NVIDIA GeForce RTX 5090 Founders Edition Unboxing

[AVATARAT]

If you think going from a 3-slot cooler with a lower TDP to a 2-slot cooler with a higher TDP isn't impressive, then you aren't thinking.

When the normal thermal paste stops working / dries too fast - the next step is LM/P-Ch - there thermal transfer is much better, but it goes into another category of cooling.
This has been known for a long time in the overclocking community, so yeah, I don't see anything impressive.
Far better is water cooling, but then public opinion (mind share) will change because water cooling is associated with very hot chips.

 

[LittleBro]

The problem is that FG doesn't solve the same problem / problems that a traditional frame does. It's not a frame produced by the engine, so no calculations are done during the FG frame. If you are clicking your mouse, it's not registering during the FG frame, the opponents aren't taking damage, aren't moving etc. It just offers smoother image. Which is great - especially in CPU limited scenarios, but im just saying we shouldn't conflate the 2, it's really not the same thing.

Exactly. It basically does not offer any advantage but smoother framerate, which results in latency penalties. That's where Reflex 2 comes in - gets updated input device position from CPU and alters/warps existing frame accordingly, renders it as the next frame without other inputs from CPU. This not only eliminates penalty that is caused by (M)FG, but also kind of "cheats" with waiting for GPU to render next native frame. I must say I'm impressed by this functionality. Of course, it will have some drawbacks, because everything has them, but currently I see consider is the best and only option to radically improve input latency with (M)FG. Though I'm not fan of FG and post-rendering image manipulation techniques, I like the idea behind this workaround. Smart.

Regardless of what you may think about NVIDIA's pricing/DLSS/packaging, I think we can all agree that squeezing a 600W-TDP chip into a 2-slot cooling solution is an impressive technological achievement. I definitely want to see the temperatures and noise results for this card.

Temps, noise and longetivity. The amount of heat generated on such small PCB is unprecedented.

I think switching to liquid metal plays the most important role in RTX 5090 FE cooler's design. Heat transfer must have improved so much that it allowed for much less heatsink dissipation area. Still, when I look at RTX 5090 pictures, I always keep getting same thought like it's just too much of a heatsink (dissipation area) loss compared to RTX 4090. And RTX 5090 adds 125W of TGP. Heat transfer between die and vapor chamber is just one part of the job, I really wonder how they improved heat transfer between vapor chamber "heatpipes" and heatsink fins.

 

[KrachB00Mente]

There is an error in the article.
DLSS 4 is not Series 5 exclusive. Multi Frame Generation is.

 

[PrimashockVR]

And nobody tried this in VR. That's a real scenario to know if this thing is worth or not. And I'm not talking with the Quest 3, I'm talking about a PCVR Hmd.

 

[Assimilator]

When the normal thermal paste stops working / dries too fast - the next step is LM/P-Ch - there thermal transfer is much better, but it goes into another category of cooling.
This has been known for a long time in the overclocking community, so yeah, I don't see anything impressive.

Liquid metal isn't a magic bullet that somehow changes physics. Yes it allows that heat to be moved away from the GPU faster, but that's irrelevant if you aren't getting rid of said heat, which what you need surface area for. And there is more heat being produced, and less surface area to dissipate it with, than the previous-generation GPU and its FE cooler.

 

[N/A]

Nothing agrees with that slide because it makes no sense, it implies the noise increases exponentially somehow close to 500W, according to them a 4090 FE probably generates one gazillion decibels or whatever at 500W lol. Just another bs graph like the rest.

Of course it doesn't extend forever, it just throttles or something. First off the 2080 Ti dual axial was 37 dBA at 250W, that may be accurate.
dual slot single flow 3080 Ti with 39 dBA at 400 W that also may be accurate. But both use smaller fans.

 

[chrcoluk]

How does PCIE pins connect to the board? is this part of information still under NDA?

Riser cable style, connector is propriety.

 

[N/A]

Riser cable style, connector is propriety.

Imagine if we could use a custom riser cable, get rid of the slot thing and the water block is just that square covering the MXM board, or a cube because the back side could use some cooling too. Your video card is a Cube now with some pipes going in and out.

 

[mtosev]

Does anyone have an idea how much will the Asus ROG Strix RTX 5090 cost? Any predictions?

 

[Chomiq]

Does anyone have an idea how much will the Asus ROG Strix RTX 5090 cost? Any predictions?

$2600

 

[Assimilator]

Does anyone have an idea how much will the Asus ROG Strix RTX 5090 cost? Any predictions?

Far more than it is worth.

 

[freeagent]

$2600

That is just the Asus tax, add a few hundred if you want the ROG flavor flave.

 

[AVATARAT]

Liquid metal isn't a magic bullet that somehow changes physics. Yes it allows that heat to be moved away from the GPU faster, but that's irrelevant if you aren't getting rid of said heat, which what you need surface area for. And there is more heat being produced, and less surface area to dissipate it with, than the previous-generation GPU and its FE cooler.

Sure, it's not magic, I agree with that, but as you can see from the video, they have improved (made bigger) the Viper's camera
I understand your viewpoint, yes they are getting better air cooling and that's nice.

But anything with 500-600w TDP usually needs water just because it's more efficient.
From a technical standpoint this cooler is very well designed and yes, impressive.
But that doesn't make the decision to cool 600w cards this way any less... strange, when there are well known and used ways.

 

[Assimilator]

Sure, it's not magic, I agree with that, but as you can see from the video, they have improved (made bigger) the Viper's camera
I understand your viewpoint, yes they are getting better air cooling and that's nice.

But anything with 500-600w TDP usually needs water just because it's more efficient.
From a technical standpoint this cooler is very well designed and yes, impressive.
But that doesn't make the decision to cool 600w cards this way any less... strange, when there are well known and used ways.

It would be strange to use anything other than an air cooler, given the space requirements that liquid coolers impose for mounting the radiator and NVIDIA's push for SFX compatibility.

 

[Chomiq]

That is just the Asus tax, add a few hundred if you want the ROG flavor flave.

I mean, Strix isn't even their top 50-series SKU, right? It's something like Luna, Eclipse, Anus?

Edit.
Oh yeah, Astral.

 

[AVATARAT]

It would be strange to use anything other than an air cooler, given the space requirements that liquid coolers impose for mounting the radiator and NVIDIA's push for SFX compatibility.

Then let's see how this works

 

[lexluthermiester]

I mean, I think it looks pretty smart. Not sure how it'll turn out inside a case (with the heat being blown onto the CPU), but they've made it look far more refined that the 4090FE.

Agreed. 2 slot design and a very sleek look. This is the first founders card I actually would like to have. I wonder what the 5080/5070 cards will look like?

 

[lepudruk]

Is it really so hard to make an adapter? Maybe ask those cable mod companies for that.

This is Thermal Grizzly WireView:

It's not a common cable mod and it can be plugged only one way. Changing power charge slot in 5000 Nvidia generation to diagonall puts all of those adapters in a trash. Perhaps only FE has a diagonall slot and non-reference cards will have the old system..?

Reviews date?

 

[dir_d]

http://imgur.com/zTREGr9

Nice BBC (Big Black Cable) @W1zzard

 

[LittleBro]

To be honest with you, I don't particular care whether you believe me or not. Ill say this and then just move on, whether you accept it or not.

When you GPU bottlenecked, CPU prepares frames faster than the GPU renders them. So CPU prepares frame 1 in 10 ms, frame 2 in 20 ms, frame 3 in 30 ms. GPU takes 30 ms to render frame 1, so by the time it's done frame 3 is already in queue. So now it starts preparing frame 2, that's already 60 ms. Then it's frame 3, now we are at 90 ms. That's what reflex stops from happening, it "syncs" the GPU with the CPU, that's why your fps drops when you enable reflex, cause instead of both the cpu and gpu spitting frames as fast as possible, they make sure to prepare and render one at a time combined.

I must admit I misunderstood your original post. Now I clearly undestand what you meant. I removed my "utter bullshit" wording from my original post.

As for those queued frames waiting to be rendered, the user input (keyboard presses, mouse movement) cannot be processed anymore in that state. This is why Nvidia introduced new Reflex 2 technology.

 

[freeagent]

Nice BBC (Big Black Cable) @W1zzard

He puts the Dong in Dongle.

 

[Xaled]

5090 Unfoundable-Edituon

 

[_roman_]

Maybe I was not clear.

Just ask a cablemod company, or a company, to make an "extension cord" so the measurement adapter fits.
Honestly It's badly designed when it does not fit any nvidia graphic card with the nvidia graphic card power supply connector. Also, what will you do, when there will be two nvidia graphic card connectors on one card? (I did checked last time the net - what that product is)

 

[N/A]

HardwareUnboxed between the lines basically said 5090 = 4090 + 25% performance for +25% more $$$. So zero performance per dollar increase gen to gen.

I believe it's closer to 50% when maximum power limits are applied, if they added 4 more GPC with 1536 shaders instead of just growing the existing GPCs to 2048. That's 30% scaling non linear and very constrained by TGP and another 15% for having GDDR7 with 33% more bandwidth.

 

[The Von Matrices]

Liquid metal isn't a magic bullet that somehow changes physics. Yes it allows that heat to be moved away from the GPU faster, but that's irrelevant if you aren't getting rid of said heat, which what you need surface area for. And there is more heat being produced, and less surface area to dissipate it with, than the previous-generation GPU and its FE cooler.

But physics doesn't have to change. The liquid metal's increased heat transfer would cause the heatsink to run at a higher temperature for a given core temperature, which would increase the heat dissipation rate for a given fan/heatsink combo or allow a smaller heatsink with the same heat dissipation ability.

This is the same reason that CPUs with thick heat spreaders need bigger cooling systems to control temperature even if the CPUs aren't producing any more heat (AM5 CPUs being a noteworthy example).