Nvidia Said No.. I said YES.

[AlexTheGreatish]

Nvidia have removed software voltage control from their newer GPUs... but we can get it back! Discuss on the forum:

 

Buy an ASUS ROG Strix GeForece RTX 4090 OC Edition GPU: https://geni.us/BDqVW

Buy an EKWB EK-Quantum Vector 2 FE RTX 4090 GPU Cooling Block: https://geni.us/FmEEz

Check out the ElmorLabs EVC2SE Voltage Control Board: https://lmg.gg/CvThC

Purchases made through some store links may provide some compensation to Linus Media Group.

 

 

 

[Shimejii]

That poor GPU  Surprised no magic smoke

[Needfuldoer]

Ah yes, the Tim Taylor approach to overclocking!

[aaronfranke]

The Corsair K100 Air Wireless Keyboard link in the description points to an amazon.co.uk link for a Logitech MX keyboard with "corsair k100 air" in the search bar.

[Ottoman420]

Does this mean we get two videos today in Europe? woooo

[terroralpha]

just an FYI, the VRMs of the asus 4090 strix and msi 4090 suprim are more than capable of handling 1000W to the core. the VRMs on these two cards are overbuilt to the point of absurdity. the two cards are almost identical, but the MSI is slightly crazier than the strix, with 26 phases going to the core compared to the strix's 24. 

 

if there is a limiting factor in feeding 1000W to the GPU, i assure you that it's not the VRM. 

[AnonymousGuy]

The most I ever hit on my 4090 was 1040W running furmark and just fucking around trying to blow it up.   2x current shunt (577W x2) and some Gigabyte 600W bios.  Probably like 1.25V.  The Strix 1000W bios doesn't fully work on MSI and kills one of the displayport outputs, and it didn't exist in the wild until a few months after this screenshot.

 

 

But yeah there's really not that much point.  4000 series isn't power limited like 3000 series was.

 

The 1000W BIOS does go further than shunts do, however.  There's secondary current sensing going on that aren't controlled by shunts .

14 minutes ago, terroralpha said:

just an FYI, the VRMs of the asus 4090 strix and msi 4090 suprim are more than capable of handling 1000W to the core. the VRMs on these two cards are overbuilt to the point of absurdity. the two cards are almost identical, but the MSI is slightly crazier than the strix, with 26 phases going to the core compared to the strix's 24. 

 

if there is a limiting factor in feeding 1000W to the GPU, i assure you that it's not the VRM. 

Correct the Strix is limited to 1600A...like 2000W on the VRM.  The 16 pin connector will melt before that.  (and that's just for the core it takes another couple hundred amps to the vram).

 

[Kweet]

How much performance gain was seen from this?

[dogwitch]

2 hours ago, Needfuldoer said:

Ah yes, the Tim Taylor approach to overclocking!

what time is it?????

ITS TOOL TIME!!!!!!!

[Gucky]

Everything they did, I can do with my STOCK 4090 Founders. Powerlimit on 600W and voltage to 1,1v (+50mv) and it runs over 3Ghz (effective clock)... with +250Mhz core and +1250Mhz on memory... and still cool on air within my SFF Nr200P case.

All that with a 750W PSU :D

[Rajackar]

3 hours ago, Gucky said:

Everything they did, I can do with my STOCK 4090 Founders. Powerlimit on 600W and voltage to 1,1v (+50mv) and it runs over 3Ghz (effective clock)... with +250Mhz core and +1250Mhz on memory... and still cool on air within my SFF Nr200P case.

All that with a 750W PSU :D

Same here on an MSI suprimx model. Video is a bit of a dud to be honest. 1000watt in the thumbnail and they never come close to that. Even with a chiller and an external voltage controller they barely exceeded what any 4090 can do.

Good job getting the power consumption up without any performance gains over a "normal" 4090.

I love the crazy projects they get up to on LTT but this one really misses the mark.

 

On a side note, any idea which settings they ran the benchmark on?

[AlexTheGreatish]

13 hours ago, Gucky said:

Everything they did, I can do with my STOCK 4090 Founders. Powerlimit on 600W and voltage to 1,1v (+50mv) and it runs over 3Ghz (effective clock)... with +250Mhz core and +1250Mhz on memory... and still cool on air within my SFF Nr200P case.

All that with a 750W PSU

10 hours ago, Rajackar said:

Same here on an MSI suprimx model. Video is a bit of a dud to be honest. 1000watt in the thumbnail and they never come close to that. Even with a chiller and an external voltage controller they barely exceeded what any 4090 can do.

Good job getting the power consumption up without any performance gains over a "normal" 4090.

I love the crazy projects they get up to on LTT but this one really misses the mark.

 

On a side note, any idea which settings they ran the benchmark on?

Well yeah.. this is what we learned in this video.  I was able to get the card to draw crazy amounts of power off screen, but there was literally no point since I couldn't make the card go any faster than it was at 580W.  Clearly the GPU itself just isn't happy going much over 3GHz no matter what you do to it.

 

Also settings were maxed out @ 1440P, RT Highest

[Cropinky]

Video games are 2000 and late.I think it would be interesting to setup a deep learning benchmark for this kind of crazy overclocking, I think deep learning workflows can actually benefit much more from this kind of overclocking than video games.

You can email me if you're interested in me making something like that. Overall i think deep learning benchmarks should become a part of every GPU review on this channel since gaming at high fps is not something that demanding these days. This deep learning workflow benchmark can be like 1 docker image and 1 python script that dumps results in a csv, nothing too fancy required.

 

https://jere357.github.io