It is clear at this point that Founders Edition and Reference Design are two vastly different beasts entirely. Founders Edition exist to make Nvidia look good. Reference Design exists to give AIBs some kind of starting point to make an "MSRP" card that allows them a razor thin margin while being significantly worse than an FE card.
FE designs are so good ,why would you spend like $400 more on a card like strix which is like 1-2fps more and 2-3 degrees cooler! Imo more AIBs will stop producing nvidia gpus in the future like EVGA
@@TechnoFreak-IN THIS design is so good. The 3000 wasn't. The PCB itself was but the cooling was "crap". With 4090 this changes. Moreover FE is more sturdy than anything else. The only downside is when you want to replace fans. Complicated design.
Honestly we can probably blame Nvidia for the partner cards having worse VRM's. Partners have much slimmer margins to work with and that's mainly due to Nvidia charging way too much for the dies, and also telling partners to plan for 600W of cooling for a chip that ended up pulling 450W. Nvidia has much larger margins to work with.
@@bbag1550 Yet there are consumers still wiling to pay for Strix, that's why they continue to make them. For everyone else there TUF cards and other AIBs.
Nvidia does a relatively good job on the VRM, and partners can't compete, why is that Nvidia's fault? Nvidia is a company that wants to make money, and they indeed want to sell some FE cards this generation, I don't think that's their fault. Everything related to semiconductor has gotten much expensive since 2020, and that plays a part in the rising price Nvidia is asking as well.
Would love to see PCB breakdowns for the Aorus Master 4090 and the Asus Strix 4090. It’s really helpful when picking out a card to water cool. No one else has content like this.
afaik the I2C bus needs to be re-locked by the card each time it starts up. and you can bypass the lock by simply shorting the SDA and SCL pins when the card is booting up. Der8auer once did that on a RTX 2080 i belive. that way the I2C bus will stay unlocked after boot and you should be able to use an EVC2 with it.
I think looking at PCBs should be a priority from the point of view that we as viewers get vital information when making our choice in purchasing a 4090, since we're paying super premium prices to begin with we don't want to get ripped off!
I didn't feel the supply levels meant I had much option on choice. Founders? Impossible to buy new. Everything else? You buy whatever available. If you are lucky. It's a sad state where you can't choose your preferred card at £2000 ish
The 4090s also don't have massive transient spikes because the core just does not want to suck as much juice as the 30 series cards. Meaning they can make the power circuitry simplified. I have shunt modded my Palit RTX 4090 and it literally does not want to use much more than low 500Ws even in Timespy Extreme at 3000MHz+ and 1.1v.
Definitely interested in the Strix and Suprim which are already up on TPU as well as TUF, Master, Gaming OC and Trio. And a shoutout to what you think is the best reference PCB to put a water block on would also be nice. Is there any specific components of certain brands that are more likely to cause coil whine? I tested several Strix 3090s and all of them had horrible coil whine, some of them earlier models, some of them later revisions but all the same in that regard…
i watercooled a msi trio 3090 it's 33cm long !!! cutting my case in half and almost touches my reservoir, which made it a pain to cable, definitely check the PCBs or simply the waterblock size the two EKs on preorder tell FE block 236mm strix tuf 260mm, from what I've seen MSI and gigabyte seems to have longer pcbs..just remember to check before you order unlike me >
@@fredEVOIX I‘m fine with my Thermaltake The Tower 900, but I can imagine your pain. I think all PCBs should be a bit shorter this time since most cards have that flow-through design. Just a bit of variance and nothing like the MSI behemoths of last gen I guess
Do MSIs Suprim next please, I can't workout how their 26+4 phase vrm works with the 2891 chip. Also, the Suprim x and suprim liquid x have slightly different capacitance on the back.
@@crittybity I barely watched any PCB analysis other than GPUs in the past and I'm certain, I heard about doublers form there. The 16 stages is meant to be "up to 16 stages". Many more configuration is available under the max count.
@@crittybity you can just not use the whole 16...? I mean, if NVIDIA can get away with 10+3. then 13+2 looks doable. also, i don't see any doublers on MSI boards... maybe they just mashed 2 phases components together into 1 without doublers? that's doable, but ideally you want doublers.
The three shunt resistors on the power connector is to balance power draw between the connectors. The Palit RTX 4090 I have only has 2 shunts. This is similar to older Nvidia FE cards as well.
It's actually impressive how small the PCB is? You talked about unpopulated phases and how some 4090 models have them, wonder if the rumored 4090ti will also include those and mostly reuse the same PCB.
I can pretty much guarantee that's exactly what it will be. Same with the rumored Titan if they actually do it but I suspect that'll actually have NVLINK.
i have so been looking forward to these pcb breakdown videos from you!! I dont know what gpu im gonna get for ln2 this gen, normally i always get a kingpin.
Looks remarkably similar to the 3090Ti founders card. Almost to the point where I'd be curious if a 3090Ti FE waterblock would work on a 4090 FE. Will never get sick of how neat these dense little cards are.
which 4090 PCB would you say is the best 4090? strix or suprim? i think the suprim has more robust hardware, but i don't know if the extra spacing between the core and VRM will have any effect on the card. thanks
What was the middle bit again? I know you were speaking English, but yeah. LOL Thank you Sir, as Dirty Harry would put it 'A Man has to know his limitations', and you went past mine in the first few seconds. Have a great weekend.
The PCB itself is a smaller part of the total cost, there's the heatsink and R&D and marketing and all of that as factors but 24GBs of GDDR6x memory alone is a few hundred, the GPU core is also extremely expensive to manufacture given it's TSMC 4nm and 608mm² so we're talking another few hundred to the total and with the giant heatsink cost i'd say we're talking somewhere between 600 and 1k usd in costs alone without going deeper on the math as i can't find a good source on TSMC 4nm wafer costs or 24Gbps GDDR6x cost per GB.
So.. is the asus tuf oc 4090 better than this? Is it worth an additional 300€ over the fe? The tuf is actually sometimes available whereas the fe isn’t so far.
G'day Buildzoid, You did Doodles on the photos, plus said "Doodoo" & "You Pee", just need to work in Boobies🤭 But in all seriousness with new AMD & Intel chipsets + AMD, Intel & nVIDIA GPUs all before Christmas you are going to be Super Busy with PCB Breakdowns, So in advance a MASSIVE THANK YOU for all the work you put in, it is so important for those looking to make a purchase & really appreciated 🥰
At TechPowerUp the Suprim X uses 17 more watts in idle and 22 more watts in video playback - i wonder why? Did they already planned the PCB for a 600 watt GPU - is that why?
as an AMD nutter, I'm 99.99% you see this PCB as full blown RTX 4090Ti coolers are ready and now it is just about binning and need for Nvidia to launch it
Any idea why they'd depopulate the phases closer to the power connectors? Seems like you'd have lower power plane losses depopulating the other side. Does this have less voltage drop right-to left in the die? Or maybe it's a cooling thing, wanting to move a few Watts of waste heat to the case exhaust?
I've spent some time going through the PCBs on the 4090. If I were to make a hierarchy: Zotac -> Gigabyte -> FE -> TUF -> MSI Gaming -> Strix -> MSI Suprim. You have to go to the FE and above to get 70A power stages. The Suprim X has 2 more stages than the Strix does. Zotac basically took the reference design even for their Amp Extreme. But then it's like "does it really matter?" when the 4090 isn't anywhere near as much of a power-slut as the 3090. The MSI cards don't have 600W bios's but it's probably going to be flashable in a week.
@@user-vx4nz8lg2i The TUF uses the same PCB as the Strix but with some power stages lopped off. I haven't seen whether it uses the same 70A power stages as the strix but...probably. I'd argue the FE and TUF are pretty comparable but the TUF is going to have better waterblock availability if that's what you're into, the FE is going to be one of the better air coolers. I wouldn't feel bad if I couldn't get the FE and had to get the TUF instead. But whatever you do...don't pay extra for the OC TUF. The Strix is just stupidly overpriced.
I honestly haven't listened carefully but from the various reviews and derbauer I seem to have understood that Nvidia more than previous gens limited the AIB versions making them irrelevant what I personnally learned with a 3090, check the PCB length before you order or buy anything some AIB have +10cm PCBs from FE for no good reasons super annoying when you watercool and it takes the same size as a 4090 strix >
Thinking about it, you haven't done official teardowns for gn for a while now right? Or I'm missing something, or there haven't been any good anything been released. Idk 🙃
The 4090's don't seem to have the large power spikes the 3090 did but I've also seen the peaks while not nearly as high do last longer .... I wonder if that was the reason for going over to the MP2891 to tame those spikes in current draw ..... It'll be interesting to see some power draw tests on AIB boards to see if they do the same or if they have the same spiky power draw as the 3090
Has anyone spotted a marking for the Layerstackup? Or seen a number of how many Layers the PCB has? Didnt see them during the vid if they are even on there. Love your Vids, way over my head in technicality but fan to listen to.
What performance difference do you get with a "good" Voltage controller, compared to a mediocre voltage controller? Why would you want good voltage regulation?
Since all 24GB of VRAM is on one side of the PCB, I wonder how feasible it would be to have the VRAM and VRMs on opposite sides of the PCB so the VRMs can wrap round the GPU. The way I'm imagining it, you'd lose the option of cooling those components from the underside, though.
AIBs suck! When there had been the initial "black-out / reboot" problem with 3000 series, Asus chose to go the all-MLCC-behind-the-core way. My STRIX 3090 has this 6x MLCC groups. As soon as the "problem" had been resolved (by drivers?), they've been placing only 2x MLCC groups + 4x poly.
Can you please review the PCB of Zotac 4090 amp extreme airo version? I want to know how their implementation is for 40 series compared to their 30 series cards.
would the pwm logic input for the mp2891 be 3.3v or 5v, making a analog dc potienttiometer to pwm converter would be simple using a raspberry pi pico or Arduino, or even just setup a web ui on the micro controller and set it over wifi and or have buttons attached to adjust it, you could have it read the stock input and apply a offset it if you wanted to
Imagine if the HOF could reliably push a kiloamp into the core... we are getting into territory where the literal solder joints and circuit board would melt down before the VRM or the GPU core does.
Are the ASUS Strix 4090s okay? If I remember correctly they didn’t have any transient power spike issues like some other models back in the 3090 launch days. (Would only be looking at models with 2 or more HDMI 2.1 ports)
@@luckerNo01 When you actually “need” fully-featured HDMI 2.1 to drive displays/TVs the GPU situation has been sucking for two years. Both AMD and NVIDIA haven’t released “good” dGPUs for 2D work that stay within a 75 W TGP, the onboard “HDMI 2.1” ports on motherboards are basically false advertising and while theoretically possible active DisplayPort 1.4-to-HDMI 2.1 adapters don’t work stable (but VRR can’t work there), meaning even if you have a dGPU with a real HDMI 2.1 port you better treat it like a baby to not lose its functionality to wear and tear, with two HDMI 2.1 ports like the ASUS or some 30 series Gigabyte models you at least have a “spare”. :(
Yeah, it's Palit. 16 (or teamed 8, not sure) phase Vcore with 50A power stages. Filtering mostly with through hole cans. Looks sad for a $1.7k top end card.
Thanks Buildzoid. But since the new 4090s don't have the VRMs on the back this generation, does it make sense to water cool the back with an active backplate?
@@DDD-xx4mg You sure are thrifty with other people's money. So what's next? Buying EK's gold plated CPU blocks just because you can afford a regular block instead of an aircooler? 🤡
@@generalgrevous1984 if you have 2 grand to spend on a gpu just to play a few video games you not going to care about and extra 150 to make you water cooling build look more pleasing. The over engineered 4090s don’t require water cooling there running manual overclock at 3Ghz around 50-60c. None of them are hitting power limits core clocks are not throttling. There hitting voltage limits and voltages can’t be increased without mods. So water cooling them is just for aesthetics as they’ll be little to none performance difference
We’re buying to upgrade 2 3090’s in a 3D workstation. We plan on buying two 4090’s to start and then adding more later if required (some of our projects take a week to render). Which 4090 would you recommend we commit to standardise our build on? Once we are past 2x 4090 our plan is to water cool otherwise it won’t fit our WRX80 build. I’m leaning towards 4090 FE’s. Thanks in advance!
Honestly yeah the FE is actually pretty good. Look up GamersNexus review of that card lots of information in there. One issue I see is how are you going to power them? Just one humongous PSU or dual PSU? I assume server motherboard and server case with built in monster PSUs right?
how else do you think they could connect all the necessary signals and power to the core? This is standard for GPUs. It's just easier to see here because they are back drilling some of the memory traces.
@@ActuallyHardcoreOverclocking oh, I never saw that. To me looks like the substrate around the chip will brake in any moment if the cooler will be on his place by detach because of that amount of vias.
God the pcb look funky xD why don't they just make a AIO as a heat sink like ====== then fans and make a low profile pump but they make it weirdly shape just to get more air pass air cooled fins
If only it was possible to actually buy this card. Seriously I hate NVIDIA for just about every reason. They announce a product, they hype it up, but then make it virtually impossible to buy that particular variant of it.
Bill can you please begin a close enough BOM cost assessment for the board itself moving to the subsystem, I need to check my own work both on a component + mechanical design basis and production economic method. I understand the dGPU cost : price / margin so I will trade this with you for the other data. mb
On through distribution and manufacturer discrete component price just wing it or I can give the models to differentiate the delta between high and low volume. mb
PCBs are getting tiny compared to the complete card... Just found one of my older cards in the attic, Asus GTX285 (PCB size worthy of a high-end card) witch was written off because of non apparent overheating issues (presumably). If you fancy it... I could mail it to you.
It is clear at this point that Founders Edition and Reference Design are two vastly different beasts entirely. Founders Edition exist to make Nvidia look good. Reference Design exists to give AIBs some kind of starting point to make an "MSRP" card that allows them a razor thin margin while being significantly worse than an FE card.
yup, one of the biggest reason why EVGA stopped!
FE designs are so good ,why would you spend like $400 more on a card like strix which is like 1-2fps more and 2-3 degrees cooler! Imo more AIBs will stop producing nvidia gpus in the future like EVGA
@@TechnoFreak-IN THIS design is so good. The 3000 wasn't. The PCB itself was but the cooling was "crap". With 4090 this changes. Moreover FE is more sturdy than anything else. The only downside is when you want to replace fans. Complicated design.
Founders Edition and Reference never been the same since 20 series.
@@DavidRain007 i kid you not 3080 VRM quality was better than some highend cards like FTW3.
Honestly we can probably blame Nvidia for the partner cards having worse VRM's. Partners have much slimmer margins to work with and that's mainly due to Nvidia charging way too much for the dies, and also telling partners to plan for 600W of cooling for a chip that ended up pulling 450W.
Nvidia has much larger margins to work with.
EVGA AGREE!
There's also the issue of time. nVidia can start PCB design work early, but doles out info to partners as late as possible.
Strix and suprim put out great pcbs this generation too but only strix ended up grossly overpriced
@@bbag1550 Yet there are consumers still wiling to pay for Strix, that's why they continue to make them. For everyone else there TUF cards and other AIBs.
Nvidia does a relatively good job on the VRM, and partners can't compete, why is that Nvidia's fault? Nvidia is a company that wants to make money, and they indeed want to sell some FE cards this generation, I don't think that's their fault. Everything related to semiconductor has gotten much expensive since 2020, and that plays a part in the rising price Nvidia is asking as well.
Would love to see PCB breakdowns for the Aorus Master 4090 and the Asus Strix 4090. It’s really helpful when picking out a card to water cool. No one else has content like this.
Get the FE. No point spending $2000+. FE is better than Strix at overclocking too.
@@Kizzster That a fact? If so, is there a video you can link to that?
Asus is nowhere near as good as it used to be up until couple years ago. That goes for their motherboards too.
And the FTW4.....oh wait.
afaik the I2C bus needs to be re-locked by the card each time it starts up. and you can bypass the lock by simply shorting the SDA and SCL pins when the card is booting up. Der8auer once did that on a RTX 2080 i belive. that way the I2C bus will stay unlocked after boot and you should be able to use an EVC2 with it.
I think looking at PCBs should be a priority from the point of view that we as viewers get vital information when making our choice in purchasing a 4090, since we're paying super premium prices to begin with we don't want to get ripped off!
THIS!!! ☝️☝️☝️
I didn't feel the supply levels meant I had much option on choice. Founders? Impossible to buy new. Everything else? You buy whatever available. If you are lucky. It's a sad state where you can't choose your preferred card at £2000 ish
@@stevenross-watt8640 Exactly.
The 4090s also don't have massive transient spikes because the core just does not want to suck as much juice as the 30 series cards. Meaning they can make the power circuitry simplified. I have shunt modded my Palit RTX 4090 and it literally does not want to use much more than low 500Ws even in Timespy Extreme at 3000MHz+ and 1.1v.
Tsmc work wonders
Bro when's your ranking 4090's coming out
i love your videos :)
Definitely interested in the Strix and Suprim which are already up on TPU as well as TUF, Master, Gaming OC and Trio. And a shoutout to what you think is the best reference PCB to put a water block on would also be nice. Is there any specific components of certain brands that are more likely to cause coil whine? I tested several Strix 3090s and all of them had horrible coil whine, some of them earlier models, some of them later revisions but all the same in that regard…
i watercooled a msi trio 3090 it's 33cm long !!! cutting my case in half and almost touches my reservoir, which made it a pain to cable, definitely check the PCBs or simply the waterblock size the two EKs on preorder tell FE block 236mm strix tuf 260mm, from what I've seen MSI and gigabyte seems to have longer pcbs..just remember to check before you order unlike me >
@@fredEVOIX
I‘m fine with my Thermaltake The Tower 900, but I can imagine your pain. I think all PCBs should be a bit shorter this time since most cards have that flow-through design. Just a bit of variance and nothing like the MSI behemoths of last gen I guess
Do MSIs Suprim next please, I can't workout how their 26+4 phase vrm works with the 2891 chip. Also, the Suprim x and suprim liquid x have slightly different capacitance on the back.
13+2 with doublers seems plausible.
@@menpee Do GPUs use doublers? I thought they were teamed starges? So 26+4 would be 13+2, which is less than 16
@@crittybity I barely watched any PCB analysis other than GPUs in the past and I'm certain, I heard about doublers form there.
The 16 stages is meant to be "up to 16 stages". Many more configuration is available under the max count.
@@crittybity you can just not use the whole 16...? I mean, if NVIDIA can get away with 10+3. then 13+2 looks doable. also, i don't see any doublers on MSI boards... maybe they just mashed 2 phases components together into 1 without doublers? that's doable, but ideally you want doublers.
Even if I don't understand most of what you are saying I still enjoy listening to whatever you are saying. I always learn something new😀
The three shunt resistors on the power connector is to balance power draw between the connectors. The Palit RTX 4090 I have only has 2 shunts. This is similar to older Nvidia FE cards as well.
Could you post the tier list for the 4090s please (:
It's actually impressive how small the PCB is? You talked about unpopulated phases and how some 4090 models have them, wonder if the rumored 4090ti will also include those and mostly reuse the same PCB.
I can pretty much guarantee that's exactly what it will be. Same with the rumored Titan if they actually do it but I suspect that'll actually have NVLINK.
@@JJFX- Titan is coming back???
i have so been looking forward to these pcb breakdown videos from you!! I dont know what gpu im gonna get for ln2 this gen, normally i always get a kingpin.
2 of those shunts are in series -> increasing resistance to measure the lower power of the memory supply without using different resistor types.
Looks remarkably similar to the 3090Ti founders card. Almost to the point where I'd be curious if a 3090Ti FE waterblock would work on a 4090 FE. Will never get sick of how neat these dense little cards are.
It does look that way, and there are a few bykski 4090 FE blocks on ali that seem to list 3090ti support so I do wonder
Thank you!
I miss those PCB analysis!
Can you do that with all the other partner cards? Please
which 4090 PCB would you say is the best 4090? strix or suprim? i think the suprim has more robust hardware, but i don't know if the extra spacing between the core and VRM will have any effect on the card. thanks
So what cards should we avoid in your opinion?
Best channel on UA-cam.
I can confirm that it is not THE best channel on UA-cam. Vsauce is the actual best channel on UA-cam, no debate
@@racecarrik 🤣🤣🤣🤣🤣
Such a nice little board.
What was the middle bit again? I know you were speaking English, but yeah. LOL Thank you Sir, as Dirty Harry would put it 'A Man has to know his limitations', and you went past mine in the first few seconds. Have a great weekend.
Would love to see cost analysis if possible.
The PCB itself is a smaller part of the total cost, there's the heatsink and R&D and marketing and all of that as factors but 24GBs of GDDR6x memory alone is a few hundred, the GPU core is also extremely expensive to manufacture given it's TSMC 4nm and 608mm² so we're talking another few hundred to the total and with the giant heatsink cost i'd say we're talking somewhere between 600 and 1k usd in costs alone without going deeper on the math as i can't find a good source on TSMC 4nm wafer costs or 24Gbps GDDR6x cost per GB.
ASUS TUF next please! J C has a high res breakdown for reference.
So.. is the asus tuf oc 4090 better than this? Is it worth an additional 300€ over the fe? The tuf is actually sometimes available whereas the fe isn’t so far.
G'day Buildzoid,
You did Doodles on the photos, plus said "Doodoo" & "You Pee", just need to work in Boobies🤭
But in all seriousness with new AMD & Intel chipsets + AMD, Intel & nVIDIA GPUs all before Christmas you are going to be Super Busy with PCB Breakdowns,
So in advance a MASSIVE THANK YOU for all the work you put in, it is so important for those looking to make a purchase & really appreciated 🥰
ROG RTX 4090 Strix next please 🙏
Edited: We need to know what's your best brand of RTX 40 - Thanks!
At TechPowerUp the Suprim X uses 17 more watts in idle and 22 more watts in video playback - i wonder why? Did they already planned the PCB for a 600 watt GPU - is that why?
Thanks for the video!!!
thanks for the video! I would be also interesting to analyse some of the cheaper 4090s e.g. inno3d. Does a reduction in price imply lower quality?
I don't want a 4090 or a 7900 XTX, I want a 9800 Pro... the new one in a year. That card back in the day dominated. Nothing close.
Seems like there is no point in buying third party 4090. Only out of stock FE can force people to pay +400$ for RGB puke cards.
With my eyes closed it sounds like Kermit is presenting this video....
Can we talk a bit about warranty here? Zotac offers 5 year warranty. Do you recommend buying their 4090s?
Please do the strix 4090 pcb!!!
Thx a lot for this Video!
i saw today that Nvidia cancelled the 4080 12gb...i guess public pressure does get results
That's not the whole story. It also means the 4080 12GB won't compete with their own 3090/3090 Ti cards they are still trying to unload.
@@brego129 yeah i see on newegg they got a ton of 3000 series cards still in stock
as an AMD nutter, I'm 99.99% you see this PCB as full blown RTX 4090Ti
coolers are ready and now it is just about binning and need for Nvidia to launch it
Any idea why they'd depopulate the phases closer to the power connectors? Seems like you'd have lower power plane losses depopulating the other side. Does this have less voltage drop right-to left in the die? Or maybe it's a cooling thing, wanting to move a few Watts of waste heat to the case exhaust?
The 12V power distribution is pretty low current do the distance doesn't matter that much
@@ActuallyHardcoreOverclocking can you do TUF next?:-)
But BZ, this is a QUAL SAMPLE, this isn't representative of the final card /s
That is a beautifully dense PCB
Almost as dense as the nvidia exec who tried to sell the 4080 12gb
Do you think theyre saving the unpopulated spots for a full-die 4090 Ti?
Can you please do the strix 4090? techpowerup have some good images you might be able to use
I've spent some time going through the PCBs on the 4090. If I were to make a hierarchy: Zotac -> Gigabyte -> FE -> TUF -> MSI Gaming -> Strix -> MSI Suprim. You have to go to the FE and above to get 70A power stages. The Suprim X has 2 more stages than the Strix does. Zotac basically took the reference design even for their Amp Extreme. But then it's like "does it really matter?" when the 4090 isn't anywhere near as much of a power-slut as the 3090. The MSI cards don't have 600W bios's but it's probably going to be flashable in a week.
So is the tuf better then founders pcb with only 18 power phases vs 20 on founders?? also is the tuf 70A?
@@user-vx4nz8lg2i The TUF uses the same PCB as the Strix but with some power stages lopped off. I haven't seen whether it uses the same 70A power stages as the strix but...probably. I'd argue the FE and TUF are pretty comparable but the TUF is going to have better waterblock availability if that's what you're into, the FE is going to be one of the better air coolers. I wouldn't feel bad if I couldn't get the FE and had to get the TUF instead. But whatever you do...don't pay extra for the OC TUF. The Strix is just stupidly overpriced.
The tuf have only 18 power phases, strix 26 phases.. so is the TUF a worse pcb then founders???
I honestly haven't listened carefully but from the various reviews and derbauer I seem to have understood that Nvidia more than previous gens limited the AIB versions making them irrelevant
what I personnally learned with a 3090, check the PCB length before you order or buy anything some AIB have +10cm PCBs from FE for no good reasons super annoying when you watercool and it takes the same size as a 4090 strix >
Which PCBs are good and which bad?
Don’t need details, just since planning to get a GPU soon, atm mostly considering ASUS due to HDMI 2.1s
What do you think about the 4090 msi trio?
Thinking about it, you haven't done official teardowns for gn for a while now right? Or I'm missing something, or there haven't been any good anything been released. Idk 🙃
The 4090's don't seem to have the large power spikes the 3090 did but I've also seen the peaks while not nearly as high do last longer .... I wonder if that was the reason for going over to the MP2891 to tame those spikes in current draw ..... It'll be interesting to see some power draw tests on AIB boards to see if they do the same or if they have the same spiky power draw as the 3090
Has anyone spotted a marking for the Layerstackup? Or seen a number of how many Layers the PCB has? Didnt see them during the vid if they are even on there.
Love your Vids, way over my head in technicality but fan to listen to.
isn't the 12vhpwr supposed replace 4 8 pins?since each 8 pin is supposed to carry 150W and the 12vhpwr is "rated" for 600?
What performance difference do you get with a "good" Voltage controller, compared to a mediocre voltage controller? Why would you want good voltage regulation?
Since all 24GB of VRAM is on one side of the PCB, I wonder how feasible it would be to have the VRAM and VRMs on opposite sides of the PCB so the VRMs can wrap round the GPU. The way I'm imagining it, you'd lose the option of cooling those components from the underside, though.
Interesting idea, maybe with an efficient VRM a finned backplate could be sufficient for cooling it
Interested in the Zotac one!
Me too (the 4090 AMP Extreme).
AIBs suck! When there had been the initial "black-out / reboot" problem with 3000 series, Asus chose to go the all-MLCC-behind-the-core way. My STRIX 3090 has this 6x MLCC groups. As soon as the "problem" had been resolved (by drivers?), they've been placing only 2x MLCC groups + 4x poly.
In an abstract way, the 4090 founders PCB looks like Godzilla's head with his mouth open.
So which 4090 is the KING of performance/thermals???
Asus? MSI? Galax? Gigabyte? Etc.
Can you please review the PCB of Zotac 4090 amp extreme airo version? I want to know how their implementation is for 40 series compared to their 30 series cards.
Nvidia skimped out on the DP2.0 connection, of course they dont fill out the VRM lol or cheaped out on other things
Looks like Palit decided to run with 4 less power stages than the FE for its 4090 Gamerock. Should potential Palit buyers be concerned?
Let’s go I love these and I don’t know why
FE is the best. But why the voltage limits?
Thx !!!
Nice analysis. But some of it can be a bit too nitpicky.
would the pwm logic input for the mp2891 be 3.3v or 5v, making a analog dc potienttiometer to pwm converter would be simple using a raspberry pi pico or Arduino, or even just setup a web ui on the micro controller and set it over wifi and or have buttons attached to adjust it, you could have it read the stock input and apply a offset it if you wanted to
Imagine if the HOF could reliably push a kiloamp into the core... we are getting into territory where the literal solder joints and circuit board would melt down before the VRM or the GPU core does.
Are the ASUS Strix 4090s okay? If I remember correctly they didn’t have any transient power spike issues like some other models back in the 3090 launch days.
(Would only be looking at models with 2 or more HDMI 2.1 ports)
AFAIK only the asus cards have 2 hdmi ports. So you’re limited to them - they’ll be okay I’m sure, but worth additional hundreds of €/$? 🤷♂️
@@luckerNo01 When you actually “need” fully-featured HDMI 2.1 to drive displays/TVs the GPU situation has been sucking for two years. Both AMD and NVIDIA haven’t released “good” dGPUs for 2D work that stay within a 75 W TGP, the onboard “HDMI 2.1” ports on motherboards are basically false advertising and while theoretically possible active DisplayPort 1.4-to-HDMI 2.1 adapters don’t work stable (but VRR can’t work there), meaning even if you have a dGPU with a real HDMI 2.1 port you better treat it like a baby to not lose its functionality to wear and tear, with two HDMI 2.1 ports like the ASUS or some 30 series Gigabyte models you at least have a “spare”. :(
The Zotac/Gigabyte $1700 cards...
Still need to see a TUF, but so far all the MSRP-ish cards are real bad looking.
I don't understand much, but I hate myself to a similar degree that Buildzoid hates doing UA-cam videos, so I suffered through the whole 39 minutes. 💋
Which manufacturers have worse Pcb's?
Probably Palit
@@HamedEmine Palit had very good PCBs last gen though.
Yeah, it's Palit. 16 (or teamed 8, not sure) phase Vcore with 50A power stages. Filtering mostly with through hole cans. Looks sad for a $1.7k top end card.
I'll say likely Zotac or Galax
Question.
I see people clock higher on FE than on other brands, even with better coolers.
Is this because Nvidia uses better VRAM?
Binned cores...
whats interesting is the spaces on the PCB for additional components that are not present, why do manufacturers do that?
For a 4090 Ti.
I did see some of the cheaper brands that hand out just 3 (max 450w) power cable has a lot less components.
Waiting for the supreme x and rog strix
i like the way you say anyway... :)
Suprim next please.
Pls do the Strix and the Suprim 4090s.
Thanks Buildzoid. But since the new 4090s don't have the VRMs on the back this generation, does it make sense to water cool the back with an active backplate?
I find the active backplate is athletically pleasing wether needed or not
@@DDD-xx4mg That it is but for double the price.
@@generalgrevous1984 when you spending 4090 💰 💰 💰 what another €150 😅
@@DDD-xx4mg You sure are thrifty with other people's money. So what's next? Buying EK's gold plated CPU blocks just because you can afford a regular block instead of an aircooler? 🤡
@@generalgrevous1984 if you have 2 grand to spend on a gpu just to play a few video games you not going to care about and extra 150 to make you water cooling build look more pleasing. The over engineered 4090s don’t require water cooling there running manual overclock at 3Ghz around 50-60c. None of them are hitting power limits core clocks are not throttling. There hitting voltage limits and voltages can’t be increased without mods. So water cooling them is just for aesthetics as they’ll be little to none performance difference
what about MSI 4090 Suprim X?
We’re buying to upgrade 2 3090’s in a 3D workstation. We plan on buying two 4090’s to start and then adding more later if required (some of our projects take a week to render).
Which 4090 would you recommend we commit to standardise our build on?
Once we are past 2x 4090 our plan is to water cool otherwise it won’t fit our WRX80 build.
I’m leaning towards 4090 FE’s.
Thanks in advance!
Honestly yeah the FE is actually pretty good. Look up GamersNexus review of that card lots of information in there.
One issue I see is how are you going to power them? Just one humongous PSU or dual PSU?
I assume server motherboard and server case with built in monster PSUs right?
@@alouisschafer7212 FE will be shit in dual GPU setup because the bottom card exhausts it's heat into the top card.
@@Wasmachineman that's why I am asking Server case yes or no?
Buildzoid 31:48 Is that not dangerous to put that amount of vias around the GPU's chip?
how else do you think they could connect all the necessary signals and power to the core? This is standard for GPUs. It's just easier to see here because they are back drilling some of the memory traces.
@@ActuallyHardcoreOverclocking oh, I never saw that. To me looks like the substrate around the chip will brake in any moment if the cooler will be on his place by detach because of that amount of vias.
Can you cover the strix?
God the pcb look funky xD why don't they just make a AIO as a heat sink like ====== then fans and make a low profile pump but they make it weirdly shape just to get more air pass air cooled fins
i want to see the 4090 from asus (PCB)
YEEES
Nice!
If only it was possible to actually buy this card. Seriously I hate NVIDIA for just about every reason. They announce a product, they hype it up, but then make it virtually impossible to buy that particular variant of it.
Bill can you please begin a close enough BOM cost assessment for the board itself moving to the subsystem, I need to check my own work both on a component + mechanical design basis and production economic method. I understand the dGPU cost : price / margin so I will trade this with you for the other data. mb
On through distribution and manufacturer discrete component price just wing it or I can give the models to differentiate the delta between high and low volume. mb
Take a shot every time BZ says "like" :p
What is a Hall of Fame card?
I think the Palit Gamerock 4090 has worse pcb design than FE.
What about calling surface mount can caps axial caps
Can you connect a EVC2 to a FE?
no memory on the back to overheat this time? great
Next Galax sg rtx4090 pcb breakdown please. 🙏🏻
ASUS Strix and Galax HOF when?
we need to know the bad ones!
I want an "anyway" counter in your videos pls
PCBs are getting tiny compared to the complete card...
Just found one of my older cards in the attic, Asus GTX285 (PCB size worthy of a high-end card) witch was written off because of non apparent overheating issues (presumably).
If you fancy it... I could mail it to you.
5:22
Haha, he said "doodoo"
please do the aorus 4090.