tier GPU and it sells as XX80 tier GPU!
@Krystian Nowak I agree that the performance isn't where it should be. With how much they cut down the pcb and limited the power. The performance is generally good. It's not great but it's not bad. For someone that wants to upgrade to the current architecture it could be compelling.
Lots to criticize about modern nvidia cards, but if you have to do a "screwless" design, this is a nice way to do it. Pretty easy to disassemble, no silliness.
If 7 inches is ultra tiny then I'm practically non existent
Wrong comment section, or is it?
just buy a 4090 to compensate
The absolute love for computer components on gamers nexus is crazy! Thorough infos!!
I'm betting we're gonna get another GPU between 4070 and 4070 ti, due to the price gap. It might be possible nvidia brings back the "super" variant of a 4070 with slightly more vram, at 16gb for $700. It'd have to be 256 bit so they might tweak a 4080 down to similar specs.
I have the 4070 Founders Edition and the design is excellent. This card Overclocks like a champ. A matter a fact out of the 12 different video cards I've had in the past 20 years this one overclocks the best.
Had this come out with 16gb of Vram, and priced at 475-525$
It would be super interesting
But at 600$, with AIB cards only being more, and only having 12gb
It’s a hard pass
Seeing a fairly power efficient, high performance, normal form factor card
Is awesome
Would have given me hope for a really cool, 120-150 watt, 4060 that can fit in SFF cases while stock
Unfortunately, that card will likely have 8gb of Vram, and cost over 400$
We need to hold on buying these things, and let prices drop to decent levels again, which enough Vram to play any game I want to
And at the end of year 2023 Nvidia still will have 80%+ market share… People just want to buy Nvidia. No matter how bad the product is/will be in price wise…
I enjoy seeing the good build quality even on this lower prices cards. At least they are not tricking the customers here by using just simple eye blending plastic. Thanks for the teardown! Always a pleasure.
Could be such a good card, if it was $450.
Very sad!
I love these types of teardowns so much. Please do more!!!!
My guess on the 552 is that the GPU has been pulled out of the production line for QA sample testing, and put back after it qualified.
Those two extra memory spots may be for the rumored 16Gb version of the card, like those listed by Gigabyte in...February?
The 4070 board might not be the biggest, but it sure has got a lot of personality.
That's an impressive GPU design IMO. The PCB is compact, there's no real BS going on to mount it to the cooler. My only pet peeve w/ it is that little decorative element covering the baseplate. I want a GPU that's just a black brick, no frills or anything like that. Just the PCB and the cooler. For what the 4070 is though and the design, they did a really good job on there. I'm surprised!
This is a, finally, sensible card. If only its price was more sensible as well....
Those gray strands are really starting to stand out, but you're not the only one. Getting older is part of life, with age comes wisdom ;)
I disagree that Nvidia are going 'Apple' when it comes to screwless designs.
The iPhone is pretty much the only phone left with exposed screws, two on the bottom.
The Macbooks are super easy to get in to because the screws are exposed on the bottom. I mean, you can't do anything once you open it, but they're easy to open!
Some people might say that a 7" PCB is plenty. Maybe quite large, in fact
That’s what she said.
7 inches is pretty big....just saying...some video cards are much smaller and still get the job done... definitely talking about video cards
its pretty average tho:(
this card definitely has a good personality
@@ORCIproductions more to a video card than length...video cards have feelings too
@@miroradics4565 it's above average...
Thanks for testing the 4060
Hope you guys keep covering the 4070 when the manufacturers versions come out. I have read some of those are lower quality than the FE, due to Nvidia last minute MSRP prices
Most AIB cards are worse quality than FE models on all 40 series cards. Only absolute top end stuff like Suprim and Rog Strix have better PCB than FE.
I have to thank you Steve. I was always slightly intimidated by the thought of tearing my GPU apart because they're so damn expensive. However my 3070 started running 20ﾟ hotter and 15% slower (18 months of 24/7 mining which I made enough to cover the entire cost of my pc) than I was used to so I had to tear it down. The thermal paste was dried out and cracking. After replacing it with fresh my 3070 is running like new again(honestly maybe even a little better then when it was new but i assume i have driver updates to thank for the 2%ish bumb in performance. Thanks Steve!
I love the smell of fresh tech review in the morning
Really envisaging Steve will all grey hair. Hope we all make it that far :)
I feel like this card shoud have been 499$ , mostly because of the 12GB Vram
Then the 4070ti 599$ with a 16gb Vram
Right now it makes no sense to prefer this card over a much cheaper 6800XT or 6950XT that has same price
Nvidia is increasing their prices mainly because they are nearing 90% market share which basically tells them they can do whatever anti-consumer move to rise their margins to the moon. I still dont understand why AMD almost doesnt market their gpus except their own webpage, I think they would gain a lot of market share just by actually marketing their GPUs which have good price and not the rx7900s which are basically the same shit nvidia does in terms of pricing
@@damara2268 AMD is being equally as stupid as nVidia. nVidia can charge exorbitant prices simply due to the fact they're selling you a multi-purpose GPU. AMD on the other hand are selling one trick ponies. The 7000 series are worse at RT, worse at compute, worse at ML, and they have inferior upsampling tech along with 20-40% lower performance per watt. The 7000 series is really only highly competitive in E-sports gaming and a few AAA DX12/Vulkan rasterization games, yet AMD thinks they're going to sell the cards for only a 10-15% discount versus competitive nVidia GPU's. Both AMD and nVidia are delusional right now, and buying anything from them right now is stupid.
For the idiots who forked out $1000+ for midrange cards during the pandemic, despite everyone telling you not to, these are the fruits of your stupidity and impatience. GG.
@@damara2268 imo AMDs market share is hard to take back because a lot of people think ai and some other software only run on Nvidia GPUs. Nvidia took over a lot of the market thanks to paying devs to prioritize cuda and their exclusive features for years.
@@xfy123 people don't only think that majority of ai and compute SW doesn't work or works badly on AMD GPUs, that's true.
But AMD GPUs are much more with gaming in mind than Nvidias that are nowadays a lot multi-purpose.
AMD should advertise it properly, they can easily take top selling charts in regular stores where gamers buy
That was fun and educational! Was nice to see a not pro tackle a pretty big job in obe shot! Was actually pretty amazed how much she learned from watching you. She retained more than I have lol
Paying €700+ (Europe) for this small pcb is a crime 😭
7" is ultra tiny now... GREAT steve. Just GREAT
The 4070 looks like a great card, which is just $100 too expensive. 3080 performance, with the added benefit of DLSS3, but with a huge improvement in efficiency, meaning lower power consumption and cooling requirements (and thus size). For $500 this would be an excellent product
anything under 16gb vram is just pathetic at this point since consoles give better performance
@@visceraeyes525 What are you smoking? The PS5 and Series X mimic the performance of a 6700xt, which is almost half the performance of the 4070. Also, the PS5 shares the vram and ram.. so 16gb of vram in the PS5 is misleading.
@@something3610True on the ps5, but 600 dollars for a card that'll have its vram start to get saturated in the near future isn't the best thing in the world, its compute power is restrained by the low memory bandwidth and capacity. To think Intel offers the most memory per dollar is insane.
That's really interesting. I really wish they would start coming out with newer cards that aren't the size of an Imperial Super Star Destroyer.
Totally unfair, it's only as big as a regular star destroyer 😅
It's too bad board partners can't attach the extra RAM chips to produce a 16GB card.
not really possible with a 192bit bus
even if it was actually possible do you really think they’d charge anything less than ~$700 for it? you guys will just complain about it being too expensive again anyways so who cares?
I wonder if it was originally intended to. Be 8 1GB chips and they revised it to a 6 2GB chips design.
Our manufacturing improvements have cut costs, and we've passed those savings into our own pockets!
14:20 That PCB can accomodate the RTX 4080 I believe since 2 vacant memory pads are there. Hopefully with competition from AMD RX 7800XT series, Nvidia will launch a further cut down AD103 , RTX 4075 with 16GB of VRAM to compete.
More likely a 14GiB version with one controller disabled.
Properly engeneered and placed screws are so satisfying.
This is actually a 4050 no doubt anymore
The Aorus Master 4070 is taking full advantage of this tiny card. Might be the coolest stock air-cooled GPU in history.
whats so special about it? it’s hideously big… looks like a 2 and a half slot design with 3 ugly gigantic fans in a row…just like everything else
2slot was also standard for 1080ti, 2080ti, titan, 3080ti etc, not only 1080
It actually wasn't really standard for the 20 or 30 series. That's when a lot of cards began moving to 2.5 and 2.8 slots (like EVGA's). It was available, but not as common or prevalent as the 10-series.
@@GamersNexus true, I was thinking about standard as in founders editions, thanks!
I would LOVE to see true single-slot watercooled cards back in vogue! And the PCB is so dense that a card that was designed for watercooling from the ground up could be made EXTREMELY compact.
Imagine a world where Framework makes modular GPU PCB shroud add-ins for PC towers, and you could just but a new PCB instead of the whole thing so you save $75 bucks or so. Imagine.
like a new cooler mount like what we have for CPUs. like how the same cooler has AM4/AMD sockets and LGA/Intel sockets mounting brackets?
@@masterdoge17 Yeah. I mean, even to go a bit broader, how we have a whole motherboard that can be swapped out of a pc case and a new one placed in that is better than the last (within size capacity of course). Just keep that thought going as far down as we can through-and-through PCs to save on waste overall in a way that is at the very least as easy to do as swapping a cpu and/or ram on a laptop.
Thanks for the tear down. I've been enjoying the morning upload time of the last few videos as well.
At 17:50; if those extra 2 memory positions were populated, this would be a 16GB card. Maybe they dropped it to 12 late in the day to fit a lower price point.
The memory bus does not support extra memory chips. They could use higher density memory chips, but can not add more memory chips.
What they could use, is to use same pcb with 4080 chip… and downlock a lot.
You can’t just populate additional memory pads. You need a memory controller to actually access it. AD104 has a 192 bit memory bus, meaning 6 32 bit memory controllers. That means that it can handle 6 memory modules (12 if they are back to back). What is likely is that this board may have originally been intended for a higher tier GPU, or the additional slots are there to improve yields on the board itself.
This is the first card since the 10 series that I actually like. Unfortunately it's just too expensive though.
That's what my wife said to me when I removed my pants!
i think the no-screws ideology comes from the idea that visible screws remind consumers that you can take things apart yourself if you need to
The current best gpu you can put in my ultra sff system is the 3060ti. I would love to see if the performance holds up with a small cooler attached
I miss my 3060ti tbh now I have a. 4070ti
Thanks Steve and crew, interesting.
Ha ha drilling the 3080.. the power of magnets :)
HOW DO THEY WORK?!
@@GamersNexus Science, but that's just made up.. so Magic?
@@GamersNexus apparently they stop working once super glue is applied?
The Quadro equivalent of this card is low-profile (RTX 4000 Ada SFF), so this is huge by comparison!
That die size though. Benefits of removing so much bus width (since it doesn’t shrink like other components) and a more advanced process node. Seems that will be nvidia’s strategy outside of the top end card.
Important to understand that the bus width is nothing more than a multiple of the number of physical RAM chips on the card. Each GDDR6X chip, regardless of capacity, has two 16-bit buses. That's why this card has a 192-bit bus (6 chips times 32 bits). The reason this 4070 is better than the 3070 and its 256-bit bus is because the memory speed is 50% faster.
Think of it like traffic lanes. What road would get you to your destination faster? A 4 lane road where the speed is 80km/h, or a 3 lane road where the speed is 120km/h?
@@warren_r no I get that haha. It’s not directly comparable across generations of memory or even higher throughput memory of the same generation. It’s still definitely a cost cutting measure as IO doesn’t shrink like it once did moving to more dense nodes. I do think that 16GB of memory on a proportionally wider bus would give this card better performance over the long term though but that would add cost and cut into nvidia’s margins and/or raise prices for end users. The cost of adding the extra IO on silicon due to how much die space it takes up would be high and for not much more in transistor count. It’s proportionally more expensive to do on a denser node and this trend continues on in the future. This is one benefit of the chiplet stuff AMD is doing although im unsure how it pans out long term, seems there are some issues with this first gen.
I know this has 50% (more memory bandwidth than last gen but it’s also more powerful. Maybe more memory and a wider bus wouldn’t make any difference on this or the Ti but even if it did, my hunch is nvidia is happy to give people enough performance at this tier to buy it but not to be a great value either
@@warren_r and to answer your question: how about 4 lanes at 120?
@@jolness1 Whether you realize it or not, you're asking why the 4070 doesn't have 16GB of VRAM. The answer is probably money -- those Micron 2GB/24gbps chips cost about 30 USD each.
14:12 where our 2 more memory chips, NVIDIA?!
Oh if the pcb is this small they gotta make a low profile for once in 5 years
That would be the only real way they could carve out a good Market for this card. As there are a lot of wealthy SFF builders that are willing to pay a premium for extra power in a small package.
They did? It's a two slot card. Not "two and a half" which is just 3 unless someone can show me a half slot PCIe card I can use in the other half.
@@mjc0961 🤦
@@mjc0961 Low profile means half-height, not the number of slots it takes up.
It’s gigantic and has a good personality, calls back etc
@15:19 in like 10 years, I see the lack of GPU label on core as a potential problem with GPU scammers. Like now when you buy 60 buck RTX2080 on aliexpress or wish, you can at least disassemble it and tell immediately what it is, even if BIOS is flashed. After these cards age to the point of being basically e-waste, people will still be scamming others with them, but it will be very hard to determine what you get, especially if Nvidia sticks with this design for the next several generations.
I dont think these cards will be e waste for a long long time, because they will always be able to play older games very well. Theyd have to be to the point of non-functioning to be ewaste imo
Times changed, these Fermi era GPUs they use to scam people were from an era where new DX versions obsoleted a card within 3 years of it's release. I'd be worried if i knew DX13 was in the works, but knowing how games focus mainly on PS5 for their development that's not going to be a problem anytime soon even if that were the case.
Its really surprising that we still dont have a mini itx 4070 card yet
As bad value as this card is compared to stuff like 6900Xt and such, it's actually quite impressive Nvidia managed to cram this performance into such physically small card, and not only manage the thermals, but also do it all at quite energy efficient levels. As greedy as they are, they do know their stuff when it comes to GPU design and manufacturing.
this card is going to be the hot ticket item for the 4000 series once it gets a price cut of at least $100 and the used ones start trickling through - a 3080 for such a low wattage and a small form factor is a killer proposition for all gamers that just want a very capable 1440p card or a bulletproof 1080p card.
Sadly Nvidia just has to murder the prices on these after we've shown them that we'll buy 3080's for double MSRP. God damnit.
One thing to note, the engineers were the ones who designed this so props to them. The marketing/higher ups/financial what-not dept were the ones who decided on the price so they're the greedy ones.
You mean TSMC right? Because most of the improvements from Ampere to Ada on both power and efficiency are a direct result of going from Samsung 8nm to TSMC 5nm. This has more than double the transistor density of a 3090, it's not like they made some incredible design improvements this generation.
@@kajurn791 TSMC only applies to the core. The overall board design, power phasing, cooling, trace layout, all that stuff is still done by Nvidia engineers. Transistors, however small or power efficient they are, can't do anything without proper cooling and power delivery and other support systems.
Asus are likely to be one of the companies to make an ITX scale 4070 GPU as an updated Dual RTX Mini, the last ones they built were the Dual RTX 2060/2070 Mini; the later being a very competent GPU.
I guarantee there’s gonna be a Phoenix 4070. Just seems too perfect not to do.
When it’s cheaper this would be a pretty cool itx build card
This somehow looks like that this are the 4080 12gb leftovers that they couldn't sell anymore after the shitstorm around the 16/12Gb vram versions, and perhaps they just replaced the resistor to make it less powerful (what would be a ridiculous move) and sell it as 4070.
This "cleanness" on the PCB separation just proved my theory that the 4090 was ready and stored in some secret warehouse for a lot longer than we might think otherwise. This 4070 is so fresh, that the thermal paste was still fluid.
kinda off topic here but, I am a trucker and I use a Bluetti ac200max power station to run a Lenovo legion 15ach6h laptop (3060) on my off duty time, and while that is sufficient for a 10hr break it falls short on a 34hr and I end up needing to idle for a couple of hours to recharge every 10-12hr (truck has a 20a inverter and have dual AC chargers for the power station , but I prefer not to, always want to try for that MPG bonus).
I do not have many new games as I mostly play Mudrunner/Snowrunner, or the Borderlands series era stuff, titles that all run at well over 120fps on this system maxed out but, pulling 300w load. I have began limiting FPS to 75/80 on the shooters and as low as 60 on the driving games, but I hit a point of diminishing returns where the system stops reducing its power draw. it never seems to drop much below 175w if the 3060 is active on any sort of 3d load no matter how simple.
I have noticed a few articles hint that the 40 series is supposed to be quite a bit more power efficient and I wondered if that applied to possible gains in low load efficiency as well.
I have had people tell me to just get an overall less powerful system but I would like to maintain as high a level as possible as I occasionally Use Solidworks and who knows there might be a new game someday that I actually want to buy.
I would love to build a SFF system as I am space limited (Vovlo vnl 760) but I am very concerned about shock/Vibration causing damage to expensive components), bonus kudos given if anyone knows of a case designed to keep components safe in such an environment?
This will be amazing for people looking to build small form factor!
It's just normal form factor
@@HellGatefr2 Yeah, but its a lot smaller than other 40 series cards, meaning it will fit better in small form factor cases
I’m pretty impressed with how quick and simple the tear down was! It’s way better than Intel Arc!
This would be cute if this card wasn't hot garbage sold at a sky jacked price.
Well it is nowhere near hot and its not really that expenisve so Idk if you got the correct information in AMD Fanboy Camp :b
@@AlpineTheHusky I literally have a 3080 in my PC. Cope and seethe fanboy,I'm simply objectively looking at the corpos.
@@kizuati I just love triggering AMD fanbois since they rarely defend their point and just go "Fuck you how dare you fanboy" or "I got nvidia stuff so I cant be a fanboy" I havent had a single bit of Nvidia hardware outside of my current 1070ti. Currently only got AMD stuff except for a single NUC.
@@AlpineTheHusky I don't even like AMD you brain damaged mong as they're doing the bare minimum with their cards right now and only look good next to NVIDIA.
Good to know you're your mummy's good boy who argues to strangers he's the smartest guy in the room in regards to GPUs.