Tldr; Placing caps close to the load makes them perform more as ideal caps, thereby reducing transients/ripple more. This is more important for small caps as their resonant frequency is affected more by distance relative to bulk caps. + an explanation on how to figure out if a capacitor will have an effect before you add it (which can be used to determine if you should remove your larger caps far away). Your statement of having caps close to the load and 'screw the VRM' is correct. The reason you want caps closer to the load is to smooth out the ripple/transient, as you surely know. Placing the caps as close to the load as possible reduces the effective ESL due to less being introduced from the distance the current must travel across the PCB. I.e a cap will work more as an ideal cap (decoupling higher up in frequency) the closer it is. This is also why the small low ESL/ESR MLCCs are placed closer to/right on the back of the chip as these handle the highest frequencies as opposed to the bulk caps which are large and thus handles the larger and lower frequency power required before the VRM figures out what to do. The resonant frequency I.e. The frequency a cap will reach its minimal impedance (thereafter starting to increase) is: 2pi*f=1/sqrt(L*C) Where L is the total inductance seen by the cap (ESL + ESL_PCB_distance+via) and C is the capacitance value. As larger caps tends to have higher ESL and obviously capacitance, their resonance frequency will be less affected by distance from the load, while the small low ESL caps will be highly affected. This is the reason bulk capacitance tends to be close to the VRM (as this is the lower priority cap) because it is the closest it can get, not because it is better to have it close to the VRM than the core; it is just less bad to have it further form the core than a MLCC. Same argument for why the bulk cap is closer to the load than the VRM. On a similar note to the resonance above the impedance of a cap at a given frequency can be calculated as: Z = 2pi*f *L + 1/(2pi*f*C)+R Here L is the total ESL for both cap and PCB, C is capacitor capacitance and R is the ESR. this can be used to calculate if a cap you add will have any effect at the frequency you are trying to suppress. For example adding a huge electrolytic capacitor far away to suppress 2GHz noise will have little effect while a MLCC close to (even though much smaller in capacitance) will. Note however that the bulk may not reduce high frequency ripple, but may help your load trancient, which I do expect is much lower in frequency. Also placing caps close to the load makes current not having to travel across the power plane disturbing anything in its path (don't know how much this affects GPUs, as I don't know how many components are connected to the core voltage power plane in between VRM and gpu core, but in general EE design this is also a reason to place caps close to the load and not the VRM). Thus in summary the closer you place caps to the load the less transient your VRM will see.
I come from an audio background where fast/large decoupling caps in certain circuits can make a huge difference in performance, and occasionally wondered if a similar thing could work on GPU's, without ever actually looking into it, but I'm super interested now. Looking forward to watching this through to the end, thanks for the long format vid and rock on!
I think this is a perfect example of efficiency. Thus the power leveling means lower input voltage. Good one droidzuu. Unless starting power is undervolting for too long even higher than 500 would benefit.
the moment I saw that PCB I thought "looks like power color" saw the number above the pcie fingers and knew that was power color. I got a Powercolor 7850 with a similar looking board. Seems to perform well even though it's also missing phases with no cooling on VRMs and OCP is either not there or doesn't work. No cooling on memory also sucks, it gets overheated with heavy memory OC (solvable with some means of sinking heat- I used aluminum flatbar as a heatspreader, and thermaltaped heatsinks on that). I think I should do a bit more with that card.. only 1 PCIe lane still works for some reason, but overall it still works. I want to grab some Tahiti cards to play with, not just pitcairn, hawaii, and polaris I have now.
we had 2 vtx3d 7850 2gb single fan they were pretty cool with there red shroud :) that was back in 2015 ish my oncle used them to crypto mine. i used to love them cause they looked cool
Astonishing. You solder caps to parts my eyes can hardly spot :D Edit: And while I watch this it comes to my mind that because of you (mostly) I understand how CPUs work (mostly). But now I see, I don't understand anything about GPUs :O
That amount of modding was quite a surprise. I thought you only added some cap to smoothen the voltage core but nope. Copious amounts of capacitors. Caught me really good 😅
compared to the ceramics by the GPU, the electrolytics you added probably still had lots and lots of ESR at the high frequencies a GPU switches its load. There are bigger ceramics that would probably have been of more use directly under the core. But adding bulk capacitance in the "far away" spot couldn't have worsened things.
Try adding a few thick stranded wires between the chokes and GPU (with all those caps still on) to decrease the resistance of the power line and report on your findings!
HOT GPU PICS :):) THANKS TO YOU IVE BOUGHT RX 570 PULSE FROM SAPPHIRE and its great. I will go all Red but for now my 4,3ghz i7 950 hold up good. (Also my rx570 peaks at 1480mhz core and 2250 memory)
The bad transient response on some/many of those graphics cards, could that be a reason for AMD to have that stock voltage so ridiculously high? Many AMD-GPU's can be undervolted with around 0.09-0.1 V. If AMD wouldn't have sufficiently strict rules then I can see AMD demand stock voltage which is a bit too high to compensate for those bad transient responses. What do you guys think about this theory?
Generally 1.45V is really the highest someone should go for daily usage. Even that is kinda high and you may want to drop it to 1.4V. Especially when you don't have Samsung B Die.
Hey buildzoid. There is a new company called colorful making x570 boards. Whats your take, does it get your praise or you will just roast it like a master sli. Please let us know. Please please please do this🙏🙏🙏🙏🙏
"I bought it because it has a red pcb" Me: (sees a gtx 1080ti hof) also me (already has a 1080ti) my retarded brain: *COP COP COP IT HAS A WHITE PCB OH MY F****** GOD REEEEEEEE*
Hey, I had my Xfx R7970 3gb Black edition Die- It one day (a month ago) started having the occasional 1hz screen flicker Which included Blue and Orange colors in it sometimes (memory?) was gaming and the pc crashed and gray screened, pc then refused to boot. After trying 20 restarts it turned on once, when it did it went to desktop and seemed fine except it was at 150mhz core 300 mem vddc 0.806 (turned off OC-should be Stock 1000c 1425m 1.113vddc constantly) Under load the card starts at 150 300mhz and it jumped up to stock however the vddc increased to 1.090 and then started decreasing until crash with gray screen (also drops to 150c 300m on crash) booted did it again, Now it won't boot at all. (tried like literally 200 times xD) Take out gpu and pc boots fine, put in other pc it didn't boot. So rip got a rx 580 8gb gigabyte gaming. BUT i wish i knew why it stopped, Any ideas???
It's just that before he said that he would not get it because of powermod likley not being applicable to navi because of bios lock. The power scaleing from what i've seen on IgorsLab is better than vega. So i think it would be cool.@@haukikannel
You do realize that you just spent a fair bit of time talking about your "probing technique" ..... I'm not sure that's too wise... ya know , don't ask don't tell...lol
Try adding a few thick stranded wires between the chokes and GPU (with all those caps still on) to decrease the resistance of the power line and report on your findings!
Try adding a few thick stranded wires between the chokes and GPU (with all those caps still on) to decrease the resistance of the power line and report on your findings!
5700XT with AMD CPU fans,BZ Limited Edition!
-How harcore overclockers choose graphic card?
-I want Radeon with red PCB.
Yeah, RED is faster! I got it! ;)
Its such a Buildzoid thing to say, isnt it?
He uhh... he took the red one.
two boxed fans on a gpu ..
Function over form
Tldr;
Placing caps close to the load makes them perform more as ideal caps, thereby reducing transients/ripple more.
This is more important for small caps as their resonant frequency is affected more by distance relative to bulk caps.
+ an explanation on how to figure out if a capacitor will have an effect before you add it (which can be used to determine if you should remove your larger caps far away).
Your statement of having caps close to the load and 'screw the VRM' is correct.
The reason you want caps closer to the load is to smooth out the ripple/transient, as you surely know. Placing the caps as close to the load as possible reduces the effective ESL due to less being introduced from the distance the current must travel across the PCB. I.e a cap will work more as an ideal cap (decoupling higher up in frequency) the closer it is. This is also why the small low ESL/ESR MLCCs are placed closer to/right on the back of the chip as these handle the highest frequencies as opposed to the bulk caps which are large and thus handles the larger and lower frequency power required before the VRM figures out what to do.
The resonant frequency I.e. The frequency a cap will reach its minimal impedance (thereafter starting to increase) is:
2pi*f=1/sqrt(L*C)
Where L is the total inductance seen by the cap (ESL + ESL_PCB_distance+via) and C is the capacitance value. As larger caps tends to have higher ESL and obviously capacitance, their resonance frequency will be less affected by distance from the load, while the small low ESL caps will be highly affected. This is the reason bulk capacitance tends to be close to the VRM (as this is the lower priority cap) because it is the closest it can get, not because it is better to have it close to the VRM than the core; it is just less bad to have it further form the core than a MLCC. Same argument for why the bulk cap is closer to the load than the VRM.
On a similar note to the resonance above the impedance of a cap at a given frequency can be calculated as:
Z = 2pi*f *L + 1/(2pi*f*C)+R
Here L is the total ESL for both cap and PCB, C is capacitor capacitance and R is the ESR. this can be used to calculate if a cap you add will have any effect at the frequency you are trying to suppress. For example adding a huge electrolytic capacitor far away to suppress 2GHz noise will have little effect while a MLCC close to (even though much smaller in capacitance) will. Note however that the bulk may not reduce high frequency ripple, but may help your load trancient, which I do expect is much lower in frequency.
Also placing caps close to the load makes current not having to travel across the power plane disturbing anything in its path (don't know how much this affects GPUs, as I don't know how many components are connected to the core voltage power plane in between VRM and gpu core, but in general EE design this is also a reason to place caps close to the load and not the VRM).
Thus in summary the closer you place caps to the load the less transient your VRM will see.
I come from an audio background where fast/large decoupling caps in certain circuits can make a huge difference in performance, and occasionally wondered if a similar thing could work on GPU's, without ever actually looking into it, but I'm super interested now. Looking forward to watching this through to the end, thanks for the long format vid and rock on!
37:56 the hot gpu pics folder
see a new video...get excited...the realize it's not the 4th installment of the ramblings about gigabyte mobos.... :***(
0:50 - 1:06
Still a better love story than Twilight
I think this is a perfect example of efficiency. Thus the power leveling means lower input voltage. Good one droidzuu. Unless starting power is undervolting for too long even higher than 500 would benefit.
the moment I saw that PCB I thought "looks like power color" saw the number above the pcie fingers and knew that was power color. I got a Powercolor 7850 with a similar looking board. Seems to perform well even though it's also missing phases with no cooling on VRMs and OCP is either not there or doesn't work. No cooling on memory also sucks, it gets overheated with heavy memory OC (solvable with some means of sinking heat- I used aluminum flatbar as a heatspreader, and thermaltaped heatsinks on that). I think I should do a bit more with that card.. only 1 PCIe lane still works for some reason, but overall it still works. I want to grab some Tahiti cards to play with, not just pitcairn, hawaii, and polaris I have now.
we had 2 vtx3d 7850 2gb single fan they were pretty cool with there red shroud :) that was back in 2015 ish my oncle used them to crypto mine. i used to love them cause they looked cool
Crap... I was hoping for rambling about Gigabyte.
THIS times 100
Astonishing. You solder caps to parts my eyes can hardly spot :D Edit: And while I watch this it comes to my mind that because of you (mostly) I understand how CPUs work (mostly). But now I see, I don't understand anything about GPUs :O
That amount of modding was quite a surprise. I thought you only added some cap to smoothen the voltage core but nope. Copious amounts of capacitors. Caught me really good 😅
I don’t know why but this look so cool with two modern AMD fan on top. A custom 3D printed housing will do.
compared to the ceramics by the GPU, the electrolytics you added probably still had lots and lots of ESR at the high frequencies a GPU switches its load. There are bigger ceramics that would probably have been of more use directly under the core. But adding bulk capacitance in the "far away" spot couldn't have worsened things.
Try adding a few thick stranded wires between the chokes and GPU (with all those caps still on) to decrease the resistance of the power line and report on your findings!
/me sees the backside ::: *Reactor Edition* looks fine to me :)
Video: Posted 10 minutes ago.
You 6 days ago: "..."
@@LuisC7 should have posted "first" :P :: answer to your comment is : early access to some videos for supporters
@@uBm3d I knew it, but I wanted to sound stupid.
I don't know anything about this but it's interesting
lol ikr
I understand some bits but yeah sometimes you sit there like wut
Red PCB!
39:44 burp
I miss my 7970. It was a great card with the exception of ASUS completely castrating the voltage control... Essentially there was none.
hi buildzoid critter. when you gona review gigabyte x570 boards? im more looking forwars to your reviews of the B550 iF there will be any of those.
Have you try supercapacitors? Holds more power for lower voltage, more fitted for CPU/GPU VCore.
The card looks like a Club 3D HD 7970 royalKing/Queen, if im not mistaken.
Use house water supply to do testing with constant coolant temp. Just run the loop from the faucet to the drain.
Linus tried that... Can't speak for BZ, but Linusc attempt didn't go swimmingly. 😉
your 7970 is covered in ticks, figures.. so red it must be pure blood!
HOT GPU PICS :):)
THANKS TO YOU IVE BOUGHT RX 570 PULSE FROM SAPPHIRE and its great. I will go all Red but for now my 4,3ghz i7 950 hold up good. (Also my rx570 peaks at 1480mhz core and 2250 memory)
Is the Gigabyte X570 motherboards edition happening ;-)
The moment I noticed the "Hot GPU Pics" folder ^^
thanks
G'day Buildzoid,
How Awesome would a RX5700XT look with this fan setup & shroud, 🤔 HMMM! I think it could be a project for me
The bad transient response on some/many of those graphics cards, could that be a reason for AMD to have that stock voltage so ridiculously high? Many AMD-GPU's can be undervolted with around 0.09-0.1 V. If AMD wouldn't have sufficiently strict rules then I can see AMD demand stock voltage which is a bit too high to compensate for those bad transient responses. What do you guys think about this theory?
i'ed like to see the gtx660 with voltage unlocked bios given the cap treatment
"Hey guys Gigazoid here, and today we're.."
Good morning
RIP my RED PCB R9-280x 2013-2019
What else can i use these wraith coolers for?
Hello Buildzoid, is 1.5 volt or a bit lower safe for DDR4 24/7? Hynix C-DIE. Thank you
Generally 1.45V is really the highest someone should go for daily usage. Even that is kinda high and you may want to drop it to 1.4V. Especially when you don't have Samsung B Die.
maybe start series where you build your own pcb
Wheres our gigabyte mobo ramblings bz?
I feel like your viewer retention is lacking. Any Editing at all with jumpshots to what you're talking about would help.
Where is the 5870? Can u oc that card with LN2 on the tube?
Get a 5700 for free and unleash it's real power
1:11 - 1:17 Ehm, what?
HOT GLUE! Lol........
Hey buildzoid. There is a new company called colorful making x570 boards. Whats your take, does it get your praise or you will just roast it like a master sli.
Please let us know. Please please please do this🙏🙏🙏🙏🙏
*goes to like video because buildzoid said something funny*
*remove like because it was already liked for another reason*
AMD HIRE THIS MAN
But only if he puts up his rambling videos.
@@whiskeymike7364 yes at twice the length
"I bought it because it has a red pcb"
Me: (sees a gtx 1080ti hof)
also me (already has a 1080ti)
my retarded brain: *COP COP COP IT HAS A WHITE PCB OH MY F****** GOD REEEEEEEE*
Hey, I had my Xfx R7970 3gb Black edition Die- It one day (a month ago) started having the occasional 1hz screen flicker Which included Blue and Orange colors in it sometimes (memory?) was gaming and the pc crashed and gray screened, pc then refused to boot. After trying 20 restarts it turned on once, when it did it went to desktop and seemed fine except it was at 150mhz core 300 mem vddc 0.806 (turned off OC-should be Stock 1000c 1425m 1.113vddc constantly) Under load the card starts at 150 300mhz and it jumped up to stock however the vddc increased to 1.090 and then started decreasing until crash with gray screen (also drops to 150c 300m on crash) booted did it again, Now it won't boot at all. (tried like literally 200 times xD) Take out gpu and pc boots fine, put in other pc it didn't boot. So rip got a rx 580 8gb gigabyte gaming. BUT i wish i knew why it stopped, Any ideas???
That R7970 literally scored higher in Superposition benchmark then my new rx 580 8gb gigabyte gaming. Not sure how to feel about that.
Pls get navi, there are ppl table registry mods!
So the owerclocking is not fun and hard ;) Bullzoid like to Mess upp things and make something that ordinary people can not so not likely to happen ;)
It's just that before he said that he would not get it because of powermod likley not being applicable to navi because of bios lock. The power scaleing from what i've seen on IgorsLab is better than vega. So i think it would be cool.@@haukikannel
that's a ghetto ass shroud
Какая максимальная частота gpu получилась? И где тесты в играх?)
Когда ждать видеокарту gtx 760 на опытах?)
First
You do realize that you just spent a fair bit of time talking about your "probing technique" ..... I'm not sure that's too wise... ya know , don't ask don't tell...lol
Try adding a few thick stranded wires between the chokes and GPU (with all those caps still on) to decrease the resistance of the power line and report on your findings!
Try adding a few thick stranded wires between the chokes and GPU (with all those caps still on) to decrease the resistance of the power line and report on your findings!
already thought about doing that on some cards.