NVIDIA Details GeForce RTX 4090 Founders Edition Cooler & PCB Design, new Power Spike Management

[sepheronx]

I'm just far more interested to see what AMD will showcase. But I doubt prices will be decent.

 

[trsttte]

I adjusted the scale with my instinct
Here it is

View attachment 262533

If only oscilloscopes had a read out of the scale being used on screen that could be shared so easily instead of mis-aligned graphics with vague statements about one peak being bigger than the other, fucking nvidia

Like, not even the time scale seems to be the same, so what the fuck is that graphic supposed to tell me!? Oh smooth curve vs wavy curve with peaks, much control very wow

 

[pt0x]

If only oscilloscopes had a read out of the scale being used on screen that could be shared so easily instead of mis-aligned graphics with vague statements about one peak being bigger than the other, fucking nvidia

Like, not even the time scale seems to be the same, so what the fuck is that graphic supposed to tell me!? Oh smooth curve vs wavy curve with peaks, much control very wow

This...

And btw, even if it's anywhere near true they must have found a magic trick because I dont see any caps on the board near the VRM.

Might be on the back, but that's less efficient

 

[Cheeseball]

AMD drivers work great. Your comment is valid 12 years ago.

More like 2/3 years ago (late 2019, early 2020), being a former 5700 XT and RX 580 owner myself. Currently the drivers do work quite well, aside from the Radeon Chill and Enhanced Sync features.

 

[GunShot]

looks exactly the same. I don't see the "redesign".


Good luck with AMD drivers

*sighs*

Are we, REALLY, still saying this in late 2022?!

*NVIDIA says (in the joker's voice) "wait until they get a LOAD of me:"

Last driver update - GRD(?) 516.94


Open Issues

- [RTX 30 series] PC monitor may not wake from display sleep when GPU is also connected to an HDMI 2.1 TV, and the TV is powered off.
-Toggling HDR on and off in-game causes game stability issues when non-native resolution is used.
- Videos played back in Microsoft Edge may appear green if NVIDIA Image Scaling is enabled upon resuming from hibernate or booting with fastboot.
- [DirectX 12] Shadowplay recordings may appear over exposed when Use HDR is enabled from the Windows display settings.
- Monitor may briefly flicker on waking from display sleep if DSR/DLDSR is enabled.
- [RTX 30 series] Lower performance in Minecraft Java Edition.
- External display may not be detected when connected via USB-C on certain Razer notebooks.
- [Forza Horizon 5] Rainbow like artifacts in game after driver update.
- [Reallusion Hub] App will crash when launched on PC using a CPU with 32+ logical processors
- On rare occasion, video playback in browser may result in bugcheck code: 0x116
- [Marvel Spider-Man Remastered] Cutscene edges may appear blurry when DLSS is enabled on an ultra-wide monitor

And a few more extra issues were added to this list much later (see NVIDIA's forum) to this embarrassing column and... NVIDIA has chosen to LIE about fixing outstanding issues, that goes all the way back to Ampere's released date, as being fixed (the LIE) just to remove these huge issues from its HUGE GROWING inferior problems. e.g. HDMI eARC issues, etc. Oddly, no Ampere "professional reviewers" ever mention any of these serious flaws in their early glorifying Ampere's reviews. Weird! /s

So, please... pump the brakes with that fanboyism!

 

[Icon Charlie]

looks exactly the same. I don't see the "redesign".


Good luck with AMD drivers

I have had ZERO PROBLEMS with my AMD Drivers on my RX 5700 and 5700XT for 2+ years. I'm not defending AMD as I know for many years that when they launch their video cards, their drivers are normally not optimal. However they have the reputation of continually working on their drivers making older video card relevant.

Because of that fact my RX 5700 is running better and doing more than now than 2 years ago. But even when I purchased the cards, I had ZERO PROBLEMS with it.

 

[Crackong]

If only oscilloscopes had a read out of the scale being used on screen that could be shared so easily instead of mis-aligned graphics with vague statements about one peak being bigger than the other, fucking nvidia

Like, not even the time scale seems to be the same, so what the fuck is that graphic supposed to tell me!? Oh smooth curve vs wavy curve with peaks, much control very wow

Unless they also framed the oscilloscopes settings in the picture it will always be some room for manipulation.

 

[Slizzo]

HAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA...........

$1600....

You now have my official permission to tell the jacket man I said "GTFOH", "ROTFLMAO" & oh yea, add in a good ole "BMA", you know, just in case he don't quite get the other 2 !

​

4080s aside, I would say only $100 increase in price 4090 vs. 3090 is actually a pretty good deal considering all that has happened since the 3000 series: inflation, NVIDIA coming back to TSMC (which would lose them their preferred customer status), and TSMC saying they're ending bulk order discounts AND increasing their costs.

Oh, and Micron increasing costs of GDDR6X as well.

 

[Tek-Check]

The use of recycled PCB also adds to explanation as to why they did not bother installing DisplayPort 2.0 ports in this gen of cards.

 

[1d10t]

Friend of mine had RTX 3080 FE complained about thermal throttling, and had to do extra mile re-paste and change to a better thermal pads. And with this same card design and higher TGP, I'm sure they'll know what to do.

 

[Dirt Chip]

If only oscilloscopes had a read out of the scale being used on screen that could be shared so easily instead of mis-aligned graphics with vague statements about one peak being bigger than the other, fucking nvidia

Like, not even the time scale seems to be the same, so what the fuck is that graphic supposed to tell me!? Oh smooth curve vs wavy curve with peaks, much control very wow

You are right, it is somewhat difficult\confusing graph to read.
But you shouldn't get so angry or at least if you want to bash them, do it on a matter that actually done wrong (it`s easy to find) and not against a valid graph (that you don't understand). It`s juts weakens your stand point and might color everything you say as bias.

 

[steen]

The use of recycled PCB also adds to explanation as to why they did not bother installing DisplayPort 2.0 ports in this gen of cards.

PEG_16x 4.0 yes, DP 2.0 not so much. DP 2.0 is ASIC limitation. I have a theory about why both are previous gen.

 

[taka]

PEG_16x 4.0 yes, DP 2.0 not so much. DP 2.0 is ASIC limitation. I have a theory about why both are previous gen.

Like that 40 series is the same 30 series? Just a die downscale from 8 to 4 nano where they can cram double transistors on same surface, overclocked and re-badged?

 

[Bwaze]

Cryptomining called. It's not going to pay the bills.


And regarding "new Power Spike Management" - forcing buyers to purchase new PSU is not "management".

 

[steen]

Like that 40 series is the same 30 series? Just a die downscale from 8 to 4 nano where they can cram double transistors on same surface, overclocked and re-badged?

Whatever design Ga 7nm TSMC was supposed to be was shelved for Tu refresh -> Ga on 8nm Samsung. This is due to requirement for separate fab rtl teams. Ga 7n TSMC morphed into Ad 4nm TSMC & they retained Ga->Ad pad/pcb compatibility, PEG_16x 4.0 & DP 1.4 blocks. It's tenuous, though.

 

[Crackong]

View attachment 262552

You are right, it is somewhat difficult\confusing graph to read.
But you shouldn't get so angry or at least if you want to bash them, do it on a matter that actually done wrong (it`s easy to find) and not against a valid graph (that you don't understand). It`s juts weakens your stand point and might color everything you say as bias.

I scaled it with MS paint so I added the FAKE there just for fun.
It was not meant for any serious calculation, just you know, 350W to 450W scale up assuming it is the 12V rail.

Don't Bash me for a 30 sec MS paint job

 

[londiste]

If only oscilloscopes had a read out of the scale being used on screen that could be shared so easily instead of mis-aligned graphics with vague statements about one peak being bigger than the other, fucking nvidia

Like, not even the time scale seems to be the same, so what the fuck is that graphic supposed to tell me!? Oh smooth curve vs wavy curve with peaks, much control very wow

Time scale seems to be very slightly worse for the 4090. Current scale is off but not by all that much, power limits are 350W for 3090 and 450W for 4090 and graphs are less different that that.

However, the context of the charts are exactly the (big) peaks. RTX 3090 has more and they are higher.
Peaks relative to sustained power use is where the problem lies.

 

[Vayra86]

AMD drivers work great. Your comment is valid 12 years ago.

RDNA1 begs to differ.

If only oscilloscopes had a read out of the scale being used on screen that could be shared so easily instead of mis-aligned graphics with vague statements about one peak being bigger than the other, fucking nvidia

Like, not even the time scale seems to be the same, so what the fuck is that graphic supposed to tell me!? Oh smooth curve vs wavy curve with peaks, much control very wow

We saw a similar one with Pascal, somehow they managed to create new spikes to fix

 

[Chomiq]

Is that a 4slot cooler??

At least 3.5

 

[trsttte]

View attachment 262552

You are right, it is somewhat difficult\confusing graph to read.
But you shouldn't get so angry or at least if you want to bash them, do it on a matter that actually done wrong (it`s easy to find) and not against a valid graph (that you don't understand). It`s juts weakens your stand point and might color everything you say as bias.

They themselves said the peak was lower on the 4000 series and tha power level was higher so the alignment you made doesn't really match. Also without numbers attached that can be literally anything, like reduce vertical scale and apply a 10x math gain, there goes the peaks and you get the "noisier" thicker line, not saying they did that, but not saying they didn't. What are they even showing, power up? A quick calculation/load?

Posting a graphic with no numbers shows nothing to anyone not willing to thrust their marketing claims like a new "PID controller feadback loop" to control spikes, how were you doing the power conversion before? If you want do marketing claims do marketing claims, if you want to present data present data, just don't try to do both and mix science and marketing together.

 

[Jimmy_]

interesting... now just see what RDNA3 as to offer with their 7xxx series cards.
That fan and size is freaking me out ... Now System will build around the GPU lol

 

[Dirt Chip]

They themselves said the peak was lower on the 4000 series and tha power level was higher so the alignment you made doesn't really match. Also without numbers attached that can be literally anything, like reduce vertical scale and apply a 10x math gain, there goes the peaks and you get the "noisier" thicker line, not saying they did that, but not saying they didn't. What are they even showing, power up? A quick calculation/load?

Posting a graphic with no numbers shows nothing to anyone not willing to thrust their marketing claims like a new "PID controller feadback loop" to control spikes, how were you doing the power conversion before? If you want do marketing claims do marketing claims, if you want to present data present data, just don't try to do both and mix science and marketing together.

You are right, it is presented not fully corect.
But having higher power level (say current) and less variance (less peak\dips) is their Claim and the graph show it pretty good.

 

[Bomby569]

At the end of the day Nvidia will keep their 75% market share (excluding igpus) and this things will fly off the shelves. Next gen will be even worst.

 

[Bwaze]

That fan and size is freaking me out ... Now System will build around the GPU lol

 

[Chomiq]

It's weird that they're hiding the holes for support brackets that are located at the end of the GPU. 30 series had them (covered with screws), this one seems to have them too but they're covered up with something.

Seeing how thick this cooler is it sure could use some support.

Although I only saw one case from Silverstone that supported it: