GeForce RTX 5090 Power Excursions Tested: Can Spike to 901W Under 1ms

[btarunr]

Igor's Lab conducted an in-depth analysis of the power management system of the new NVIDIA GeForce RTX 5090 graphics card, including the way the card draws peak power within the tolerances of the ATX 3.1 specification. This analysis should prove particularly useful for those still on older ATX v2.51 PSUs, and plan to use the included power adapter that converts four 8-pin PCIe power connectors to a 12V2x6. Igor's peak power analysis shows that the RTX 5090 is capable of excursions as high as 627.5 W for 10 ms to 20 ms durations; as high as 738.2 W in 5 ms to 10 ms durations, as high as 823.6 W in the 1 ms to 5 ms category, and as high as 901.1 W in spikes under 1 ms in duration.

An excursion is a brief increase in power draw beyond the continuous power delivery limit of the connector (600 W in case of the RTX 5090's single 12V2x6 input and adapter that converts four 150 W 8-pin PCIe inputs). There is nothing particularly alarming about these numbers, and the excursions part of Igor's analysis fall within the specification of the ATX 3.1 standard, which calls for excursions of up to 200% (1200 W) up to 1 ms. Any PSU meeting the ATX 3.1 specs that even has a continuous power output of less than 1200 W will be capable of handling these spikes. It's only with the much older generations of PSUs, such as ATX v2.51 (mid-2010s) that excursions can trigger OCP. Find other great insights in the Igor's Lab review linked below.



View at TechPowerUp Main Site | Source

 

[tpa-pr]

That is a BIG jump, wow. Even though it's for sub 1ms that's a lot of wattage to put through the connector.

 

[LabRat 891]

Let the 12V-2x6 burninating, begin! (again)

 

[Zazigalka]

generational spike instead of generational jump.

 

[HisDivineOrder]

Seems pretty obvious. Hard to imagine they won't have more melted cables, but now across the 80 and 90 series.

 

[londiste]

So, about 50% above the sustained power usage? This is basically the standard spikes, controlled well enough.
It does have implications for PSUs if the one you use does not have the wattage or trigger OCP too fast.

This being said, it is worse than the spikes 4090 had which were absolutely excellently controlled. And hopefully AIBs manage to do at least as good of a job as Nvidia with this.

Seems pretty obvious. Hard to imagine they won't have more melted cables, but now across the 80 and 90 series.

For melting we are looking at the 500-600W sustained usage as the problem. the 901W is more of a good headline rather than a technical problem.

 

[Legacy-ZA]

*Sigh*

Not again, I don't trust the power connector to not fail. :S

 

[omnifuneral]

Absolutely no way this thing doesn't melt connectors with regularity. Godspeed to anyone with a PSU under about 1200w, too...

 

[Zazigalka]

*Sigh*

Not again, I don't trust the power connector to not fail. :S

yeah, not with 650w. 4090 was like 400w realistically, most of the time around 350.

 

[Timbaloo]

For melting we are looking at the 500-600W sustained usage as the problem. the 901W is more of a good headline rather than a technical problem.

People don‘t get this. They see 900W and go „zomg power connector“… They lack basic understanding of physics.

 

[LabRat 891]

People don‘t get this. They see 900W and go „zomg power connector“… They lack basic understanding of physics.

If the cable/conductors aren't losing heat faster than they're gaining, they'll eventually runaway under sustained load. -Regardless of peaks.
Not to mention, physical migration/movement from thermal expansion/contraction...

 

[AKBrian]

as high as 823.6 ms in the 1 ms to 5 ms category

Minor typo.

 

[Timbaloo]

If the cable/conductors aren't losing heat faster than they're gaining, they'll eventually runaway under sustained load. -Regardless of peaks.
Not to mention, physical migration/movement from thermal expansion/contraction...

This thread is about spikes though, and these really don’t matter from a thermal point of view.

 

[kondamin]

So you are buying a 3000 usd video card and not getting a new PSU…

 

[Steve Ballmer]

Let the world burn!

 

[Bwaze]

 

[GodisanAtheist]

Guess power spikes are back on the menu for RTX5xxx series...

 

[Outback Bronze]

Ouchies!

 

[wolar]

Thankfully atx 3.1 can sustain that right ?

 

[londiste]

Thankfully atx 3.1 can sustain that right ?

When it says under 1ms spike, this by definition means it does not need to be sustained.

 

[AVATARAT]

The removed Hot Spot is one of the reasons this card is not a thermal paste, but a liquid metal or phase change pad. In the long run 400w+ bakes the paste, and here we have 500-550w with almost double spikes.
Out of sight, out of mind - it just works!

Spoiler

 

[JustBenching]

Thankfully atx 3.1 can sustain that right ?

A good 1k or 1.2k psu should be able to without triggering ocp without the atx specification required.

 

[Gameslove]

That here say, very very big disappointment...

 

[_roman_]

I need to catch up.

I read about on igor in german yesterday.

Igor wrote that a 1000 watt power supply is enough.

-- I personally think that the average wattage counts over the connector.

Note ms = is a thousands of a second. 1000 milliseconds equal a second.

I have not yet bothered reading. That takes a lot of time for the 5090. Different websites. I have not yet watched any videos. Differetn viewpoints and such.

I think. how many of those 1000 of 1000 milliseconds are over the 575 Watts? Than make a percentage.

Note: AFAIK: that 5090 graphic card has 575 Watts. The connector is rated for 600 Watts. The pcie slot is rated for 75 Watt.

 

[motov8]

600w is too much for a room without air conditioning. This is an oven. I don't want to know how someone doesn't have air conditioning and has 28'C in the room in the holidays playing on a 200-250 watt graphics card like me.