GPU PCB Analysis: Founders Edition RTX 3080 Build Quality & VRM Capabilities
video description
Date: 2020-09-27
Related videos
Comments and reviews: 10
SuperG316
Most Obvious thing to know if someone know tech is about wich traces have the highest priority.
This layout is made by PCB desgn by expert above reference design. The developer of these board are top designers. So to question them on the why make no sense.
The first thing of importance are the high priority traces. And that are the GDDR6X memchips traces to the GPU wich needs to be lenght match. That why Ram chips are closest to the GPU. Also the gpu bga design pin ( balls ) might be ideal placed on both outer sides of the chips to support easier or optimised routing and would be special multi layer PCB with power delevery layer but also match deilectrium so they can take parasitair capacitance between layer into account. I expect mayby multi power layer with thick cuper. But to know that you need to have cross section of the PCB, wich destroy it. Also is the placement of the parts of each logic VRM and Buck spread over to both side this means that the power trace and circuit trace length between part are very short. It make sense that these PCB are very expensive to develop and produce and SMT to.
Also to put the high frequency smallest C close to the gpu pins the bypas C complexes are place on the other side of the PCB to the BGA chip.
Also the 500K switching is trade of the higher switch freq the smaller capacitance can be to point where efficiency du to switching espacialy by high loads. for low loads 5Mhz might be enough but keep the parts smaller. L stores enegy magneticly C electricly at 5Mhz it need buffer it for 1/10 time of what 500Khz need to do for switching riple.
It example how expert design is done.
reply
Most Obvious thing to know if someone know tech is about wich traces have the highest priority.
This layout is made by PCB desgn by expert above reference design. The developer of these board are top designers. So to question them on the why make no sense.
The first thing of importance are the high priority traces. And that are the GDDR6X memchips traces to the GPU wich needs to be lenght match. That why Ram chips are closest to the GPU. Also the gpu bga design pin ( balls ) might be ideal placed on both outer sides of the chips to support easier or optimised routing and would be special multi layer PCB with power delevery layer but also match deilectrium so they can take parasitair capacitance between layer into account. I expect mayby multi power layer with thick cuper. But to know that you need to have cross section of the PCB, wich destroy it. Also is the placement of the parts of each logic VRM and Buck spread over to both side this means that the power trace and circuit trace length between part are very short. It make sense that these PCB are very expensive to develop and produce and SMT to.
Also to put the high frequency smallest C close to the gpu pins the bypas C complexes are place on the other side of the PCB to the BGA chip.
Also the 500K switching is trade of the higher switch freq the smaller capacitance can be to point where efficiency du to switching espacialy by high loads. for low loads 5Mhz might be enough but keep the parts smaller. L stores enegy magneticly C electricly at 5Mhz it need buffer it for 1/10 time of what 500Khz need to do for switching riple.
It example how expert design is done.
reply
itech
Thanks for the video, but this is so sad, these companies getting cheap and treating their customers like an idiot, we as consumers should never bother understanding this, we should pay and get the product we won't,
We shouldn't have to worry and get PhD in every single side of our life, we shouldn't have to get a PhD in monitor science and make months of research to find a monitor without a dead pixel, BLB, flicker or any other artifacts.
We shouldn't have to spend time and learn PCB to get a graphic card we buy so we don't fall to the trap of getting a faulty graphic card.
These companies need to be sued, they are using cheap component to maximize their profit and they are not spending time and money on QA department, how hard is to hire 10 guys to play 10 games for 10 days 10 hours a day to find this issue before selling thousands of units to the mass market??
I wouldn't be surprised if the difference between the cheap and the high-quality components for these 6 capacitors is less than 5 each.
reply
Thanks for the video, but this is so sad, these companies getting cheap and treating their customers like an idiot, we as consumers should never bother understanding this, we should pay and get the product we won't,
We shouldn't have to worry and get PhD in every single side of our life, we shouldn't have to get a PhD in monitor science and make months of research to find a monitor without a dead pixel, BLB, flicker or any other artifacts.
We shouldn't have to spend time and learn PCB to get a graphic card we buy so we don't fall to the trap of getting a faulty graphic card.
These companies need to be sued, they are using cheap component to maximize their profit and they are not spending time and money on QA department, how hard is to hire 10 guys to play 10 games for 10 days 10 hours a day to find this issue before selling thousands of units to the mass market??
I wouldn't be surprised if the difference between the cheap and the high-quality components for these 6 capacitors is less than 5 each.
reply
Icarus
For the video, it was interesting to hear about your findings and seeing what the PCB quality is was cool - but I guess I am a bit too uninitiated in the tech sphere since most of what you said flew over my head. Guess I will be doing some side research along with figuring out which major I want to take (looking at computer engineering and physics - but I definitely will need to know what these terms mean).
On a side note that isn't related to the video subject at all, there seems to be leaks of RDNA 2 having base clocks in the 2 ghz - 2.5 ghz range (the 2.5 is supposedly on a 40 CU card, which is crazy). What are your thoughts on that? (this is to anyone who sees this comment)
reply
For the video, it was interesting to hear about your findings and seeing what the PCB quality is was cool - but I guess I am a bit too uninitiated in the tech sphere since most of what you said flew over my head. Guess I will be doing some side research along with figuring out which major I want to take (looking at computer engineering and physics - but I definitely will need to know what these terms mean).
On a side note that isn't related to the video subject at all, there seems to be leaks of RDNA 2 having base clocks in the 2 ghz - 2.5 ghz range (the 2.5 is supposedly on a 40 CU card, which is crazy). What are your thoughts on that? (this is to anyone who sees this comment)
reply
Federico
Thank you very much for these reviews.
You have already reviewed Zotac and MSI.
And according to you this is the best PCB for the 3080 video chip. And by far, since the other manufacturers tend to use cheaper components (capacitors, voltage controllers etc...) in order to save money and make profits. MSI is a bit better than Zotac but anyway not very impressive too.
Now, my question is: how can it be that the best card is also the cheapest?
How can it be that Nvidia uses the highest quality components and sells the card for 200 less money?
Does this mean that most of the production costs are due to the heat sink and aesthetic features? What do you think?
reply
Thank you very much for these reviews.
You have already reviewed Zotac and MSI.
And according to you this is the best PCB for the 3080 video chip. And by far, since the other manufacturers tend to use cheaper components (capacitors, voltage controllers etc...) in order to save money and make profits. MSI is a bit better than Zotac but anyway not very impressive too.
Now, my question is: how can it be that the best card is also the cheapest?
How can it be that Nvidia uses the highest quality components and sells the card for 200 less money?
Does this mean that most of the production costs are due to the heat sink and aesthetic features? What do you think?
reply
Prisoner
it would be nice since these cards are 700 and 1500 if you could rate the cards from worst to best when you have opportunity to get the PCBs for the 3080 and esp 3090 1500+ card from the AIBs or direct us to somewhere that does tell us which AIBs skimped on the cards and which cards are really for the enthusiasts who want to overclock the cards. AIBs who skimped on the card should not be rewarded by enthusiasts (usually the first adopters) by overpaying for a card that is marketed as OC and has to be unclocked and/or undervolted to just run.
reply
it would be nice since these cards are 700 and 1500 if you could rate the cards from worst to best when you have opportunity to get the PCBs for the 3080 and esp 3090 1500+ card from the AIBs or direct us to somewhere that does tell us which AIBs skimped on the cards and which cards are really for the enthusiasts who want to overclock the cards. AIBs who skimped on the card should not be rewarded by enthusiasts (usually the first adopters) by overpaying for a card that is marketed as OC and has to be unclocked and/or undervolted to just run.
reply
Mike
Profit maximized and cost optimized are slightly different. Profit maximized is a function of increasing margin reducing cost on economies of production scale for every next unit produced. Think volume discounts on input costs. Or, producing more in less time. Cost optimization reduces cost on the overall, typically by reducing some variable cost manpower is a typical cost center, subsystem component cost in your example, to increase net profit although taking into account (despite) the potential of encountering some cost / quality offset. mb
reply
Profit maximized and cost optimized are slightly different. Profit maximized is a function of increasing margin reducing cost on economies of production scale for every next unit produced. Think volume discounts on input costs. Or, producing more in less time. Cost optimization reduces cost on the overall, typically by reducing some variable cost manpower is a typical cost center, subsystem component cost in your example, to increase net profit although taking into account (despite) the potential of encountering some cost / quality offset. mb
reply
TVRB
I bet that they are using the same PCB with all of the cards: 12 memory chip lots with 2 unpopulated and all of the chips being 1GB would make the 10GB RTX 3080 but all of them populated with 2GB chips and with the missing LR22 's will make the 24GB RTX 3090. I wouldn't be surprised if the RTX 3070 is the same PCB with 4 unpopulated memory chip lots and at least one less LR22
Edit: Just had a look at the RTX 3090 teardown, that may just be the case.
reply
I bet that they are using the same PCB with all of the cards: 12 memory chip lots with 2 unpopulated and all of the chips being 1GB would make the 10GB RTX 3080 but all of them populated with 2GB chips and with the missing LR22 's will make the 24GB RTX 3090. I wouldn't be surprised if the RTX 3070 is the same PCB with 4 unpopulated memory chip lots and at least one less LR22
Edit: Just had a look at the RTX 3090 teardown, that may just be the case.
reply
zengrath
This is all great. I appreciate all this info. , i'm in the waiting camp myself to get 3080. i appreciate all info. My only question is when things settle down what sources will i have to study the PCB quality of various manufactures so i know when i do buy one i'm getting the best quality card? Will Gamers nexus do some sort of chart later on? I know this will take time obviously, or is there a site/source i should take out later for this info?
reply
This is all great. I appreciate all this info. , i'm in the waiting camp myself to get 3080. i appreciate all info. My only question is when things settle down what sources will i have to study the PCB quality of various manufactures so i know when i do buy one i'm getting the best quality card? Will Gamers nexus do some sort of chart later on? I know this will take time obviously, or is there a site/source i should take out later for this info?
reply
Vic
As SHOCKING as it is... but belive it or not - T.U.F. seem to Not cut any corners when it comes to components... and i have a Massive Respect to ASUS for that.
Whats RLY dissapointing... is that a LOT(in FACT - MOST..) of our Favorite Partners cards.... HAVE been affected by a mojo Lets make them as CHEAP as we can and rip off our costumers as much as we possibly can ..YAY..
Dissapointing....
reply
As SHOCKING as it is... but belive it or not - T.U.F. seem to Not cut any corners when it comes to components... and i have a Massive Respect to ASUS for that.
Whats RLY dissapointing... is that a LOT(in FACT - MOST..) of our Favorite Partners cards.... HAVE been affected by a mojo Lets make them as CHEAP as we can and rip off our costumers as much as we possibly can ..YAY..
Dissapointing....
reply
The
I think you are confusing through hole and SMD capacitors. other board partners RARELY use through hole capacitors (through hole means they go through the entire pcb via a hole and are soldered on the other side). i think you simply mean the cylindrical capacitors vs these cuboid capacitors, since those cylindrical capacitors are pretty standard from what i know.
reply
I think you are confusing through hole and SMD capacitors. other board partners RARELY use through hole capacitors (through hole means they go through the entire pcb via a hole and are soldered on the other side). i think you simply mean the cylindrical capacitors vs these cuboid capacitors, since those cylindrical capacitors are pretty standard from what i know.
reply
Add a review, comment
Other channel videos
26:48
