NVIDIA RTX PRO 6000 Workstation Runs Much Hotter Than 5090 FE

[KimRRRich]

I recently tested the NVIDIA RTX PRO 6000 Workstation and was shocked by how hot it runs. Under heavy workload (FurMark) — but without any fan directly under the GPU —the temperature climbed to 93°C (— and I had to stop there for safety reasons. That wasn’t even its thermal limit.

By contrast, on the same hardware setup and under the same conditions, my RTX 5090 Founders Edition topped out at only ~73°C.

That’s a 20°C difference, despite the power consumption being almost the same:

• 5090 FE: ~575W
• RTX PRO 6000 Workstation: ~600W

My best guess? The 5090 FE might be using liquid metal cooling, while the PRO 6000 likely uses traditional thermal paste, which would explain the large gap in thermal performance.
Please excuse any inaccuracies or non-standard testing methods — just sharing personal observations in case others find them useful.

 

[Youseethis]

Correct.

 

[ir_cow]

Why are we stress testing with Furmark? That isn't a normal workload.

 

[mrpaco]

there is no safety reasons to stop, the card will just throttle on the thermal limit set by the bios, which is way lower than the theoretical thermal limit of the die/memory

it's pretty awful that nvidia didn't use liquid on that card with the price it has though.

 

[ir_cow]

it's pretty awful that nvidia didn't use liquid on that card with the price it has though.

its pretty smart. The card needs to last a lot longer in a workstation environment. LM just dooms anything once it dries out in a few years.

 

[R-T-B]

its pretty smart. The card needs to last a lot longer in a workstation environment. LM just dooms anything once it dries out in a few years.

Does it actually dry out? I was under the impression it didn't do that unless it was reacting to something.

Anyways if it isn't using liquid metal, that could explain the temp difference.

 

[Onasi]

A card running a GPU with more active cores and less dark silicon within the same power envelope while also running significantly more VRAM is hotter. More news at 11? And I doubt the difference is solely due to different TIM. In any case, FurMark results are basically irrelevant in practice anyway.

 

[Selaya]

^ for teardown/confirmation that it is indeed just using ... ptm from the looks of it?, and not lm

 

[ir_cow]

Does it actually dry out? I was under the impression it didn't do that unless it was reacting to something.

Anyways if it isn't using liquid metal, that could explain the temp difference.

It does for CPUs, but I honestly don't know about GPU. Only read a few things over the years of people having to reapply LM to the GPUs. I don't know the context past that.

 

[londiste]

Why are we stress testing with Furmark? That isn't a normal workload.

Why not? We are testing for the worst case scenario, when card hits its power limit and Furmark does that quickly and consistently.

 

[JoeTheDestroyer]

Does it actually dry out? I was under the impression it didn't do that unless it was reacting to something.

Anyways if it isn't using liquid metal, that could explain the temp difference.

I just retired my 7y/o Kaby Lake system. I used liquid metal both between die and heatspreader, and between heat spreader and copper aio (for a 20degC reduction in temps). Never repasted, worked fine until the day I turned it off for the last time.

The issue of liquid metal "drying out" is greatly exaggerated, imo.

 

[R-T-B]

I just retired my 7y/o Kaby Lake system. I used liquid metal both between die and heatspreader, and between heat spreader and copper aio (for a 20degC reduction in temps). Never repasted, worked fine until the day I turned it off for the last time.

The issue of liquid metal "drying out" is greatly exaggerated, imo.

I think it can happen but depends greatly on the materials. I think pure copper is far more prone than nickel plated.

 

[ir_cow]

The issue of liquid metal "drying out" is greatly exaggerated, imo.

Happened to me with the 12700K delid. 6 months and the temps were worse than stock. The IHS is exposed copper inside where to indium is / was.

 

[Tyler-98-W68]

Thanks captain obvious, are you happy now that you got that off your chest, just had to register to post that?

 

[ir_cow]

Thanks captain obvious, are you happy now that you got that off your chest, just had to register to post that?

Brutal. Such kind words.

 

[JoeTheDestroyer]

I think it can happen but depends greatly on the materials. I think pure copper is far more prone than nickel plated.

Anticipating this argument is why I mentioned my aio was copper, meaning copper cold plate.

It's got to be some other factor that gives me vs. ir_cow such an extreme difference.

Only thing I can think of is different mixture of materials. The experience I related dates from the early days of liquid metal as a tim, mine was from brand I'd never heard of (which, iirc, turned out to be a scientific supplier).

Though I do also have more recent experience with Thermalgrizzly liquid metal, between 7950x3d and optimus nickel plated water block. That was good for atleast 2 years, might have gone longer but had to pull it loose to replace dead motherboard.

 

[Dr. Dro]

A card running a GPU with more active cores and less dark silicon within the same power envelope while also running significantly more VRAM is hotter. More news at 11? And I doubt the difference is solely due to different TIM. In any case, FurMark results are basically irrelevant in practice anyway.

This is exactly what I was thinking. The core on the RTX 5090 is significantly cut down, and the Pro 6000 also has the full amount of L3 enabled (96 > 128 MB), and that's a lot of L3$ transistors that never see any work in the 5090. Stands to reason that it runs much hotter.

I'll also go out on a limb and make an educated guess that the Pro 6K has a much more conservative acoustic profile, since it's intended to be used in business settings where a GPU's noise might be actively considered a major annoyance, the RTX 5090 FE was never known for being a quiet card and in fact, received an acoustics overhaul through driver updates post-launch to quieten it down a little bit.

 

[R-T-B]

Well yeah enabling more silicon is going to run hotter, but honestly, the 5090 is pretty enabled already. I don't see it making a whole huge lot of difference. At most (and this is being pretty generous), we're looking at another 10th slice of the core being enabled. Not enough to really be a incredible heat producer... there is more at play IMO. It feels to me more like a thermal impedance at play.

I think the remaining difference could be the difference between LM and conventional TIM, and longevity may be the reason, as mentioned.

 

[eidairaman1]

Why are we stress testing with Furmark? That isn't a normal workload.

Power virus, consumer level gpus typically will throttle back due to it.

 

[Visible Noise]

New reg doesn’t know what a power virus is.

Sigh.

 

[tpa-pr]

Thanks captain obvious, are you happy now that you got that off your chest, just had to register to post that?

Not sure why the topic warranted such a caustic response, it's pretty tame and inoffensive.

Anyway, the temp difference and TIM difference is interesting to know. I assume that Nvidia's logic is cards like the 6000 will be used in well-ventilated cases with plenty of airflow and as others have said, PTM lasts forever. Liquid Metal seems a bit on the temperamental side whereas PTM is just set and forget, which it what business users would want.

Would the Pro cards have a different throttle temp compared to the consumer cards?

 

[Dr. Dro]

Not sure why the topic warranted such a caustic response, it's pretty tame and inoffensive.

Anyway, the temp difference and TIM difference is interesting to know. I assume that Nvidia's logic is cards like the 6000 will be used in well-ventilated cases with plenty of airflow and as others have said, PTM lasts forever. Liquid Metal seems a bit on the temperamental side whereas PTM is just set and forget, which it what business users would want.

Would the Pro cards have a different throttle temp compared to the consumer cards?

Doubt it. The throttle point for the 5090 seems to be the 90-100 range as well. If anything, they made the Pro 6K intentionally quieter, but since it has the same flow through dual slot cooling solution + more of the chip is enabled, that causes the card to heat up a ton more.

 

[Outback Bronze]

Guys,

Just deleted the last 3 posts. Points will be given if this bickering continues.

Thanks.

 

[rusty caterpillar]

I recently tested the NVIDIA RTX PRO 6000 Workstation and was shocked by how hot it runs. Under heavy workload (FurMark) — but without any fan directly under the GPU —the temperature climbed to 93°C (— and I had to stop there for safety reasons. That wasn’t even its thermal limit.


View attachment 407076


By contrast, on the same hardware setup and under the same conditions, my RTX 5090 Founders Edition topped out at only ~73°C.



View attachment 407077


That’s a 20°C difference, despite the power consumption being almost the same:

• 5090 FE: ~575W
• RTX PRO 6000 Workstation: ~600W

My best guess? The 5090 FE might be using liquid metal cooling, while the PRO 6000 likely uses traditional thermal paste, which would explain the large gap in thermal performance.
Please excuse any inaccuracies or non-standard testing methods — just sharing personal observations in case others find them useful.

NVIDIA RTX PRO 6000 has 96 GB VRAM compared to 32GB on 5090, higher density PCB, therefore heat by proximity is higher.

I don't know exactly how wide is the PRO 6000 but, I guess is around 13-14cm. A targeted airflow of 12cm fan will work nicely.

In order to do that you have to take this shroud bits off



as they are not linked to the back plate with pads or thermal putty so the heat transfer in between them is too slow and set a 12cm Noctua A12/25 or a G2 /25mm in pull at about 1200 RPM directly on the backplate sitting on his own rubber mounting pins that comes in the box.



Set the Noctua A12 /25 in pull on the red square. You can leave the glossy shrouds bits on but, the results will be not that good as with those bits off.
If you prefer to leave the glossy bits of shroud ON you have put some thermal putty in between them and the main backplate of the card to greatly increase the heat transfer in between metal shroud and backplate and the fan will do the rest.

On the other hand increase the RPM of the rear case fan, maybe to maximum.
Is worth checking if stock thermal pads is not already baked and brittle, I don't on what temps you run the GPU and for how long.

I will certainly do all this if I had and RTX 6000 PRO, hope it helps.