The Elmor EVC also works for a huge amount of other cards. If you ever want to try something like this you can get it here: elmorlabs.com/product/elmorlabs-evc2se/ (this is not paid, but I love Elmors work!)
It is so weird that you just hit 100K subscribers. Man with a bigger knowledge than other youtubers. I wish you all the best and hope your channel will quickly grow to millions. I like your videos , great job!
thanks for featuring this @der8auer... but does the elmor EVC work for the 4090 founders edition? because maybe it uses different VRM controller (the new 16 phase one). compared to these other AIB cards?
another question... does the asus TUF share the same PCB as the STRIX? (and with same voltage VRM controller?). Since I can afford the TUF here but not the STRIX, is too expensive. Of course you cannot know this without having both cards to compare.
They've intentionally designed these cards so you cant undervolt them(thats the take i get with the less preformance doing it) and thus have them last thats evil and another reason to pass
@@DuBstep115 actually cuz most of them are just reading the spec sheet and running some benchmarks, and creating bs thumbnails, anyone can do that, but it takes experience and electronics knowledge to thoroughly review something like is the VRM design good, power delivery, components used etc even though der8auer dosent get everything right myself being an electronics enthusiast i just really appreciate that the content at least goes into the hardware
As always another Great Drill Down. You always get to the point. I think you need a couple million Subs. Unfortunately there arent enough geeks. Thanks for all you do. Thanks!
As someone who cringes at new GPU prices, I would never attempt this on my own. But seeing you do it is like watching someone drifting in a Ferrari, it's ludicrously enjoyable.
I have worked in I.T for nearly 19 years full time and I still don't quite understand everything you talk of. I rewind, pause, Google, etc and I am really thankful and impressed. Thank you
that is because the rest of channels and overclockers are still investigating on how to break world records, roman is just first, you kow that gamers nexus eventually will do it on ln2, probably jayz2cents too between others
@@zappulla4092 iirc they do some basic stuff, but yes, a engineer and overclocker like roman is the only one with the time, money and knoledge, follower by few other youtubers, like the guy on actually hardware overclocking, i think he matches in knowledge roman steve knows alot of things but interpret pcb designs is not yet on his reach
Its always a joy to see you do modifications, you're so dedicated.............thats what I like about a person doing what he likes with the utmost dedication.
I almost always undervolt my cards a little bit, the last small % of performance are rarely worth the extra power unless you really need them for something special.
Couple of protips for soldering annoying ground pins: 1. a higher power soldering iron (higher thermal power in watts, not hotter in degrees) will do the job better because it can apply more thermal power to the joint than the large plane can conduct away, thus overcoming the thermal conductivity and causing a higher local temperature increase for a smaller non-local temperature increase. same principle as TDP of a CPU vs. TDP max of a heatsink. the flipside of this is that the iron's temperature control feedback loop needs to be higher quality as to not cause overshoot. thermal overshoot causes your iron temperature to temporarily exceed the target temperature, and an iron that's too hot will burn off the flux and potentially cause pad delamination. 2. preheating the board with a little hot air as a "thermal soak" reduces the problem because the delta between the board temperature and the iron temperature is decreased. you're essentially saturating some of the thermal capacity of the plane so that it can't "absorb" as much heat from your iron (simplifying a bit here), which again results in higher local temperature increases vs. non-local temperature increases during soldering. this is my preferred method because you can use a decent general-purpose iron (e.g. FX888D) and a decent general-purpose hot air gun, rather than needing to buy a costly high-power iron with good temperature control. run the hot air at 150C or so, high flow, and warm up the board area in a circular motion, starting wide and far then moving inward and closer. it shouldn't melt plastic or lift any SMDs. only takes about 20 seconds of preheat to make these joints a ton easier. I'm guessing you know most/all of this, but folks in the comments might not and I often see people trying to compensate by turning the iron temperature up! (super risky!)
The lower voltage same clock also happened when undervolting the 30 series. You basically end up clock stretching where it's not fully stable but it's stable enough not to crash
Other tech tubers can't tell if a cable has solid cores or stranded cores. We are lucky to have der8auer and give us real world scientific experiments.
They didnt leave much room for improvement for their AIBs. Higher prices and very small improvement and smallish profit for the AIB. Nvidia going the way of Apple now.
@@smugmode I don't know about sinister but it smacks of a weird mix of arrogance, panic and paranoia. Calling their new "feature" DLSS 3 is even misleading, since it's DLSS 2 with a separate additional "40 series only" insertion of faked frames. Looks great for benchmarks and makes comparisons difficult, of course.
Damn!!! Why board so short? Feel sorry for the textolite on a video card for 1600+ Euros? And it's Asus Strix! I can't believe my eyes, why is this happening and what is the reason?
Thank tou Roman for taking time to do actual testing on these devices. Also it's a convention red/black is always supply, you can take ground anywhere on the card.
Thank you for quoting weight data; this confirms my suspicion that these cards weigh more than an entire 15" laptop, lol. I'm excited to see what these cards will do on water cooling/sub-zero...
The point of the three adjacent through-hole pads next to the voltage readouts is so you have an accurate local ground reference to measure from. Due to the high current flowing through the ground plane (primarily return currents) there will be a small voltage drop across it, so if you measure from the far side you're not necessarily getting an accurate reading. The 90mV you measured between those those pins and the grounded screw pad is exactly that - both pads are connected to ground, but the ground plane has a nonzero resistance/impedance so there's a voltage drop between the left and right sides. If you measure that ground plane drop during operation you'll find that it shifts a little as the current draw increases and decreases. If you connect a differential oscilloscope probe across those points you'll likely see some high frequency noise caused by the dI/dt from each VRM phase switching in sequence. The voltage of that noise is the summed vector magnitude of the currents flowing through the plane multiplied by the impedance of the plane. In designs where there's a high-current section alongside a low-current section containing sensitive (often analogue) electronics, and where all the signals between those two areas are differential and isolated (e.g. with an optocoupler or galvanic isolator), they'll often split the ground and power planes for those two regions and connect the high-current and low-current ground planes with a ferrite bead to prevent the ground plane noise leaking out into sensitive circuitry.
I don't care that much for overclock anymore these days (loved the days where it could easily give you insane gains). Still like to see your videos for it's very informative and you never know, there might be a chance there is performance increase to be gained.
I run my old AMD Ryzen 7 3700X 3.6 GHz at 4.2 GHZ with the box cooler. When you only have to run the Ryzen Master program, the extra performance is well worth it.
@@Varangian_af_Scaniae Yea but back in the day with some old intel CPUs you could go from like 3.2 ghz to something like 4.6ghz with just simple OC and all you needed was some typical air cooler. Back then overclocking was very worth it and made the cheaper CPUs great deal.
seriously. overclocking the crap out of my 6800xt, air cooled, gains so little for such a huge cost. honestly i dont know if thats a good or bad thing. probably good since the assumption is that out of the box you are getting the most out of your card without doing extreme overclocking with super cooling which in my case is overkill unless im doing it for the fun of it but not practical at all.
awesome video brother as always you definitely take it to another level. love the fact you open something up and see what it's about. looking forward to seeing your next video
The only one that does Overclocking on another level, everyone else is basically only relying on software only. Cant wait to see what he will manage to achieve in future with this card.
The voltage reduction loss tracks Ohm's Law. If the voltage drops, but the resistance remains the same, there's a corresponding drop in current. When one is talking about 3 billion or so operations per second, any slowdown in the ability to charge & discharge gates on the tiny FETs in the die, will tend to net a performance decrease. Gotta watch those thermals though.
Would also really appreciate a review looking at the amount of reduced power draw under constant loads while lowering temperatures, TSMC’s 5 and 4 nm processes seem to like that quite a bit.
Thanks der8auer, you are absolutely awesome. I really enjoy how you ask questions about hardware and then test your hypothesis in the same video! Others simply ask the questions and shrug, never to find the answers themselves. You have a true engineers mindset, THANK YOU! I'm just curious, now that we've removed the gaming GPU bottleneck at 4K, is it possible you could test some CPUs at 4k? I haven't seen others do this with the new 4090s. I'm wondering if we have a new value CPU king at 4K? Especially wondering if 5800X3D still stacks up well against Ryzen 7k series or how the 5900x compares! Do we have any values or must we simply pair high end GPU with high end CPU? Thanks!
Damn man this made me miss the 8800 GT/GTS/GTX days where we used to solder on a little turn knob to adjust the voltage :) Or pencil mod on old X1950 for 1.4V :) keep up the high tech overclocking bud !!
I do remember one day, when my friend bought a GTX 580, he did what i did not expect.. he got an another GTX but this time, broken 590 which is dual chip card, he just cut the mosfet board out of this card and made every connection needed for get a working 580 card, he also moded a bios and hoked some programming probes to the 8 pin chips on the board to manually increase voltage to the card. He pushed this 580 to its absolute limit, core clock was 900MHz higher than what card gives you by stock and the powerdraw was increased from 250W to 390W. The card was faster than dual 590 in some games and benchmarks and equally fast like GTX 780 back in these days. And also he removed stock HS and replaced it with water cooling..
I think you had a bad silicon on that Strix, my Gainward Phantom GS does at Time Spy Extreme 1.0V - 2940MHz 1.02V - 3000MHz 1.06V - 3075MHz 1.1V - 3105MHz BIOS-modding would be nice on non-strix cards to Strix bios, because I am stuck at TDP-limit of 515W.
@@cptnsx Time Spy Extreme is TDP-limited at Graphics Test 2, I drop down there to 1.0V with 515W powerconsumption. Beginning of Graphics Test 1 the 3105MHz @1.1V is consuming 515W. So to answer your question (at TimeSpyExtreme), between 1.0 and 1.1V the graphics card can consume 515W, it depends only from graphics-demand. This is the same for games. Forza Horizon 5 consumed lot less power (when I tested long time ago with RX 6900 XT), than Metro Exodus. Propably same story with the RTX 4090. --> You could fine tune your graphics card for each game individually :P I prefer getting the most insane graphics-demand, tune the graphics card with it until stable, and then play every game with that fine tune. This way no matter what kind of game I play, my fine-tune settings will always hold (if not, then reduce core-clock with 15MHz steps).
@@bezol5845 Thanks for info. I was wondering if you had to surpass the 450 watt limit to get those numbers and if so by how much. How were your temps?
At what point will you melt the power connector. I think that will be your next limiting factor. I don’t think you’ll see crazy watt numbers like on the 3090ti with LN2. We’ll have to water cool the power cables and connector moving forward because there is no way in hell that connector and cables will sustain 800-900w. Like this is 12 pins vs 24 with as far as I know the same gauge on the wires but the connector pins are half the gauge. I would buy one of those connectors that’s on the pcb, wire a variable load to it and see how far you can go before the stuff starts melting before melting it on a 2000$ GPU.
Wireview could come with a very small extention cable from the wireview to the gpu. you could use if the gpu power connector was backwards. Maybe better than making a hole new model.
Regarding the 6:40 soldering. That is not good what you did. You should have your own wires that you solder to the desired spot and they should have an ending with a simple connector. And then on the device you solder 3 connectors that fit inside the connectors of the wires from the board. This way, if you don't solder the correct wire to the board you just unplug and plug the correct one.
Is it possible that the card doesn't actually target 600W? We know the power connectors are there but maybe Nvidia did something in the bios to prevent reaching 600W due to various tech tubers expressing their concerns about such high power draw for the average user. Looking forward to the power mod video, as well as when you have your direct voltage meter ready for the STRIX. 😁
@@APhamx7 They literally canceled the '4080 12GB' over bad press. And they are reportedly going to lower the price as well. That is due to tech tubers and the tech community's collective response. So what is unfeasible about them hard-locking the card at silicone or bios level to not actually reach 600W after everyone had been concerned about the high power consumption since it leaked?
The cooler was created with Samsung's node. They ended up switching to the tsmc 4nm node but kept the coolers since they had all the money into them. So they are over kill for these chips.
Could you have a look at the Inno3D OC? It is artificially capped at it's 450% power limit regardless if you use a 4 head or 3 head connector and I'd like to see what you can do with it. I'm a little disappointed at the moment and are considering taking it back, but no one's looked at the card other than kit guru and they didn't go into too much detail on the power limit.
@@zappulla4092 yes. Buying a product you know nothing about the first day it releases is a terrible decision. How did you not understand? Use your brain. Something people lack in 2022.
I wonder if I should buy 4090 or wait for the possible 4090ti. Also would be nice to see what the temps are with ek-waterblock, since I need to get one due to my old 1080ti is watercooled too, so would be too much work to not just add new card into the loop.
Would undervolting the processor on the gpu help or would you just hit the wall of the boost clock speed set in the card’s bios?
2 роки тому
it makes sense... with 1V 350W for GPU itself is already 300A, which is insane... but if you push it by 0.1V and still want same clocks, I'd assume current won't go down, so I'd expect 10% 385W but it seems like higher voltage also means higher current with this? because that's a significant increase I also have to wonder if the efficiency of VRM itself is the same for 1.1V and 1.2V, because nvidia said they managed to squeeze 30W just from having better VRM than previous generation on founders card... because if your VRM drops in efficiency and you are limited by power draw before (not after) change of voltage, you might hit the power limit just because your VRM itself draws too much power and kinda related... I really miss the voltage measurement points on my old R6970 Lightning where you didn't have to solder anything, there was adapter with holes were made to fit regular multimeter probes and it would hold it in place, it was awesome
It's almost as if they're holding these cards back. Maybe until we get newer equipment to go along with it. Are they worried about their components under too much stress? Or the air cooling system is an adequate enough?
i seen another tech tuber that had a 4 way power adapter into the card giving it the framework to draw more power allowing it to overclock to 133% power
How much power can the PS connector supply? It would be interesting to see the VD under load. It might be a good test to see if an automotive voltage stabiliser and battery could help when going extreme with overclocking.
3000 series was the same with undervolting, i was seeing differences at the same clocks on a different voltage. although not as noticeable in games on my 3080
i have been searching about this topic for the last two hours and gave up , then i go to youtube to find something to sway me away from this and see der8auer uploading this
The Elmor EVC also works for a huge amount of other cards. If you ever want to try something like this you can get it here:
elmorlabs.com/product/elmorlabs-evc2se/
(this is not paid, but I love Elmors work!)
It is so weird that you just hit 100K subscribers. Man with a bigger knowledge than other youtubers. I wish you all the best and hope your channel will quickly grow to millions. I like your videos , great job!
thanks for featuring this @der8auer... but does the elmor EVC work for the 4090 founders edition? because maybe it uses different VRM controller (the new 16 phase one). compared to these other AIB cards?
another question... does the asus TUF share the same PCB as the STRIX? (and with same voltage VRM controller?). Since I can afford the TUF here but not the STRIX, is too expensive. Of course you cannot know this without having both cards to compare.
They've intentionally designed these cards so you cant undervolt them(thats the take i get with the less preformance doing it) and thus have them last
thats evil
and another reason to pass
which one do you recommend more? Aorus master or Strix?? I am planning to build with one of those two and can't really decide
you are on a completely different level than other tech youtubers.
You have know Teclab channel, they are the best moders.
Because the market for delidding CPUS and soldering 2000€ GPUS is like 0.00001% of the population
@@DuBstep115 but that's the fun part :D
@@der8auer-en just now realised you do these videos in both english and germany, not just dub it. Hats of to you, insane dedication
@@DuBstep115 actually cuz most of them are just reading the spec sheet and running some benchmarks, and creating bs thumbnails, anyone can do that, but it takes experience and electronics knowledge to thoroughly review something like is the VRM design good, power delivery, components used etc even though der8auer dosent get everything right myself being an electronics enthusiast i just really appreciate that the content at least goes into the hardware
As always another Great Drill Down. You always get to the point. I think you need a couple million Subs. Unfortunately there arent enough geeks. Thanks for all you do. Thanks!
Thanks a lot! Will invest in more snacks for Shiek and Makita :D
@@der8auer-en
@@halrichard1969
@@der8auer-en wait, why is one of your cats named after a power tools brand?
@@HDJess It's a cat, what's he gonna do? Name it Phil, or Steve, or Tom? That would just be weird...
As someone who cringes at new GPU prices, I would never attempt this on my own.
But seeing you do it is like watching someone drifting in a Ferrari, it's ludicrously enjoyable.
It’s impressive that manufacturers are dialing in their chips so well. The golden days of crazy headroom seem to be in our rear view mirror.
Great stuff. I love when you push the limits of hardware!
That's what she said
LOL. Limits.
I have worked in I.T for nearly 19 years full time and I still don't quite understand everything you talk of. I rewind, pause, Google, etc and I am really thankful and impressed. Thank you
This is the type of content no one else is doing. Interesting. I like it! 👍
that is because the rest of channels and overclockers are still investigating on how to break world records, roman is just first, you kow that gamers nexus eventually will do it on ln2, probably jayz2cents too between others
@@arch1107 they don’t do voltage mods like this. They might use LN2 but they openly admit they don’t know the pcb parts like Roman.
@@zappulla4092 iirc they do some basic stuff, but yes, a engineer and overclocker like roman is the only one with the time, money and knoledge, follower by few other youtubers, like the guy on actually hardware overclocking, i think he matches in knowledge roman
steve knows alot of things but interpret pcb designs is not yet on his reach
This is just on the whole new absurdity level of power consumption
Was looking forward to you doing a voltage mod since you put out your last video! Excited to see how far you can push this card with a power mod 😀
Its always a joy to see you do modifications, you're so dedicated.............thats what I like about a person doing what he likes with the utmost dedication.
This is why I love der8auer.
Undervolting did make the GPU a bit slower, but 17% more efficient
I almost always undervolt my cards a little bit, the last small % of performance are rarely worth the extra power unless you really need them for something special.
@@kevinmalk
How far did you go?
My RX470 went bollocks if I lowered the voltage beyond -30mV. I've heard people go as low as -100mV. Is that true?
Dont think 90 series buyers want efficiency
@@cor74 sure, they was inefficient from the moment they bought it 😂
That GPU is just a huge overkill.
@@Talkshowhost23 nothing is overkill if you're rich enough
Couple of protips for soldering annoying ground pins:
1. a higher power soldering iron (higher thermal power in watts, not hotter in degrees) will do the job better because it can apply more thermal power to the joint than the large plane can conduct away, thus overcoming the thermal conductivity and causing a higher local temperature increase for a smaller non-local temperature increase. same principle as TDP of a CPU vs. TDP max of a heatsink. the flipside of this is that the iron's temperature control feedback loop needs to be higher quality as to not cause overshoot. thermal overshoot causes your iron temperature to temporarily exceed the target temperature, and an iron that's too hot will burn off the flux and potentially cause pad delamination.
2. preheating the board with a little hot air as a "thermal soak" reduces the problem because the delta between the board temperature and the iron temperature is decreased. you're essentially saturating some of the thermal capacity of the plane so that it can't "absorb" as much heat from your iron (simplifying a bit here), which again results in higher local temperature increases vs. non-local temperature increases during soldering. this is my preferred method because you can use a decent general-purpose iron (e.g. FX888D) and a decent general-purpose hot air gun, rather than needing to buy a costly high-power iron with good temperature control. run the hot air at 150C or so, high flow, and warm up the board area in a circular motion, starting wide and far then moving inward and closer. it shouldn't melt plastic or lift any SMDs. only takes about 20 seconds of preheat to make these joints a ton easier.
I'm guessing you know most/all of this, but folks in the comments might not and I often see people trying to compensate by turning the iron temperature up! (super risky!)
Congrats on 100k on the English channel, btw! Loving this "for science" content. It's great :D
I can't wait to see you power mod the card!!! You're freaking awesome dude!!! It's so fun to watch you work, & figure things out!
This is just crazy, cant wait for the next video! :D
The lower voltage same clock also happened when undervolting the 30 series. You basically end up clock stretching where it's not fully stable but it's stable enough not to crash
Probably why my gpu doesn't crash but flashes random rainbow blobs sometimes
Other tech tubers can't tell if a cable has solid cores or stranded cores.
We are lucky to have der8auer and give us real world scientific experiments.
Looking more and more like nVidia have tuned these cards right up to the max. Great job on the video!
They didnt leave much room for improvement for their AIBs. Higher prices and very small improvement and smallish profit for the AIB. Nvidia going the way of Apple now.
It's more sinister than that
@@smugmode I don't know about sinister but it smacks of a weird mix of arrogance, panic and paranoia. Calling their new "feature" DLSS 3 is even misleading, since it's DLSS 2 with a separate additional "40 series only" insertion of faked frames. Looks great for benchmarks and makes comparisons difficult, of course.
@@halrichard1969 Tim Cook deserves the right hook
@@cherry_blast2984 Never thought much of Tim Cook. Not an innovator just a care taker.
Damn!!! Why board so short? Feel sorry for the textolite on a video card for 1600+ Euros? And it's Asus Strix! I can't believe my eyes, why is this happening and what is the reason?
Thank tou Roman for taking time to do actual testing on these devices.
Also it's a convention red/black is always supply, you can take ground anywhere on the card.
Thank you for quoting weight data; this confirms my suspicion that these cards weigh more than an entire 15" laptop, lol. I'm excited to see what these cards will do on water cooling/sub-zero...
The point of the three adjacent through-hole pads next to the voltage readouts is so you have an accurate local ground reference to measure from. Due to the high current flowing through the ground plane (primarily return currents) there will be a small voltage drop across it, so if you measure from the far side you're not necessarily getting an accurate reading. The 90mV you measured between those those pins and the grounded screw pad is exactly that - both pads are connected to ground, but the ground plane has a nonzero resistance/impedance so there's a voltage drop between the left and right sides. If you measure that ground plane drop during operation you'll find that it shifts a little as the current draw increases and decreases.
If you connect a differential oscilloscope probe across those points you'll likely see some high frequency noise caused by the dI/dt from each VRM phase switching in sequence. The voltage of that noise is the summed vector magnitude of the currents flowing through the plane multiplied by the impedance of the plane. In designs where there's a high-current section alongside a low-current section containing sensitive (often analogue) electronics, and where all the signals between those two areas are differential and isolated (e.g. with an optocoupler or galvanic isolator), they'll often split the ground and power planes for those two regions and connect the high-current and low-current ground planes with a ferrite bead to prevent the ground plane noise leaking out into sensitive circuitry.
I have no idea what you just said, but it sounded smart.
Glad you touched on the undervolting question.
I don't care that much for overclock anymore these days (loved the days where it could easily give you insane gains). Still like to see your videos for it's very informative and you never know, there might be a chance there is performance increase to be gained.
I run my old AMD Ryzen 7 3700X 3.6 GHz at 4.2 GHZ with the box cooler. When you only have to run the Ryzen Master program, the extra performance is well worth it.
@@Varangian_af_Scaniae Yea but back in the day with some old intel CPUs you could go from like 3.2 ghz to something like 4.6ghz with just simple OC and all you needed was some typical air cooler. Back then overclocking was very worth it and made the cheaper CPUs great deal.
seriously. overclocking the crap out of my 6800xt, air cooled, gains so little for such a huge cost. honestly i dont know if thats a good or bad thing. probably good since the assumption is that out of the box you are getting the most out of your card without doing extreme overclocking with super cooling which in my case is overkill unless im doing it for the fun of it but not practical at all.
@@megapet777 is it worth overclocking the cpu though? personally i dont see any difference in gaming. i have the 3700x as well.
@@godw1ll99 i have 5600x and its like 2-5 fps increase. Really no point, same with GPU rather just leave it
Nicely done, and now I get why they locked the voltage.
LFG! Always bringing the good content!
I've been waiting for this!
Can't wait to see the next video with the power limit gone 🍿The amount of power this card will draw is going to be insane with the limits removed
really impressive! great job! well done!
awesome video brother as always you definitely take it to another level. love the fact you open something up and see what it's about. looking forward to seeing your next video
The cooler is amazing! What a fantastic heater for the winter lol.
The only one that does Overclocking on another level, everyone else is basically only relying on software only.
Cant wait to see what he will manage to achieve in future with this card.
a tips for that small voltmeter, add a diffuser sheet to make it easier to see under light.
The voltage reduction loss tracks Ohm's Law. If the voltage drops, but the resistance remains the same, there's a corresponding drop in current. When one is talking about 3 billion or so operations per second, any slowdown in the ability to charge & discharge gates on the tiny FETs in the die, will tend to net a performance decrease. Gotta watch those thermals though.
Would also really appreciate a review looking at the amount of reduced power draw under constant loads while lowering temperatures, TSMC’s 5 and 4 nm processes seem to like that quite a bit.
Thanks der8auer, you are absolutely awesome. I really enjoy how you ask questions about hardware and then test your hypothesis in the same video! Others simply ask the questions and shrug, never to find the answers themselves. You have a true engineers mindset, THANK YOU!
I'm just curious, now that we've removed the gaming GPU bottleneck at 4K, is it possible you could test some CPUs at 4k? I haven't seen others do this with the new 4090s. I'm wondering if we have a new value CPU king at 4K? Especially wondering if 5800X3D still stacks up well against Ryzen 7k series or how the 5900x compares! Do we have any values or must we simply pair high end GPU with high end CPU? Thanks!
Master class in overclocking. Cheers for the share mate!
Just love the soldering on a 1v old flagship gpu ❤️😁 power to the cutting edge tec tubers!
When will we get voltage hack?
Damn man this made me miss the 8800 GT/GTS/GTX days where we used to solder on a little turn knob to adjust the voltage :) Or pencil mod on old X1950 for 1.4V :) keep up the high tech overclocking bud !!
Every generation, you are the only dude i fully trust as far as testing the new hardware.
thank you again!
What about GN? They also have great reputation and they do pressure tests, Schlieren imaging etc.
I do remember one day, when my friend bought a GTX 580, he did what i did not expect.. he got an another GTX but this time, broken 590 which is dual chip card, he just cut the mosfet board out of this card and made every connection needed for get a working 580 card, he also moded a bios and hoked some programming probes to the 8 pin chips on the board to manually increase voltage to the card. He pushed this 580 to its absolute limit, core clock was 900MHz higher than what card gives you by stock and the powerdraw was increased from 250W to 390W. The card was faster than dual 590 in some games and benchmarks and equally fast like GTX 780 back in these days. And also he removed stock HS and replaced it with water cooling..
when I watch your vids I get reminded why I pursued my course
I love that the subtitles have "Founder's Edition" as "Furnace Edition"
Thanks you for doing the English version of your videos.
I think you had a bad silicon on that Strix, my Gainward Phantom GS does at Time Spy Extreme
1.0V - 2940MHz
1.02V - 3000MHz
1.06V - 3075MHz
1.1V - 3105MHz
BIOS-modding would be nice on non-strix cards to Strix bios, because I am stuck at TDP-limit of 515W.
@BeZol what are you watts at each of those settings?
@@cptnsx Time Spy Extreme is TDP-limited at Graphics Test 2, I drop down there to 1.0V with 515W powerconsumption.
Beginning of Graphics Test 1 the 3105MHz @1.1V is consuming 515W.
So to answer your question (at TimeSpyExtreme), between 1.0 and 1.1V the graphics card can consume 515W, it depends only from graphics-demand.
This is the same for games.
Forza Horizon 5 consumed lot less power (when I tested long time ago with RX 6900 XT), than Metro Exodus.
Propably same story with the RTX 4090.
--> You could fine tune your graphics card for each game individually :P
I prefer getting the most insane graphics-demand, tune the graphics card with it until stable, and then play every game with that fine tune. This way no matter what kind of game I play, my fine-tune settings will always hold (if not, then reduce core-clock with 15MHz steps).
@@bezol5845 Thanks for info. I was wondering if you had to surpass the 450 watt limit to get those numbers and if so by how much. How were your temps?
@@cptnsx Yes, +11% powerlimit added (this is the maximum mit Gainward Phantom GS)
I want to see that power limit mod. Sounds next level
Good work Sir, I can see you smiling :)
There is no communication on those extra pins. They are just grounded to enable the extra power input.
At what point will you melt the power connector. I think that will be your next limiting factor. I don’t think you’ll see crazy watt numbers like on the 3090ti with LN2. We’ll have to water cool the power cables and connector moving forward because there is no way in hell that connector and cables will sustain 800-900w. Like this is 12 pins vs 24 with as far as I know the same gauge on the wires but the connector pins are half the gauge. I would buy one of those connectors that’s on the pcb, wire a variable load to it and see how far you can go before the stuff starts melting before melting it on a 2000$ GPU.
Mystery of same clock but worse performance sounds like memory instability that error correction is fixing to keep the card stable.
That's exactly what happens when I get my 4090 rog strix oc over 3090 MHz - black screen. Sometimes an ad shows up as well.
How the hell do you only have 138k subs? That’s a lot but should be much more for what you provide…
Wireview could come with a very small extention cable from the wireview to the gpu. you could use if the gpu power connector was backwards. Maybe better than making a hole new model.
This man over here about to unleash the demon
Would be curious to know what kind of effects pushing the limits like this has on longevity of the gpu.
This is insane 🔥🔥🔥 great work
Regarding the 6:40 soldering.
That is not good what you did.
You should have your own wires that you solder to the desired spot and they should have an ending with a simple connector.
And then on the device you solder 3 connectors that fit inside the connectors of the wires from the board. This way, if you don't solder the correct wire to the board you just unplug and plug the correct one.
Is it possible that the card doesn't actually target 600W? We know the power connectors are there but maybe Nvidia did something in the bios to prevent reaching 600W due to various tech tubers expressing their concerns about such high power draw for the average user. Looking forward to the power mod video, as well as when you have your direct voltage meter ready for the STRIX. 😁
The Samsung node that these cards were going to use was targeting 600w. The newer tsmc 4nm process allowed them to hit less heat and power.
I don’t think Nvidia hardware engineers changed their design due to tech tubers opinions.
no. The card was never targeting 600W - you are mixing up bullshit leaks and intentional misinformation from AMD-fanboys with actual reality.
@@APhamx7 They literally canceled the '4080 12GB' over bad press. And they are reportedly going to lower the price as well. That is due to tech tubers and the tech community's collective response. So what is unfeasible about them hard-locking the card at silicone or bios level to not actually reach 600W after everyone had been concerned about the high power consumption since it leaked?
@@ModernOddity728 renaming something is far easier than changing the entire silicon design
Strix: *Uses all the power*
This guy: *Somehow uses more*
Really want to see some of these cards pushed to breaking point with shunt mods and water cooling!
People are still Reviewing and building with 4090 but this guy is on another level
You're a madman! Love it
Thanks for you efforts!
amazing mod, roman. thanks bro
Make a little mesh table for your cards so you can run them face down. A baking grill could work.
this channel for me is high end just likeGamers Nexus hardware unboxed and digital foundry Thanks for the content.
Did I miss how the frequency affected FPS at those voltages?
Reminds me of the old Intel P4 days and the Abit IC7-Max doing volt mods.
mateeeeeeeeee 10/10 I hate touching my components let alone this. Keep pushing those limits
The cooler was created with Samsung's node. They ended up switching to the tsmc 4nm node but kept the coolers since they had all the money into them. So they are over kill for these chips.
looks like clock stretching
maybe lowering the clock offset will give you more performance
I still can't figure out why they don't put the power connector at the end of the card to keep cable flex and plug strain to a minimum
Theses cards have some magic. How can we uncover it? Other people are seeing lower clocks produce more performance if the card is cooler.
Thank you for your videos!
omg 600 watt power limit not being enough is insane
can't wait for the power limit mod + water cooling to see some insane benchmarks
Wondering where you got the Digital LED multimeter..!
No idea what you did there, but.... interesting stuff! 😀👍
Thanks
Fyi 900w is the rating for most small bathroom heaters... Yikes
winter is coming :D
Just Remember guys With Insane power comes Insane responsibility!
Jeremy Clarkson, for once, seems appropriate:
"Speed, and power."
Could you have a look at the Inno3D OC? It is artificially capped at it's 450% power limit regardless if you use a 4 head or 3 head connector and I'd like to see what you can do with it. I'm a little disappointed at the moment and are considering taking it back, but no one's looked at the card other than kit guru and they didn't go into too much detail on the power limit.
This is why you dont pre order or order before reviews. You made poor life decisions. Next time use that brain.
@@tacticalcenter8658 decisions?
@@zappulla4092 yes. Buying a product you know nothing about the first day it releases is a terrible decision. How did you not understand? Use your brain. Something people lack in 2022.
I wonder if I should buy 4090 or wait for the possible 4090ti. Also would be nice to see what the temps are with ek-waterblock, since I need to get one due to my old 1080ti is watercooled too, so would be too much work to not just add new card into the loop.
Buy the 4090, sell it and buy 4090 ti
Would undervolting the processor on the gpu help or would you just hit the wall of the boost clock speed set in the card’s bios?
it makes sense... with 1V 350W for GPU itself is already 300A, which is insane... but if you push it by 0.1V and still want same clocks, I'd assume current won't go down, so I'd expect 10% 385W
but it seems like higher voltage also means higher current with this? because that's a significant increase
I also have to wonder if the efficiency of VRM itself is the same for 1.1V and 1.2V, because nvidia said they managed to squeeze 30W just from having better VRM than previous generation on founders card...
because if your VRM drops in efficiency and you are limited by power draw before (not after) change of voltage, you might hit the power limit just because your VRM itself draws too much power
and kinda related... I really miss the voltage measurement points on my old R6970 Lightning where you didn't have to solder anything, there was adapter with holes were made to fit regular multimeter probes and it would hold it in place, it was awesome
How come the 16pin power connector didn't melt at almost 600watts?
so what was the maximum board power wattage of it?
It's almost as if they're holding these cards back. Maybe until we get newer equipment to go along with it. Are they worried about their components under too much stress? Or the air cooling system is an adequate enough?
i seen another tech tuber that had a 4 way power adapter into the card giving it the framework to draw more power allowing it to overclock to 133% power
Can memory be overvolted as well?
im just a amateur but i really miss overclocking with just unlimited sliders.. let me melt the card down if i want to 😞
But if you drop voltage then surely the performance can only decrease slightly at some point ?
This video is amazing information and guide for overclockers of RTX 4090 its gonna help my Asus Rog Strix 4090 overclock to 3105mhz
This is some next level stuff my friend YES let them know what hardware is all about !
nice, card looks great, more powa!
How much power can the PS connector supply? It would be interesting to see the VD under load. It might be a good test to see if an automotive voltage stabiliser and battery could help when going extreme with overclocking.
things got a higher tdp than the 295x2 from back in the day...
3000 series was the same with undervolting, i was seeing differences at the same clocks on a different voltage. although not as noticeable in games on my 3080
i have been searching about this topic for the last two hours and gave up , then i go to youtube to find something to sway me away from this and see der8auer uploading this