Hidden NVLINK? RTX 4090 Aorus Master Teardown and Overclocking to 3GHz

[@rustler08]

Good to see how small the board is, water cooling the 4090 will give you a very tidy package compared to the monstrosity that is the stock cooler configuration

[@UhOhUmm]

So when you watercool the radiator pump and fans do not count? They vanish into the air right.

[@YukiSorrelwood]

@@UhOhUmm Pump and Rad are still a thing, but we are talking about going from a 3/4/5 slot card to a 1 slot card, a massive reduction in weight on your PCI-e slot, improved airflow, and most people who are going to disassemble a 4090 and put a block on it, are almost always the same people who have a whole loop built for their CPU and they are adding a GPU to the loop now. This helps for small form factor cases....and a lot of just mid sized cases too, the card is the size of an Xbox Series S, its huge. Even with a pump and rad and fans, its still going to be smaller overall.

[@fredEVOIX]

WATCH OUT !!! no they are not that small only a handful of them are...I got screwed on that with my 3090 the pcb is 10cm longer than the FE I spotted several 4090 with long pcbs too

[@RifsxD]

​@@UhOhUmm atleast the bulk is shredded by a big amount on top of the mohterboard and about the radiator and pump they have their own separate compartments which makes the setup a little bit clean instead of the whole brick taking up most of the motherboard space

[@TheEchelon]

@@UhOhUmm At least you can tuck away the radiators or configure it however you want, smartass.

[@ImDubbleYou]

Please ALWAYS include thermal pad sizes. Beyond nice to be able to find those prior to tearing down a card. Thank you!!

[@peteelderman8606]

8:17 lmao wait a sec he did measure the pads. Cmon man 😹

[@les_railgun]

@@peteelderman8606 he meant to include pad sizes on further videos

[@peteelderman8606]

@@les_railgun hmm

[@theecstatic9686]

Cut your own pads to size it doesn't matter....

[@manusnz]

@@theecstatic9686 I don't think OP is asking for the length/width dimensions, the thickness is far more important though.

[@50H3i1]

I love der8auer videos . Because when everyone doing reviews and same things . Your videos are different from everyone and also interesting topics .

[@malborboss5710]

der8bauer is a GIGACHAD

[@LordHypnosis]

Same, a lot of the content Roman produces is really unique and you always come away having learned something. All you need is the Holy Trinity of der8auer, Gamers Nexus and Actually Hardcore Overclocking - in-depth analysis and none of the tech bro buffoonery.

[@50H3i1]

@@LordHypnosis agreed . And for just straight up true benchmarks hardware unboxed

[@andreasotto1977]

Igorslab stated the other day, that before switching to TSMC 4N node, nVidia calculated some 600W+ power draw and AIB partners designed their coolers accordingly. And after the switch and recalculation, it was too late to design / order new coolers. So what we see here, is a generation of GPUs with comically oversized coolers, that come with a generous overhead in heat dissipation, I guess.

[@kosmosyche]

Honestly, all things considered, it's better this way than the other way around.

[@zacharytaylor8523]

@@kosmosyche It is, of course the AIBs aren't going to be too happy as there is less incentive for people to buy higher end models if all models have decent cooling

[@seculi7757]

It`s a mining card. Games are just not able to fully load the card.

[@pearce05]

@@seculi7757 not a mining card, it's a workstation/ai card

[@jannegrey]

@@pearce05 Without Titan drivers it isn't as useful. Though I agree - it still makes sense if you can utilize it that way, it just seems that NVIDIA is taking away usefulness with every generation (NVLINK in this case).

[@popullarmmax]

Increasing the performance by 10% while going 200w more power...this card at 300w is a beast

[@damonm3]

This is my takeaway as well. My card while vsynched peaked at 65% power draw in a 2 hour gaming sesh in destiny 2 at 4k 120 fps at max settings. So pulling less power than my 3080 for a slightly better exp. This game doesn’t need a 4090 to run well but obviously the new UE5 games coming will🔥🔥🏆🏆🏆

[@savifrank6323]

@@damonm3 What upcoming UE5 games are there??

[@damonm3]

@@savifrank6323 there’s a list. Not sure exactly.. but they’re (some) for sure coming in the next year. We need some new aaa games for sure. Lots of remakes…

[@savifrank6323]

@@damonm3 looks like the new stalker will be on UE5 that looks the most interesting to me

[@EarlywineJC]

Thank you for measuring the thermal pads. There was so much confusion about best size on last generation, knowing the multiple thicknesses on the card is very helpful.

[@notjulesatall]

They're probably just reusing the design for Quadro or Tesla cards, and disabling NVlink in the BIOS. It's common for NVIDIA to differentiate hardware only at BIOS or driver level.

[@rkan2]

How about if the rumors were true that Nvidia didn't update the spec for the AIBs and they just went with the original reference designs?

[@Oz-gv5fz]

Reuse 3090ti board, include the display output port 😞

[@1BigBen]

or the RTX 4090Ti may have it

[@JJFX-]

Could be anything and may also just be reusing a PCB design. Personally I think Nvidia is considering a Titan this generation which would explain the loss of NV link. Aside from what I assume will be an insane amount of memory, that's the only feature that'll really make anyone willing to drop upwards of $3K on a card.

[@terrablader]

Let's nod it to quadro , anyone remember those days of getting x5 Open GL performance for CAD

[@hanes2]

They removed it on the new (Quadro) RTX 6000 Ada. Saying apparently it’s not needed anymore. I’m very interested how it will stack up (literally)

[@jlgroovetek]

Great video. It seems you're the first to tear down the Aorus Master. Certainly 100% same PCB as the Gaming OC, but as you left all the pads on it was hard to see if it has same number of phases installed in its VRM.

[@CoalitionGaming]

When you said the card is still tiny without the cooler, I was just in awe of how wide the PCB still is regardless of that lol. That could be problematic for a custom loop actually since terminals on the side add even more width to a card.

[@tonymorris4335]

this is what I'm worried about with my 011 dynamic xl. I contacted ek to see why they're doing a side port on a card that can barely fit in most cases but without the coolers have plenty of end room for them.

[@Kepe]

@@tonymorris4335 Yeah some water block makers are making blocks with terminals at the end of the block, increasing length instead of width.

[@hectorm8649]

Comino has a beautiful professional block for this card with the terminals at the end instead of the side.

[@damonm3]

What a time to hit 100k!! 🎉🎉🎉 congrats🏆🍻🍻🍻

[@der8auer-en]

Thanks 😬

[@B3TR0Z]

"I want to see Ada naked" - der8auer 2022

[@navarropuff187]

took me by surpise when I first saw the video when you held the Aorus GPU and it was bigger than your head, Roman. Nice teardown video! Really admire the work you are doing for PC tech.

[@JasonDarbel]

Always round time stamps at the end of these videos, makes everything better Much love from France and great content!

[@samarthtibdewalGL]

0:17 You got us in the first half

[@Toxicity1987]

That NVlink stuff could be because the AD102 and the GA102 have a very very similar footprint and could be mostly compatible to each other. so manufacturers could use the old PCBs with some minor changes and just didn't cared about the old NVlink traces since it doesn't bother anything.

[@miha493]

Real reason why nvlink was removed, in my opinion, money. These AI enthusiasts, 3d designers and other very bad, in eyes of Nvidia management, people's who need big graphic buffer but din't want to buy 5 times more expensive Quadro GPUs, should suffer. Even if it's just your hobby or just it's small startup without a few extra hundreds of thousands of dollars, now you should pay full 10 000$ for upcoming Quadro version of 4090, but with ability to connect cards.

[@halrichard1969]

Agree.

[@alexamderhamiltom5238]

agree

[@SuperTort0ise]

With pcie advancing so much shouldn't it be easier to just use pcie and not have nvlink, also how would you fit 2 4090s you could water cool but it's still 1200 watts. (⁠＠⁠_⁠＠⁠;⁠) But yeah fuc Nvidia they're not doing this for any of that they just want money.

[@ChristianStout]

Wrong. The new Quadros don't have NVLink either. Going forward only the Tesla-class cards have NVLink.

[@Jumbo_Smooth]

@@SuperTort0ise All companies are in business to make money...

[@TheSlickmicks]

Yes! Please test the changing of the thermal pads. I would like to see that. I wonder how cool this can get.

[@dime4026]

Why waste time with a minor change in thermal pads. Just water cool it peasant

[@extreme123dz]

@@dime4026 Water cooling is for fucking poor trash peasant, nitrogen is the way to go!

[@demetter7936]

@@dime4026 Because its interesting to see? Neckbeard comment.

[@dime4026]

@@demetter7936 come back when you can afford a 4090 and a water block peasant. Mine is in the mail. waiting for ek to release a water block to go on my 5hp water chiller

[@demetter7936]

@@dime4026 Bro is bragging that he spent $1400, you need to be humbled and realize there's people many many people much richer than you.

[@godtiermedic]

Hi Roman I swear you are the most respected tech person in the world. I hope that we can get you to 10M subs one day. You deserve it. Now to try to oc my 4090. Thanks for years of amazing videos

[@metalmaniac788]

Would love to see a video where you replace the thermal pads and test again to see if better quality pads help.

[@Themisterdee]

Thank you for the video. With the voltage limited then AIBs are just furniture / Styling variations .you can see why EVGA walked away.

[@SashimiSteak]

This had been on my mind as well. Anyone can download afterburner and tweak the card themselves. There's almost no incentive to buy AIB models.

[@Jrood1989]

@@SashimiSteak you can get an ugly card for more money, seems beneficial. I think AIBs will slowly depart like EVGA did

[@rawswisherzstreams6606]

Always bringing the best, professional content in english so much more accurate and cleaner than all the American or Canadian youtubers

[@MathieuFRAPPIER]

Try a shutter speed of 1/30 to make the spinning lights more realistic

[@evanescence2]

Here's the reason why you can't add your own GIF: In RGB Fusion 2.0, when you select your GIF and resize it, there's no "cancel or ok" button showing up at the bottom of that window. That's because the application wasn't coded to automatically scale that particular window to your screen's resolution (dumb oversight) Here's a quick fix: Lower your screen resolution a lot, until you see the "cancel and ok" buttons, then you'll be able to add your own GIF :)

[@PCBWay]

Nice content 👍

[@emu071981]

I still cannot get over how big these 4090s are. I always considered some of the custom 2080 ti's to be big cards and the 4090s just blow them away in terms of board widths and cooler sizes lol

[@Quizack]

The first time I saw a 2080Ti I thought it was comically large. Now that the 4090 is released I’ve realised that Nvidia does not care about size at all. As long as the cooling needs more mass, they’ll happily supersize it. It doesn’t bode well for future GPU sizes

[@ruikazane5123]

For voltage adjustment...a rather hardware-level way would be to go straight to the power supply controller like the older days. (though probably use I2C and write to registers instead of jumping resistors or putting a pot) Requires a card that have adjustable voltage supply. Good luck on OC!

[@meeponinthbit3466]

And then Nvidia will start going the route of Apple and using serialized chips (v controllers) and encrypting their control bus to the GPU to prevent hi-jacking.

[@DimkaTsv]

@@meeponinthbit3466 Shunt mod exists. It looks gross, but it is a way that extreme overclockers remove limits from voltages. There are drawbacks ofc, though.

[@meeponinthbit3466]

@@DimkaTsv pretty sure those shunts are just for current sensing resistors on the 12v rails for total power measures... 40s aren't power limited there, so that wouldn't matter. It's the VRM output that needs tweaking, and with digital VRMs, that gets real tricky if Nvidia decides to be jerks about it.

[@blubaustin1]

@@DimkaTsv that's for current sensing. For voltage someone needs to either reverse engineer the voltage controller, or bypass it completely with some external regulator.

[@DimkaTsv]

@@blubaustin1 yeah, sorry, i am bad then. But.they are doing this one as well. Even though external voltage for sure doesn't look pretty

[@mme725]

Gosh, can't even imagine SLI 4090. That much power would be insane.

[@Kattemageren]

Sure people have said this on every new GPU generation reveal

[@user-vk2cd9qw7i]

@@Kattemageren only for the past few launches.

[@BBWahoo]

1200 watts

[@mme725]

@@Kattemageren d'oh that's probably true lol. I only said it because of the potential nvlink connector. Besides that I've been operating on the assumption that the likes of SLI and crossfire have been dead for ages....unless Intel tries it out 🤔

[@ic_evox136]

I wonder if Nvidia is reserving NVLink for 4090ti

[@gtbrown001]

So could you theoretically scrape off the top layer, cut out the connector profile, reflash to BIOS to enable NVlink? This would be useful for some compute applications. You know, the kind nvidia wants us to buy "quadro" or datacenter cards for?

[@samsak1239]

Here is the funny thing Nvdia does not support NVlink in any of its rtx series which includes the formerly known "Quadro" lineup. So now it's limited to only their tensor cards which are Data center cards so even paying thousands of dollars for "Quadro" to do professional work it isn't enough for Nvdia. You need to buy datacenter cards to have NVlink enabled.

[@blubaustin1]

well there's what the person said below plus you can't flash a modified bios on a nvidia card, it just will refuse to boot.

[@wolfzert]

7:06 yep, always GB VGAs have this problem with cooling the circuits causing a burn around the year or less if you intesify the load on the vga

[@QuentinStephens]

That NVlink connector will be there for the Quadro / Professional cards. Now we know that they use the same PCB.

[@BenQuigley]

so they are actually retiring the name Quadro it seems, at least for this generation. also even the Quadro class cards don't have nvlink anymore

[@QuentinStephens]

@@BenQuigley Both are news to me. Thank you.

[@KooYu]

The PCB is so small that with the proper "sleeper" cooler it can pass as a 4050ti or something.

[@arsenal4444]

especially since someone undervolted and underclocked theirs 130W lower while only losing about 5% performance

[@fredEVOIX]

the FE once watercooled is the size of a 1050ti..I did comparisons but AIBs usually have longer pcbs

[@garyhenry7193]

Yet another superb video, this guy knows his stuff.

[@TheLevitatingChin]

I water cooled my 4090 with a 420mm Rad and its a perfect fit. It brings the card size right down and thermals/noise are very low.

[@kuricature]

@der8auer EN Could you take a minute to let me know what the compatibility is with your cases? Were you able to fit the AORUS Master in the O11D Evo? Vertical? Upright?

[@toonnut1]

I always look forward to your videos

[@zakelwe]

In the old days it was soldering on a variable resistor to one of the IC legs and a few turns on a screwdriver for more gpu volts. Though I guess the contact point is tiny nowdays :)

[@virinom]

Also we were using pencil to lower the resistance of shunt resistors for bypassing power limit. Good old GTX times.

[@Bayonet1809]

In the old old days Nvidia didn't restrict voltage control.

[@frizzlefry1921]

@@virinom Or the pencil marks between contact points on Athlon cpus for overclocking.overlooking. At least temporarily to test for stability.

[@someguy4915]

Most ICs these days are BGA so there's no contact point, it is sandwiched between the PCB and IC, no space to connect a wire unfortunately.

[@zakelwe]

@@someguy4915 Thanks, I was guessing so much. Maybe one of the graphic card company's could get a niche on that. Sadly not EVGA and Kingpin. Kingpin rocked on xtremeoverlocking back in the day after macci, OPPainter et al.

[@titanium4167]

Me after modding Resident Evil 4 0:17

[@jmlinde]

At the rate these cards are growing there won't be any room in frame for derbauer in just a few years.

[@tamer.d.official]

So I have done some testing with my Asus TUF 4090 OC and what I have managed to get 2910 MHZ with 100% Powerlimit settings in Afterburner +195 Core Clock + 200 Memory my GPU Temps are around 55-60 Celsius. Im quite impressed about my card to be honest :D

[@robertfontaine3650]

I would love to see someone solder up the nvlink connections and test if they are live.

[@gianlucapx]

The new H100 gpu has 4th gen NVLink, so NVIDIA hasn't abandoned the technology, because it has lot of advantages. Maybe those lanes are there because we will see NVLink coming back in a 4090 TI version, or in the future evolution of the RTX 6000 professional cards, that should use a very similar pcb. Meaning that they left the door open in case there is demand for that.

[@savifrank6323]

The RTX 6000(Full AD102) doesn't have Nvlink so I'm really not sure they would add it to the 90ti/Titan

[@Cypeq]

Other Channels: first 4090 partner card review! Roman: TEAR IT DOWN!

[@KEWords]

These are all laid out in cads. Could be something in a base board template that survived. Like a footer placeholder in a document that makes it to printer.

[@TimLongson]

Great video as usual. Can you please ask the manufacturer if they will be producing a pre-custom water blocked version, like some companies (like Asus) did with some RTX3000 cards in collaboration with EK water blocks? If so it would reduce a huge amount of e-waste & make people who are nervous about putting a block on for the 1st time far more likely to buy one. Thanks in advance. :)

[@halrichard1969]

MSI has one.

[@TimLongson]

@@halrichard1969 isn't that just an AIO integrated version, rather than a card you can plug into a custom water loop?

[@halrichard1969]

@@TimLongson Ah yes that it is an AIO version I believe its a 280.

[@TimLongson]

@@halrichard1969 which unfortunately would be useless for a custom water loop. Custom water loops can provide much more cooling, heat headroom for overclocking and run more quietly than any AIO - the more radiator surface the more passive cooling there could be, so I am planning a three 420mm radiator custom loop in a Cooler Master HAF 700 case. But thanks for trying.

[@TimLongson]

The more radiator surfaces the more passive cooling there could be (as well as more cooling, heat headroom for overclocking and running more quietly), so I am planning a three 420mm radiator custom loop in a Cooler Master HAF 700 case. But I am reluctant to install water blocks on a GPU myself, firstly because of all the e-waste (throwing away all the fans and cooling pipes which were included in the price paid for the GPU), and secondly because of fear of installing it sub-optimally or even damaging the GPU whilst doing it which could render the warranty invalid.

[@pete2097]

Pretty sure it was reported that ada would be a drop in replacement for ampere, I'd argue that the board was even designed with ampere in place then just switched out when it came to manufacturing

[@lickrish3930]

that's mental THAT 4090 WAS HUGE YESTERDAY WOW I LOVE IT.

[@Hephera]

ooh that fanblade-mounted LED is interesting. I wonder if in the future we'll see this extended to multiple LEDs all the way along a fan blade so you could generate text or images or even moving video on the fan as it spins

[@N4CR]

OF course it will go that way. but you can't do it now, because then you can't advertise that feature on next gen/future models...

[@seanet1310]

The size and shear thermal capacity of these coolers is just mind boggling compared to prior gens. Hopefully voltage can be unlocked more so they can be pushed to the limit

[@GibRealistic]

You could give a try using nd filters to slow down shutter speed and get smoother looking moving leds.

[@All_I_can_say_is_Wow]

That NVLINK is quite interesting. It shows that these cards have been done for some time. How long has it been since they said they wouldn't be supporting NVLINK? Almost a year?

[@royalrepublican3480]

just got prices for the 4090 locally, ASUS GeForce RTX 4090 24GB ROG Strix Gaming OC is casually 2544.98 euro or 2465.50 dollars. quite a lot

[@stepa_33]

Excellent work, thank you for this!

[@Johnjrambo75]

The size of the Aorus 4090. It's like a breezeblock. You could kill a large man with it.

[@peigen7151]

1:50 well you could record a small video in slow motion, would that not show the effect better?

[@imperatorja9502]

Nah, They are going to unlock Nvlink later, call it A6000, A7000, Titan or whatever and sell it for $6k.

[@BenQuigley]

I mean current gen "Quadros" don't have nvlink, Nvidia said it just wasn't needed anymore

[@hobblyjig]

I was excited to get a 4090 today. But they sold out in seconds and are all over eBay at $1000 over MSRP. It’s the 30-series issues all over again. I’ve been building PCs since 1999 and I think this may push me out of the hobby for good.

[@startedtech]

Unfortunately scalping like that is common for new hot releases for everything these days, not just PC stuff. Give it a month or two.

[@savifrank6323]

Nvidia already said they are intentionally keeping stock low through 2022

[@thericethatsmilesback5464]

I did a undervolt on my 4090 Zotac OC, it plays 2 percent better than stock at 80 watts less, 7 degrees cooler too!

[@frostbite1991]

water cooled 4090 in a mini itx build would be cool to see.

[@rossharper1983]

Those RGB fans seem pointless, how many cases can do a 3.25 slot card vertical mount and still give it room to breath

[@Mateus01234]

only now i noticed how tiny roman is next to those heatsinks

[@concinnus]

I wonder how much you'd gain if the voltage actually went up by the requested +100mV, instead of the 55mV it seemed to gain.

[@fleurdewin7958]

1:53 Any effects on such a power hungry card blowing hot air on the PCH and maybe M.2 SSD since the fins are directed to the PCI-E slots ? X570 boards with fans might suffer from this, the PCH is already hot.

[@sascenturion]

Interesting stuff ...enjoyed that 👍😀

[@leitomitch9778]

damn that card is massive, interesting you said about the NV-Link, just imagine if they did enable that and you had two of these cards in your system... what would be the heat and power output be??? .. that would definitely would have to be watercooled but thanks for the teardown

[@paulkendall6069]

It would be interesting to see if a modder could get NVlink working as well as raising Voltage allowance.

[@cj09beira]

if i am not mistaken they likely just changed the board shape and hid the pads, it might be possible to simply file down the extra pcb material carefully and endup with a working nvlink, question is who has the balls to try

[@BenQuigley]

raising the voltage is easy, a shunt resistor mod and boom, done

[@cj09beira]

@@BenQuigley those shunts sense the Current, and are part of the "Power" limiting circuit, it wont change voltages at all

[@kevinerbs2778]

@@cj09beira there isn't connection on the die either.

[@cj09beira]

@@kevinerbs2778 really?, then RIP, only custom pcb can save us then, btw source of the info?

[@ostryjanusz]

@1:52 you can with an ND filter :)

[@BoomixDe]

You could try to dig up the NVLINK and try to connected it.

[@vik1ng313]

0:17 👀👀👀 Me too der8auer, me too...

[@iSammax]

It's really a shame we didn't get to see PCB analysis for this one. What components are being used for GPU voltage here?

[@Quetzalcoatl999]

My bet is Nvidia removed NVLINK just for this initial release of cards. They'll release the same cards with NVLINK enabled next year to catch another wave of buyers.

[@camofelix]

Being as many AIBs are reusing PCBs from the 3090 series cards, could well be that those traces are remnants from those cards.

[@AlfaPro1337]

How are they reusing PCBs from 3090 when maybe the 4090 BGA packaging is so different from 3090 BGA packaging? That's like saying ASUS reuses FM2 PCB for AM4 platform, because 'just one connector is found on older platform'.

[@UNSTOPABLE40XBOX]

@@AlfaPro1337 i believe that they where using the 3090 ti boards when they where testing power and stuff.

[@AlfaPro1337]

@@UNSTOPABLE40XBOX Dude, even a 3090 Ti BGA package isn't the same as a 4090 BGA package. Did you even read? That's like saying FM2 traces fits on an AM4 socket.

[@UNSTOPABLE40XBOX]

@@AlfaPro1337 i never said they put 4090 chips on 3090 boards. Did YOU even read what I posted? I was just clarifying what FCLC said.

[@UNSTOPABLE40XBOX]

I'll clarify. If I remember correctly, they used 3090 ti boards when they were designing / developing the power delivery for 4090 GPU's. The 3090 ti was kind of like the testing grounds when they were prepping for the 4090 GPU's.

[@TeeraLucksanapiruk]

Maybe you can try to see if you can get that NVlink to work

[@infurnucide]

My Aorus 3080ti just got software supporting it one month ago. 9/22 it was not recognized by Aorus Engine or RGBfusion 2.0 until then. Not sure if they would give you an answer, but I doubt they intend to provide software support in the near future.

[@squatty9994]

I hope you got that card dinner before stripping it down and exposing it to us like this!

[@cmill4836]

I wonder if the voltage limit is due to the sense wires on the new ATX 3.0 connectors. If you are using a new power supply with ATX 3.0 will it allow the voltage tuning to be unlocked?

[@someguy4915]

The sense wires have nothing to do with voltage adjustments (just plain old 12VDC is delivered), likely they couldn't get it stable with variable voltages and so just locked it until they can.

[@cmill4836]

They do impact power limits though, I know that is different than voltage limits, but it seems we still don’t fully understand everything these new sense wires impact. GamersNexus just posted their review with a clip from a future video where a Nvidia rep was talking about just that. If all wires in the spider adapter are connected, you can adjust power limit over 100%. If only 3 are, you can’t.

[@someguy4915]

@@cmill4836 The way the sense wires work is exactly documented in the ATX 3.0 spec, if gamersnexus is suggesting it is a mystery, that's just clickbaiting (not sure what exactly they said, haven't watched their videos). There are 2 sense wires related to power delivery, they are either left floating (not connected to ground) or tied to ground. Those are the two options. That gives you 4 variations between the 2 wires for 4 different power modes, none affect voltage. Any voltage tuning/OC would have to be handled entirely by the GPU itself, the PSU has no role in that beyond supplying enough current and the ATX 3.0 sense wires do not play a significant role here beyond being identical in function to the jumpers that we used to use in configuring the primary/secondary CD drive and floppy disk etc.

[@someguy4915]

@@cmill4836 Like I said that would be something inside of the GPU, that has nothing to do with the PSU. The PSU design/spec is very simple in this; there's 4 'communication wires' (just jumpers). Two are for a PSU ready signal and GPU/PSU detection (just to detect that there is 'a' GPU, not which type or anything fancy like that. The other two are for current demand, used by the GPU connector to determine whether 150, 300 or 600 watt should be delivered by the PSU. There is no 'third wire' to disconnect or not to enable a power limit adjustment, that is purely inside the GPU and only because NVidia makes it that way, not because of PSU spec, requirements or limitations. Like I said before, the ATX 3.0 spec is readily available, read it yourself and you will see that there's no hidden secrets or unknowns/mysteries/conspiracies, it's all very clear and quite boring ;) NVidia somehow chose to go with the 12VPWR connector instead of just using the 48VPWR that ATX 3.0 also has, it would've delivered 600 watts far more safely without all the issues that Molex is reporting about the PSU and GPU connectors melting within 10 insertions...

[@davekelloway3337]

I had a 3090 Xtreme for a while, the software for that screen was a disaster. If I wanted to change it I would end up having to DDU out everything, reinstall the software and make the change. Then it would stop recognizing the card again. If you get it, expect the screen to be the animated Aorus and not of any value.

[@fredEVOIX]

as heavy and cumbersome corsair icue is (because it does all possible hardware in 1 app) my experience with msi,g.skill,gigabyte,thermaltake,asrock is that their software are betas at best, barely useable and in the end I disabled all their rgb and dropped their special features because I didn't want to use those awful apps

[@WarbossRB]

Got a TUF and hit some top 10 HoF scores on 3DMark :) I am happy

[@FranCal92]

No memory junction temps on this cards? Or is it a GPU-Z issue?

[@Hanginthere4100]

63 degree ? I guess "bionic shark fan" is work well. Waiting for dark power palit and what is the result

[@ishmaelisaac4579]

you said "see Ada naked" right as someone walked past me in the laundry mat. I got a weird stare. just a funny moment.

[@WXSTANG]

It would be cool to see the card hooked up to an oscilloscope for power draw. Wondering if there are power spikes like the 30 series.

[@savifrank6323]

GN already tested for it and the spikes were not as major as 30 series

[@TheBlueBunnyKen]

I would like to see how much the temps drop when you change the thermal paste and pads

[@heeryanus]

Maybe if you ramp the fans to max to drop the temp, you could achieve slightly higher clocks?

[@jannegrey]

He said that he couldn't.

[@PrettyboyPa]

No there isn't enough voltage to clock higher it's not a matter of temperature at this point.

[@heeryanus]

@@PrettyboyPa Yes but at colder tems you can clock higher at the same voltage. So since +170 wasn't stable now, maybe it wouldbe at slightly lower temps

[@Krmpfpks]

Where some see bad design or manufacturing defects, der8auer sees overclocking potential!

[@cjbtlr]

I wonder if they gave you a binned unit given the two different pastes.

[@atemoc]

This card is so huge...

[@christophermullins7163]

I wonder why they limited it.. perhaps to add value to a 4090 ti that IS voltage unlocked?

[@GewelReal]

you're supposed to use the curve in MSI afterburner

[@Bayonet1809]

Nvidia limits the voltage of all their cards.

[@dangingerich2559]

2:00 - Is that a CarBot animation? It looks like his art style.

[@puciohenzap891]

58*C at 400W? Holy SH*T I ordered the 360mm Waterforce Xtreme, should be even more insane, maybe sub 50*C at load with Noctua 12x25 pushpulls.

[@Backup_Jack]

can you use liquid metal on it to see how much better the temps are.

[@axi6ne8us]

Anyone knows where I can buy this shirt he's wearing?

[@chassecourt8824]

I have just been comparing the heat pipes from an Aorus 1080 Ti Extreme and an Aorus 4080 o/c I notice that for some reason the heatpipes on the 1080 seem to be much larger than those on the 4080 which seems counter intuitive.

[@Nepturion88]

Roman, I would like to see how much better the ROG Strix 4090 would be with better thermal pads. Seems like the Strix runs way hotter than other brands, I'm curious to why....

[@fliggopolis]

Any news on when the Wireview adapter is going to be available?

[@paco1669]

Thanks for this!

[@NarekAvetisyan]

As a 3D artist and a Blender user I'm super pissed with the removal of NVLink. This means no more memory pooling with 2 GPU's in 3D apps. It seems like the more money we give them the less features we get nowadays. Absolute greedy bastards!

[@rdmz135]

They want you to pay more for a Quadro.

[@kevinerbs2778]

@@rdmz135 the new Quadro Rtx6000 has not Nvlink. The older Quadro RTX A6000 had it. the difference is the A i nthe name of the card.

[@Xzonedude]

What brand calipers are those?? They’re so nice

[@rancid_]

Please test with better pads and then difference with a custom water-cooled loop (if possible) would be awesome. Seems the aio 4090s are about 5-6 degrees cooler.

[@xXDarthBagginsXx]

The girth of that cooler makes me glad i still have my CaseLabs M8 - finally a card that can fill that case out, lol