Overclocking the 4090 should be illegal!

Get your JayzTwoCents Merch Here! - www.jayztwocents.com
○○○○○○ Items featured in this video available at Amazon ○○○○○○
► Amazon US - bit.ly/1meybOF
► Amazon UK - amzn.to/Zx813L
► Amazon Canada - amzn.to/1tl6vc6
••• Follow me on your favorite Social Media! •••
Facebook: / jayztwocents
Twitter: / jayztwocents
Instagram: / jayztwocents
SUBSCRIBE! bit.ly/sub2JayzTwoCents

[@ArchonPain0]

I know Jay didn't say it but EVGA is so lucky it got out, if the FE is better than a Strix, Nvidia is royally Fing the board partners.

[@DaBigBrozer]

Shouldn't the partners git gud? I don't think they should hold Nvidia back

[@buckminst]

@@DaBigBrozer The partners should produce the best cards they can, sure. But it's hard for them to compete with Nvidia when Nvidia gets first dibs on the best quality GPU cores, and RAM. And has control over what the partners can do to their VBIOSes.

[@Liatin1]

@@DaBigBrozer sure they should but nvidia gives them nothing but the board design. No drivers to actually test their designs properly

[@manicdan481]

@@DaBigBrozer The partner cards keep the chips way cooler, so its not like they are slacking. The problem is nvidia charges so much for the chips that there is no margin unless you sell a gpu way over founders msrp. Even better is nvidia is also voltage capping these cards, so they actually cant draw up to 600w without more volts. There really is so little reason to buy a more expensive version of a gpu for performance, only on noise or style or if you are going to volt mod and extreme OC the thing.

[@Fantafaust]

@@DaBigBrozer they can only do so much since Nvidia forces them to stay under certain prices

[@ReveredDead]

It was so cool plugging my computer into the GPU.

[@valienn]

if this is a joke pls explain

[@koksem]

@@valienn High end GPU like rtx 4090 are HUGE and EXPENSIVE af meaning that pc feels like a component for the GPU, a machine that allows GPU to run.

[@valienn]

@@koksem oh ty

[@ReveredDead]

@@koksem Yos took the words right out of my mouth. 4090 is a behemoth.

[@xaviertrese]

@@ReveredDeaddid you used to play cs

[@hellraserfleshlight]

Jay - The error correcting on GDDR6 will happen before you start seeing artifacts. You can't just boost until you see artifacts, you have to run a benchmark after each frequency bump and see when the scores start going down. The errors will affect your scores before you see the artifacts.

[@MaksKCS]

Also him increasing memory by 5Mhz is useless when memory frequency increments go by 15Mhz. At least that's the case for msi afterburner.

[@aboutthat413]

@@MaksKCS My memory goes up a bin every 4 mhz, but my core clock goes up every 15mhz on 3080

[@sergioav7278]

What benchmark program should I use for that? I'd like to learn to push ram frequency until corrections start to happen.

[@jurakerleha]

@@sergioav7278 Best program was for RTX3000 series or even 2000 series Port Royal,I used that

[@SmartMoveGraphics]

Cool sounds smart to me

[@cruzanman37]

If anything AIB should see this as Nvidia is working toward phasing them out

[@fredEVOIX]

after evga, reference design seems to have become "worse design", nvidia keeps the best bins for themselves etc....is a pretty big problem for most of the world because FE are near impossible to buy outside NA and nvidia makes nowhere near enough of them, seems like nvidia are drunk and succes and blind to the fact that AIBs sell 100x more cards than them the ratio right now in my country is 19/0 aib/fe nvidia just says "buy another brand" on their own website never seen an FE available and knows no one who ever got a 3000 FE

[@scarletspidernz]

@@fredEVOIX NVIDIA for sure gonna have their own assembly plant and cooler manufacturing+ distribution hubs/connections within 10yrs. Now that their cooler out preforms AIB's, all that other stuff is easier to get in place and they'll be just like Apple

[@kiptt7082]

@@fredEVOIX 19 to 0 ratio? what country are you in that has only 19 gpu's and 0 math teachers?

[@yumri4]

As he said "it's all RAM" I would be more so into what RAM chips they are using is the issue instead of the assumed by you to be better GPU chip quality on the nVidia card.

[@AeroDr]

Highly unlikely. Nvidia wouldn't want to deal with marketing, logistics, distribution and end user support for the entire market of nVidia GPUs. AIBs do a lot of the legwork in dealing with consumers, distributors and retailers.

[@mrhappy8966]

u have to imagine evga has a 4090 we will never get to see just collecting dust in a warehouse or lab somewhere

[@dancomando]

Someone will sell it. Some day

[@doraafelfedezoofisol]

or they just sent it back to nvidia

[@captainkirk3000]

You don't have to imagine, because they do lol

[@VesperAegis]

EVGA doesn't like money. I'm convinced of it now.

[@remix763]

@@VesperAegis they dont like working with nvidia

[@jondanimal]

I have been a car guy for years and this year stepped into building PC's. Your car reference was perfect and I wish more people could do those types of references for me!

[@zachrichard7039]

He nailed it. Jay must be one of us

[@JayJay-ky4cc]

@@zachrichard7039 he is he has a channel called JayzTwoCars💪

[@SteveDFM]

It went straight over my head, but I'm glad it made sense to car guys :D

[@whuspr]

@@JayJay-ky4cc Jay Leno's Garage

[@Arhey]

Well he drives a RS6, so i assume he is kind of car guy himself

[@Mahi-jg5dx]

I always go with GPU tweak with my ASUS cards. I feel like it makes more sense. I understand MSI Afterburner is still very much what people prefer but GPU Tweak 3 is not bad at all and it always feels more right to use the software the manufacture of the card provided.

[@Sc1Z]

the best part about your car analogy is that if my fuel system in my car fails short ( and it is the weak link) my car changes the ignition based off a multiplier, starting at 1.00 and goes down... my car underclocks its self.

[@Masoch1st]

thats only at the pump tho. if its at the injector, you wont see it fast enough often, unless you have egt sensors in every exhaust runner, and even then, its questionable.

[@Sc1Z]

@@Masoch1st its 100% the pump, the 1.6L ecoboost block and lower end is good for 600HP that I have seen, but the stock fuel pump is only capable to 400 and that's best case scenario, that's getting upgraded next year. its only like a stack to get a new pump, lines, and upgraded injectors. this year was new LSD, S280 turbo, quick spool exhaust ( gotta come off and get wrapped thought, I was impatient) and base tune. BUT you are 100% correct, keen eye for detail my friend.

[@Masoch1st]

@@Sc1Z i have a 1000hp TTRS ;) happy trails friend. be safe out there

[@calebhohneke8482]

@@Masoch1st You getting that 1000hp out of the original engine? Love that the TTRS's are 5 cylinders...Gives them a nice unique flavor/sound. Stoked for ya man that you got such a badass car!

[@dhrida5518]

@@Sc1Z the egnition going down you mean the timing? Also there is alot of ways to get more fuel into the cilinder, higher pressure valves on the fuel rail, bigger injectors, an extra injector rail, plenty of ways, but lets say you want to push more its just a matter of time untill you know the limits, better to play safe than sorry...

[@StripperGlitterMedia]

The comparison to cars on the dyno made SOOOOOOO much sense! I appreciate that..... Keep up the good work!

[@arieatona]

I always love these overclocking videos, please keep doing them! These are some of my favorite videos from your channel. Thanks for covering the 40 series, I can’t wait to see how far you can get!

[@SebLukaTech]

14:30 WARNING 🖥🖱Jay: "See what its doing now? I'm pretty sure that because of the previous run..." Translation: 💣🧨OVERCLOCKING MAY DAMAGE YOUR CARD🔥🚒( He broke "another one" ) ☮PEACE☮

[@hombre_loco]

@@SebLukaTech yeah this guy got for sure for free the cards a normal person got to buy the cards

[@neomoz]

Through my own testing, I found the lower temp problem Jay had is related to the GDDR6X. I was experiencing random black screens on cold system starts, even though my card passed stress tests and benches at +1500 memory offset when it warmed up a little. So with blasting the fans I found once the GDDR6X memory temps fell below 35C, memory starts throwing errors at speeds above 1450mhz, this is consistently reproduceable. It is weird because generally silicon performs better the colder it is, at 24Gbps these memory modules behave weird and need a sweet spot temp to run well, I've found that's around 60c. I run the ram daily now at +1400 offset and it's been solid(23.8Gbps). The FE cards don't cool the memory as well as AIB cards, so that would explain why it did better on memory, it's keeping the ram in that sweet spot range better and avoiding this cold bug problem.

[@ufsteropolstero6014]

Jay's personality has this effect of making relatively boring things like watching other people wait for benchmark results fun indeed. That's a remarkable gift imho.

[@mastixencounter]

🤣🤣

[@_Gnome.]

I think he did liposuction, that's why his body looks so terrible

[@TaintedCamper]

No wonder the board partners are pissed if the fe card performs so well what value add do they have

[@DJJOOLZDE]

RGB

[@TheRikardo91]

@@DJJOOLZDE yeah and overall aesthetics I suppose? All cards look different and different people have different tastes. :)

[@technicalfool]

Well Nvidia don't offer pre-waterblocked cards yet, and apparently FE is a bit of a noisemaker. Plenty of room for something more quiet, or something that doesn't take up every slot in the case. Plus oddly enough some people care about looks, and FE might not come in their preferred shade of hot pink :P

[@FreelanceDev4life]

RGB

[@lycanthoss]

@@TheRikardo91 yeah except I think the 3000/4000 design is the best. Most AIB designs make me cringe. The FE cards look better because they are so simple.

[@chubbysumo2230]

Derbauerr already power modded a 4090 with an Elmore Labs voltage controller, and very quickly hit a power limit, it was only very slightly voltage limited, and volt modding did not really get him a lot of extra headroom. Maybe with colder temperatures you could run more voltage, and he probably will, but it seems that Nvidia is running these right on the very ragged edge of what they're capable of. Or what they've bios limited them to.

[@dontrowbridge3316]

Thank you for the content! Personally I've found overclocking is like stretching. You can work your way up higher stable, But when I turned on my computer the next day it won't boot. I'm guessing that's what you saw with the initial AC run.

[@antivanti9805]

I had an old GTX card, don't remember exactly but may have been back in the 500 och 700 series, that was only stable at maximum overclock when the card was hot. Tuning it while hot would get me to a stable place but if I let it cool down I started getting artifacting when the temperature dropped below like 75 degrees. Run the benchmark for a while and as the temperature went up the artifacts dissappeared. Made it really annoying to find a truly stable overclock for the card

[@Gatorsrok]

Cool

[@AdelaeR]

@@Gatorsrok No, hot.

[@DM16_]

4090 has a digital voltage control so you need elmor's evc to use the i2c. I think LN2 will be necessary for higher frequencies tho :)

[@narcoti]

Wasnt the power limit the new limiter ?

[@tresnugget]

@@narcoti It'll hit power limit with a volt mod but without volt modding it probably won't hit power limits, unlike previous generations.

[@megapet777]

I think its faster with lower clock speed and lower temp because in the 40 series there is bunch of hidden clock speeds too. And I think they might be boosting higher due to the lower temp.

[@jeremybarber2837]

I love the blanket method. It's a whole new world for tech this round.

[@tall1oN]

Funny that you also have cold instability, i have an issue with my 7950X being unstable with -20 curve optimizer on a fresh cold boot, but being stable in warmed up stress tests. I have given up for now but i hope it will resolve with new bios.

[@kdrgaming3344]

I was able to get the VRAM on an EVGA 3080 to +2000 and be stable, a Gigabyte to +1500, an ASUS to +1500, a different EVGA to +1200, an Aorus to +1100, an HP to +900, an MSI to +900, and a Dell to +800. I was amazed at how high you could overclock the memory on those 3080's and the HUGE range of values that different brands of cards would crash at. EVGA was always king at making these video cards as stable as possible. I am going to miss them. Every single one of those cards didn't have heat-pads behind the VRAM going to the metal backplate EXCEPT the EVGA cards. And because of that EVER SINGLE CARD except the EVGA ones would throttle unless they had heat-pads going from the VRAM to the fucking backplate... God damn, I spent a lot of money on fucking heat-pads. I really am going to miss EVGA.

[@jurakerleha]

I think I have tested 10 various RTX3090's from Palit RX3090 GamingPro which could do 1395MHz in benches,in gaming or rendering in Octane1200MHz was 100% stable,then I got another GamingPro and that wouldn't do more than 1000MHz,anything above 1000MHz would crash,replaced pads for Thermalright etc and didn't helped much,temps I see on VRAM max 58C and GPU itself was hitting 38C,then I tested Asus Strix which was one of the best in VRAM OC and core as well,that GPU was running 2245MHz and +1500MHz on VRAM,Nvidia 3090 FE was okay I guess but wouldn't hit more than +1000MHz on VRAM,core fully stable is +180MHz,then EVGA FTW3 was okay only with XOC BIOS,without it would hard throttle and VRAM was not the best at +700MHz

[@anthonyblacker8471]

It looks like it's that time of year already! Can't wait to see some LN2 on the 4090!!

[@MrGringo466]

I said it b4. the voltage slider does do stuff. but only if the voltage/freq curve has been setup in msi afterburner. I can show you what I mean if you wish to get in a chat with me. but it DOES do something. you can also lock freq rather than have it jumping around. there is also a feature to clock based on temp.

[@denbot.cycling]

I had an issue with lower temps on a RX590, think it was down to holding a higher state for longer, it could hold it for short bursts but not a constant

[@praetorxyn]

What I want to see (don't think anyone has covered it yet) is noise and thermal comparisons between the FE and the Strix. I couldn't care less if the FE gets a higher score, I typically buy Strix cards because they run much cooler and quieter than the FE.

[@SSoul0]

so i have the Strix and Im using a 5900x and playing at 5120x1440p at full Ultra/Psycho Raytracing. Card stays below 65C and fans spin around 65% as well. I have my case open because I dont want to stress the new adapter while I wait for my cablemods power cable to get here. Its not loud in the 7000D at least for my purposes. With the glass anel closed I dont really notice it. Coming from a EVGA 3080 10gb, its a world of difference and its specifically why i chose Strix as well. I hated how loud and how hot the card got even undervolted. Running my strix at a power limit of 80% and its still an absolute monster.

[@lawyerlawyer1215]

@@SSoul0 dude , I was about to fall asleep and saw this , and now you have me worried about my connector , should I open my case , the cable is , touching the crystal indeed 😢 To fucking wide this cards 😂

[@SSoul0]

@@lawyerlawyer1215 yea you have to make sure the bend isnt extreme. Cablemods is making a 90 degree connector for the GPU in a week or two so that should be interesting when its out. And its mostly for the adapters, you should be fine if you actually have the single 12v cable

[@lawyerlawyer1215]

@@SSoul0 im using the adapter , I don’t have one of those new psu As for the bend… well it’s not like I pushed the side panel hard till it fitted. But it’s no doubt touching it and applying some pressure. Wish I could upload a picture to get an opinio I think I’m going to open it. Breaking my 2000€ card isn’t in my current plans 😂

[@SSoul0]

@@lawyerlawyer1215 Yea just be careful. Not sure how you use your 4090 but as long as you're not overloading it CONSISTENTLY for long periods of time, you should be fine, its just for an adapter, the bend should not be extreme.

[@The_ViciousOne]

Hey Jay, stupid question (probably) but I wonder, did you check the behavior of video- and effective clock speeds in HWM when OC? Apparently these seem to be somewhat tied to core voltage. Which seems to have a huge impact on performance when undervolting, even tough core clocks are the same... (unlike 30 Series) I wouldn't be surprised if those react to higher voltage as well... Just an idea though... 🤔

[@bitsbfg1810]

Wow that is trippy. Can't wait to see the others come up with videos explaining why this is so different.

[@EnlightMC]

I haven't been this geeked out since I got the 2080Ti, sort of lost interest in the last couple of years. But watching this brings back so many memories

[@cognitiveinstinct2929]

Jay, it could be possible that the memory performing better is making it so the gpu is doing less work. Meaning that the memory or at a minimum the GPUs ability to utilize the memory is the bottleneck.

[@Mr.Morden]

Can't wait for the 4090 Tie to start tripping breakers.

[@ffwast]

At the current rate of wattage creep that would be the 8090ti

[@HanSolo__]

This sure will be TIE Fighter.

[@kyleaschaber]

what a terrific video, well done. Keep it up

[@RWoody1995]

Just had to pause the video in appreciation of that moment at 4:55, I want more tech channels to explain things in terms of what happens when you're on the dyno ;D

[@welcometobobs]

I love when jay compares computer and cars

[@ZMacZ]

Maybe the overclocking mechanism has som sort of stabilization mechanism. What I mean is that if it has to skip a cycle, in order to compensate for errors, it simply runs slower, since skipping a cycle lowers the processing speed. It's like the diff between 6 or 7 clockcycles per instruction on the CPU. If the answer isn't done by 6, it skips to 7. Then if the temp drops, it tries for the 6 again, which then is a boost on top of the overclocking settings. (like 16% or such.), and then ofc it crashes since now it tries to run that much faster, instead of skipping to the next cycle, which then in turn allows the rest to be done correctly as well. You can check for it by comparison in framrates at less or non overclocked speeds, and then calculate what the outcome should be at overclocked rates. If the numbers come up lower for testing, it starts skipping more. Also the most latency between any processing and memory is reading the memory. But most of that is difference between clockcycles. If the memory is at clockcyle like 3000.4 and the processor is at 4000 exactly, the computation can't start until reaching 3001 for the memory, after which the processing unit will have to wait till 4001 or even 4002 as well. Memory can only be read if the bus is already showing the correct result.

[@CmdrSoyo]

afaik the 16Gb Micron Memory Chips on the 4090 and 3090Ti have negative cold scaling. at some point you will have to place a memory heater on the card or try to increase the memory voltage to heat them up more (they might also scale with voltage and run a bit faster) i'm also wondering if the FE is using higher spec memory modules than the AIB cards like that one 6900XT did where it had something like 17Gb/s modules while the other cards had 15Gb/s.

[@nellynelson9387]

Hey guys great video as per usual. Optimum tech did a good video on undervolting and found video hz speed dropped but the core clock said it was the same. He was using hw monitor and 3 different core clocks to look at. Could pushing the ram or the core be causing the voltage to drop to low in one or the other? Does that make sense?

[@germanmade1219]

His undervolting method does not work with 40 series cards so his video is very misleading.. As other people have stated, had he used the proper method (just reducing overall power target slider in msi afterburner) it would have a lot better results. Techyescity does a video where he undervolts the 4090 properly.

[@PryvatCyan87]

What cant it run straight out the box? These things are beastly and I hope AMD stays with them with the 7000 series

[@AnIdiotAboard_]

If what Enterprise OEM's are to be believed next gen AMD Compute and GPUs are 60% less power with a massive and i mean massive FP64 advantage. And for us in the datacenter we itching to get our hands on them. Could finally see the ousting of NVIDA from the datafloors.

[@j7powndandtrolled]

i literally hate this gpu just because they FKING colored it blue and red. Like really asus i thought we where don coloring cards with fixed colors

[@eitanmalul4877]

@@j7powndandtrolled don't buy it yet.. everyone knows that RGB version is best because it adds FPS and fixed card colors reduce it.. so wait for it's release.

[@ron200088]

@@j7powndandtrolled Yeah, so do I. Wanted to buy 2 of them to hook them in SLI, because they’re so affordable. Found them for $2500, a real bargain. But the colour scheme made me decide against buying them. Heck, probably would have gone for quad SLI if not for the colouring scheme. ;)

[@shadow105720]

@@ron200088 I thought they didn't have the nvlink

[@1Dshot]

Always nice to see your process

[@jonahswim]

My dad builds servers and once told me that your biggest enemy is time for data to move like length of cables etc. seems negligible but it can add up. I wonder if the low temp is contracting the board decreasing distance data travels.

[@TheJackelantern]

Watching these videos is the closest I will ever come to a real gaming PC.. These are just insane! I am still on my rig from 2009 haha

[@warrax111]

I am still on Geforce 210. :D Well, it serves good as display adapter.

[@TheJackelantern]

@@warrax111 I am on a Radeon HD 4550 .., probably not much better. The best game I can play is half-life with medium settings. :) .. I can't keep up with the gaming industry..., beyond my means. Though a lot of the older games are really and truly gems.

[@tonymorris4335]

@@TheJackelantern You should be able to pick up some cheap used hardware that would be a big jump up. An RX480 or something should be reasonable and a huge improvement. You don't need to chase the bleeding edge to still run modern games.

[@TheJackelantern]

@@tonymorris4335 At some point I might upgrade. Though anything 2018 - 2022 is beyond my price range on the mid-highend. If I upgrade it will have to be an entirely new build from the ground up as I don't see the worth in the options currently available for my existing system. Even an i7 2700K would be a huge CPU jump for me lol

[@timothysmith160]

Stop spending all your money on butt plugs and you will soon have enough cash for a new gaming rig!

[@lesabre1972]

I love watching you do this it's so great

[@videocruzer]

Modern Physics has that, the very hi voltage and frequencies opening and or expanding. love how you made that transition.

[@Idlehampster]

I miss the days when this channel was smaller than Barnacules and video titles were clear.

[@EyesOfByes]

I wanna hook this up to a waterblock and put the radiator outdoors during a nice Swedish winter. Yummy

[@jeffro5032]

Haven't built a new gaming rig since 2013! Still running an OC 2600K / AMD R9 Fury / Sabertooth 77 board. ( Replaced dual 7970's for the R9 in 2016 ) I see some nice stuff finally dropping in price , where I might finally build a new rig later winter time. Going all AMD though I believe.

[@allenbrodess8510]

Lower temps you can do more work with same electricity.. a lightbulb in the pole requires measurably less power for the same brightness graphics cards are no exception.. voltage also tends to be more stable and even microfluctuations can be detriment to max sustianable workload . Part of why i stay to my sub 60c rule..

[@YippingFox]

I honestly wonder if you were to underclock the core and then possibly clock the memory higher, would it lead to a higher score?

[@iikatinggangsengii2471]

pointless on nvidia, iirc its more sensitive to core oc, recall my 460 can oc to 880/more i dont remember (from 725) and its like a new card

[@Make-Asylums-Great-Again]

No

[@pixels_per_inch]

Based on my experience, higher core and lower memory gives the best results

[@Cheeseypoofs85]

i watched steve's teardown of the FE and it looks like its top notch build quality. thermal pads and TIM are all like perfectly applied. no glue and actual metal frame. i was always anti-FE, until the 40s finally upped their game

[@sp00n]

@@michaelmcconnell7302 And with a LPL reference 🤩

[@ERROR-hf3wx]

Nightmare to teardown, with plastic clips on the side, to get to the other side of the pcb.

[@sp00n]

@@michaelmcconnell7302 Lock Picking Lawyer. Go look him up, it's fascinating (and frightening at the same time).

[@sklls6718]

unrelated but I've recently noticed that I watch every video this man posts and shows up on my feed, but I never actually listen to what he says or watch the video. I just put his videos on and walk around the house doing stuff, using the video as background noise lmao

[@raywashington1029]

Thank you for what you do, Jay. Did you use gpu tweak 3 with the FE card or is it only for Asus cards?

[@CyberJedi1]

I wonder if it makes a big difference in games these overclocking numbers, if it does I might try overclocking mine, right now I'm just using the silent bios and chilling at low temps with plenty of fps to spare. I actually undervolted and got the same performance at 300-320W, I'm very happy.

[@interlace84]

Were you also happy paying what you did for that GPU though? Over here it's above $2000 for a single gaming pc component.. even if I had that, there's no way I'd be "very happy" with those prices.

[@eduardosanchez7827]

Nowadays it’s not worth to OC the GPU, it’s too much hassle to get like 5% extra fps while loosing a lot of stability. Once you get over 120 fps it’s very hard to see any difference in gameplay (wich doesn’t mean you shouldn’t get a high refresh rate monitor)

[@SSoul0]

@@interlace84 If he has the means to and he wants to why not? Im very happy with my Strix 4090 as I can push all the frames at 89% of 4k and not break a sweat and my 3D renderings and renderings overall are quicker than before. This card, provided i take very good care of it, is going to last me YEARS!

[@mcgman8058]

the difference in gaming is minimal at the expense of gaming stability.

[@interlace84]

@@SSoul0 that's quite the gender assumption you're making there with OP's avatar and all 🤫

[@TheHarleyEvans]

i feel like the strix was designed to run horizontally (with the Mobo vertical as it would be in the case) something as silly as it's orientation could be the key factor in it's cooling capabilities (as it's vapor chamber)

[@mcgman8058]

condensing, evaporation and the phase change are going to absolutely ignore gravity. Just like if you stood an industrial chiller vertically it would absolutely care less. Its going to follow the thermodynamics, not gravity.

[@TheHarleyEvans]

@@mcgman8058 the convection part of fluid dynamics is a thermal dynamic function which relies heavily on gravity, those fin stacks need to be place opposite the directional force of gravity to achieve maximum thermal transfer between the vapor chamber and the fins. and they can only achieve that if the card is oriented in the right direction

From my experience, temperatures may affect how voltage regulator actually behaves (on both CPU and GPU) which may cause crashing with lower temps. I had this happened a few times. Hot silicon, rock solid. Cool it down a lot, crashes all the time. If you are able to heat it up without crashing, rock solid again but as soon as it goes to idle and cools down, it may crash again. For sure VID or just ASUS voltage regulators take the temps into account and the voltage you see might not be exactly what you have on the core. By the way, I have no clue if that works, but wouldn't Ctrl+Shift+Win+B (GPU restart) be enough to flush corrupted memory?

[@catnaptv_plushshow]

I love watching Jay and co it’s like hanging out with the boys

[@Hi-levels]

Could it be the shrink effect from low temps with 5nm node?

[@ffwast]

I get a shrink effect from low temps too.

[@gerrya4818]

@@ffwast I WAS IN THE POOL!!

[@benwhite4554]

That card is comically huge.

[@VesperAegis]

"And what of the rest of the computer?" Senator 4090 RTX Palpatine: "I AM the computer!"

[@aos32]

I think I would rather go for a basic RTX 4090 model if I was going to buy one. You could get a quite decent watercooling loop for the price difference that the ROF strix version costs.

[@mikeg2491]

Yea it was humorous when I was building out a prebuilt and there was absolutely no room for any other pci-e cards if you added the 4090

[@999jay999]

@@mikeg2491 We going backwards, computers will need to get larger to accommodate the GPU alone... Future 6090 / 7090 the GPU will be a huge Desktop Tower alone..

[@WyattOShea]

The opposite of my pp 😢.

[@willfullyinformed]

I ended up with Core clock at 125hz, and 2,000 Memory clock. Pretty satisfied considering it's just air cooled in my PC, and I've never had a card get anywhere near 2,000hz on mem. Great video!

[@cl11334]

Nice results! What gpu do you have?

[@Aaron-zl5gq]

It’s to bad the FE cards are impossible to get, pretty much only through Best Buy here which had 0 in stock day one

[@troyyerman7282]

What you were saying about lower temperatures allowing more performance from a given clock speed is actually true and an accepted scientific fact, fundamental in the field of electromagnetism. Lowering the temp of a conducting material reduces its electrical resistance, e.g. the super conductor train in Japan uses wires and coils which have been cooled to extremely low temps. Very astute!

[@Wub892]

Air gun and shop air noticed got extremely cold on my fingers. Loud but I wonder how cool you can get an air radiator with shop air carefully placed

[@eternalbeing3339]

A graphics card being that big should be illegal.

[@Necroknix]

I remember I originally subbed to this channel because Jay was a car guy and a PC guy, by the way I liked your analogy. I wonder if most car guys are PC gamers and just don't know it yet.

[@lanlockpcgaming6292]

nice to seee you having fun doing what you like for a change

[@charlyb.9406]

msi combustor with activated artifact scanner is a nice tool for finding the vram limits. should be around 1400-1600mhz. and also i use nvidia inspector tool to reach highest possible clock speed. for the highest limit you need a binned gpu reaching out of the box a 220 mhz or more overclocked curve which leads to stable 3dmark peaks over 3050 Mhz (air cooled).

[@jackhansen7305]

Is the strix using large thermal pads like the tuff is? Do you track the ram to see if it’s thermal throttling?

[@TheOmei]

Would love to see Kingpin on the board with his one off EVGA 4090.

[@yanshaoumo6076]

same here

[@DrivenKeys]

I always enjoy watching Jay overclock. He's what convinced me to get 3D Mark and have fun with it. I want to buy your stuff, but I need tshirt colors other than black. I'll buy two deskmaks if you make them for left handed mice. Until then, keep up those incredible ifixit ads.

[@RegazozoGaming]

I saw this exact same behavior on my 6900 XT. Ended up setting a 2525 Max Clock with a 1135 voltage 2150 ram speed and maxed the power slider also slightly increased the fan curve. In Timespy the clock sits right around 2460 the ram at 2138 pulling around 1080-1100Mv and gives me a GPU score of 21800 +/- 50, temps were round 62-65 GPU and 86-89 hotspot. Upping the clock speed or voltage anymore than this spiked the temps and my score ended up dropping 600-800 points. As Jay said something else on the card is benefiting from the lower temps and that is giving better performance boost than core clock.

[@Bobmarleytronjr]

Also I like how you use car analogies to help explain

[@savifrank6323]

Debau8r was able to unlock the voltage and the card immediately became power limited

[@themagicjules9489]

The GPU Tweak is gonna be acting differently in comparison with Afterburner, also the GPU Tweak has different voltage profiles maybe giving a try to that might have change the results. The 0.050mV on the Performance mode might increase something between 30-60mhz to the clock, that’s interesting considering that the FE is somehow at the same level I believe than it is safe to assume the Rog Strix P-Mode isn’t a really P-Mode it seems a kind of “placebo”. Also, the voltage fluctuations not always make things crash it will be for sure depending on how voltage is regulated on the PCB. Anyway…

[@darknase]

0.050mV = 0.00005 V You probably mean 50 mV or 0.05 V Millivolt (which mV is the abbreviation of) just means 1/1,000 of a volt and is chosen as unit in the UI as it gives more granularity and readability (b/c without separators).

[@12Burton24]

The really bad thing is water cooled card with preinstalled blocks are still more expensive even they dont have expensive vapor chambers and th air cooled cards picked up with material they use for the giant air cooler.

[@bipbop3121]

I recently got that water bottle (Odwalla), also in marshmallow. I still spill occasionally tipping instead of sipping. Thicker doesn't really help it in cup holder, still need to test too loose cycle cat cage, but built stays better than polar & I save strap down for massive bottles Blackburn, so eh. Taking a long time to not taste odd at end of water, though cleaned well. Not as cool as long as my Built. Only advantage so far is metal still better than my polar and with the handle down, it's easy less likely to land in a way that'll result in open water spill compared to Built.

[@lexi-ru9sz]

true only thing is it needs to be changed when it starts tasting sweet

[@Hardwaretuned]

Yes sir Jay I like your car references and I hope ppl actually understand what you are talking about. Luckily most of newer ecus these days will pull timing and cut power on a fuel pressure loss or sudden spike.

[@emoluv54865]

I liked the ending

[@yournotdatguypal]

Jay you da man! 💯💪🏽

[@noahkramer9797]

Thanks for the car analogies

[@MaciekEm13]

Imagine the score if Jay had managed that cable mess on the desk 🙃

[@AndrewAffolter]

Imagine being able to actually buy a 4090 not from scalpers. Especially an FE.

[@alpenfoxvideo7255]

FE is the only 4090 that fits in 3-slot SFF cases and while still cooling very well, it's tempting a lot of buyers

[@HanETG]

looking good jay!

[@patrick_goldman]

What might be an issue on a 4090 that causes 3DMark to freeze the whole computer at the end of a benchmark? I've had it go through the entire benchmark process and force me to hard restart my PC at the point when 3DMark should be showing the scores.

[@equinoxe3d]

DerBauer's review showed the 4090 being far over its optimal point on the power curve so I'm surprised at how much headroom is still left to overclock. It was surprising how well it ran at 60-70% power limit, which was 100W less power and could make using one in a SFF system somewhat manageable.

[@Alex-zi1nb]

lol what SFF system is a card that size fitting in??!!!!!

[@changen4125]

@@Alex-zi1nb The PCB is tiny. Watercooling it is just fine. It becomes a 1 slot, half length card. lol Even an Aio version will fit most SFF cases.

[@equinoxe3d]

@@Alex-zi1nb Yeah pretty much no SFF in its current form, but at 100W less it could have used a smaller 2-slot cooler. It's just weird seeing nVidia redlining it out of the gate, especially if a Ti version comes out later

[@ImotekhtheStormlord-tx2it]

4090 is just an overclock toy but nvidia fanboys wont agree with me. the card itself has 0 logic. big as fuck, defective adapter, large TDP, loud, easy to bottleneck, pricey....

[@gultar4513]

Jay you keep saying these cards are voltage limited and they are not. Watch Der8auer's video. If you raise the voltage even 50mV above what the software lets you do you run face first into the power limit. He is currently working on a Power Mod for his card so that he can push it beyond the 600W max power.

[@scottw1871]

Love your videos and about to build a high end PC in Australia and wondering would you go AMD 7950x or Intel 9 13900k? Was looking to go AMD 7950x with a Gigabyte 760e Aorus Xtreme and a Gigabyte GeForce RTX 4090 Aorus Master. Any assistance would be appreciated.

[@fragalot]

I've always noticed since the days of the GTX 900 series, that the Heaven Benchmark is more forgiving than a game is. I can set a VRAM and GPU overclock that's super stable in Heaven , or Valley Benchmarks, but once I load in a game (even an older one from that era), glitches and crashes galore.

[@EinSwitzer]

Get that chip to 92c to 93c it starts to show weirdness , then cool it so block the fans and force a overheat then let it cool itself slowly as a while

[@grxvytrain5794]

Waiting on my 4090 went with the liquid cooled MSI

[@DaleFCB10]

Nobody likes a showoff my guy

[@Dji00]

@@DaleFCB10 fuckin chill.. he's just stoked. Let him be. You sound bitter.

[@JohnDoeC78]

Same, waiting for the Suprim X to be in stock.. Pre orderd 👌

[@EchoedR]

@@DaleFCB10 commenting that they got a product the video is about is not showing off

[@grxvytrain5794]

@@DaleFCB10 Lol! Show me where Nvidia hurt you.

[@stunhades1842]

Jay knows car stuff man hes the goat!

[@alexandergrace]

On my 3090 i noticed more stability and higher scores lowering memory to free some voltage for higher core OC.

[@fahadal-asmari6893]

Now I know why EVGA left that makes way more sense than anything.

[@whitewizzard5245]

What do you know about EVGA leaving?

[@Iwhoiam999]

@@whitewizzard5245 Nvidia is actually making good cards now as far as thermals and performance. Basically phasing out 3rd party vendors. Nvidia can sell their cards for far cheaper too. 3090ti is a good example of that

[@mcgman8058]

I do not think it's the ASIC quality that made them leave, I believe it was NVidia's price capping and them losing a couple hundred dollars per card in this saturated market. you cant compete with with a company like ASUS, who can lose tons of money on their flagship and make up the losses with marketing more profitable products.

[@fahadal-asmari6893]

Like what is the point of 3rd party vendors at this point?, That's what I'm saying and EVGA already knows.

[@whitewizzard5245]

@@Iwhoiam999 Thanks Alen. I'm new to the GPU seen so thanks for the information =)

[@basictechgirl9848]

I tried overclocking and getting at best top 10. I did for the 2070 super for awhile but for my 3080ti I got a bad card. I can't even break 13000. These 4090 cards hitting 28000 to 30000 is crazy.

[@pastafazool5951]

My trio is hitting 2980 with just oc scanner OC 😆 these cards are insane indeed

[@craig9365]

Do you find this fun? I'm not sure I understand why one would waste their time with this? But I'm also viewing this from a "does this increase the performance of the card" and it seems to be overall "yes" but the yes is so small that the effort and time to find it being stable isn't even worth it. So is this just one of those I'll never understand type of things?

[@metalhead4404]

@@craig9365 I get what you’re saying, but you need to think of it this way… Imagine a 4090 on the shelf for $1500 and it’s average FPS in 200, next to it is another 4090 selling at $1500, but it’s average FPS is 205, which would you choose? The overlocking is simply increasing your cost to performance ratio. It’s a no brainer.

[@ahmedhamm5464]

@@metalhead4404 also while making the life of the card a bit shorter so its somthing the think about

[@metalhead4404]

@@ahmedhamm5464 Any piece of silicon comes with a separate life expectancy. No one knows when it will die. I upgrade my main PC every 3 years. Never had issues. If you’re a person buying silicon at these prices today, you aren’t concerned about keeping it for 5+ years, because software alone will simply outpace it. And if you are concerned then why are you buying a $1500 GPU in the first place. Also, that’s what warranties are for.

[@QtheMisanthrope]

@5:09 my man just compared constant voltage on a GPU to a return style fuel system on a car lolll I love this channel.

[@christopherlenahan3906]

With the IPC/Watt efficiency of the 4090, could a moderate performance gain be achieved without making a space heater out of my pc? Or is that within the realm of the card running with G/Free-sync on and fps limited? It has been said, "don't undervolt the 4090." but what happens if you only want 144/165fps.. I'm all about cranking my hardware to see what my paste/w/c/settings can do, but my 5800x3d/6900xt watercooled system runs everything 165fps at ultra or near ultra and I'm actually looking at increasing efficiency to reduce fan speeds. Just a video topic question maybe, "upgrading for a quieter system".

[@potatopotatopotatopotatopo8746]

I absolutely love Jay's overclocking videos. I wish he kept overclocking after his 3090 kingpin card died from being lapped

[@roughprecisionplayz-6434]

I took a lot of time and care putting it together and it runs really well

[@do_minguez]

can you test the gpu tweak vs afterburner on the FE card also? would like to see if asus realy add something to its software or its realy only a difference with asus cards.

[@alex-gx6bs]

5:15 i never knew he was knowledgeable on cars like that

[@saian123]

he has a racing car or two- there's a video where he shows off his home where you see that.

[@GamerErman2001]

Can't wait to see Newegg bundles that contain a 13900K or 7950X, a 4090 and a window unit air conditioner. Meanwhile in Canada people will be selling their space heater to buy a 4090.

[@wolphin732]

I have a 3080 TI... and room a little cool? start gaming or process a few panoramic photos... My space heater is likely to just collect dust, and I can have a cool room at night.

[@christianponopp8756]

Put a 3000VA APC smart-UPC unit before your PSU to smooth out the power from the powernet. A huge machine starting next door could drop voltage and causing the pc to crash while chasing overclocking records.

[@ygdefine322]

Maybe your temps before were making power draw not matter, but with the lower temps the RT cores want to ramp up more and through lowering core clocks you allow more power for the RT cores. Which have always been the bottleneck.