Overclock.net banner
1 - 20 of 209 Posts
Discussion starter · #1 · (Edited)
Let me help you dodge a bullet. Don't buy Gigabyte 4090 Xtreme Waterforce. It is made like crap.

I bought two 4090. Another one is Strix. Others are all sold out. They have been used for two weeks. Here is my impression:

1. This version of Gigabyte 4090 is built like plastic.
(1)The shroud looks cheap.
(2) The copper plate is machined with scratches. It doesn't cover components such as the VRM of the memory the way air-cooled GPU does.
(3) A part of the north core VRM is only covered 1/3. It's more like Gigabyte adjusted that part of the PCB upside down but forgot to communicate with the guys who made the copper plate. The components that are less than half-covered like this will have higher temperature, higher chance to degrade overtime to the point of burndown.
(4) One of the thermal pads is off-centered.
(5) The 360mm radiator is made of light aluminum instead of copper.
(6) The fans are not new. They seem to be refurbished from the previous 3090Ti. I can still see the debris left on the blades washed by electronic cleaning fluid.
Image

Image

Image

Image


2. It's not exactly quiet when the fans go to 1300rpm. The fans are 3000rpm. They are only slightly quiet at 900rpm which is 30% of the max speed. And the fans keep spinning at 30% all the time. They don't stop. In comparison, ASUS Strix let the fan stops when idle. Strix is also quieter when at the same load.

3. This is critical. The default fans of Gigabyte hit to max like a jet every now and then under load. Something triggered the annoying behavior. The BIOS update didn't fix it. When you control the fans through software like MSI afterburner. The fans go to the double speed of the set parameter. Multiple people have the same issues.

4. The power is limited to 500W. After the BIOS update, the power is limited even down to 490W. No overclocking protentional. The website of official Gigabyte says it is OC version. But the package of the box has no mention of the OC version. It is just a normal PCB despite the premium price.

5. The temperature reaches to 58c-63C easily at 490W. Any more than that the fan will spin louder.

6. The possible culprit of the fans hitting the max is a sudden temperature rise. Some components such as the partially covered VRM or the uneven pasted memory are not properly cooled. They can heat up to 100C threshold multiple times such as the memory temperature. No more BIOS updates can fix high temperatures like this.
Image


7. The thermal pad on this GPU is also a cheap adaption due to incompetent design:

(1) Though the original pad has good thermal conductivity, it is brittle enough to deform that you need to replace them after every disassembly.

(2) The distance between VRM chips is 2.78mm. The copper plate distance for each type of chip is 3.00mm. The main power delivery chips have a 1.5mm thermal pad. So the side chips require a 1.28mm (1.5mm+2.78mm-3.0mm) thermal pad.

Image

Image


(3) There is no conventional 1.28mm thermal pad. This can cause a lot of variations to overheat components. The thermal pad is usually the thinnest at the end. And the north VRM at the end is only 1/3 partially covered. This is why a large portion of burned chips reside at the edge of a PCB. A lot of things can go wrong with a design like this. Who the hell thinks this is a good idea?

(4) The actual pad comes off with the plate is 1.20mm short of touch after disassembly. If using a 1.5mm thermal pad instead, it also causes less pressure on the memory and core.

(5) So you have to hammer a 1.5mm thermal pad down to 1.28mm rated at 15W/mK. You don't buy a premium card to disassemble it multiple times to deal with hassles.
Image

Image



8. Overall the cooling capacity of a cheaply made copper plate is limited. It is engineered in a way the plate doesn't even cover all the VRMs as the heat dissipation of that plate is less than 500W. The thermal pad is also uniquely required at 1.28mm. These Gigabyte engineers don't do proper calculation, they do ill calculation to make core temperature looks good with a risk to burn components elsewhere. The overall result is the fans hit like a jet in the combination above. This is also why a so-called water-cooled GPU with incompetent cooling capacity in disguise has a worse power limit down to 500W instead of 600W compared to the air-cooled GPU.

=======================

The conclusion:

The value of this card is bad. It is not comparable to the ASUS Strix in terms of almost everything such as build quality, noise level, OC performance. Strix is built like a piece of art in comparison. The premium product is all about the user experience with as little annoyance as possible. The Gigabyte AIO 4090 has a crap-tier user experience not even mentioning there is a risk of burn of 100C components in the long run. Gigabyte didn't test everything.

So don't buy this GPU. A thin piece of loose bare cheap-machined copper with scratches won't cut it when other components need to be cooled down properly.
 
Discussion starter · #4 · (Edited)
Discussion starter · #5 ·
Aluminum radiator was a given, but the rest is rather disappointing for a card that retails at $1900.
Some of the issues might be solved by further BIOS updates.
I doubt the pump can last for 2 years as the card is made cheap all around.
So far Gigabyte's experience in water cooling still stays within the niche products such as 3-way Waterforce GTX 980s a long time ago.
 
I don't believe any of the words past the first 3 were necessary for the thread title.
 
Mine works just fine, you will never hit that 510-520W power limit unless hammering it with 3Dmarks, Furmark or stuff like Quake RTX.
Complaining about the radiator being aluminium is unfair as 99% of AiO's have aluminium rads. Also the VRAM temp on mine is about 70*C, far from 100*C.
Looks like you've got a bad card.

I won't touch any Strix card with a 10 foot pole as they coil whine like mad.
 
Discussion starter · #14 · (Edited)
Mine works just fine, you will never hit that 510-520W power limit unless hammering it with 3Dmarks, Furmark or stuff like Quake RTX.
Complaining about the radiator being aluminium is unfair as 99% of AiO's have aluminium rads. Also the VRAM temp on mine is about 70*C, far from 100*C.
Looks like you've got a bad card.

I won't touch any Strix card with a 10 foot pole as they coil whine like mad.
I have both of them to compare.

Claiming how little power is used on 4090 only reveals that you don't even use half of the memory or do proper OC.

And 500+W is easy if the BIOS is not cut down to this low as the whole thermal design is just as limited as it can be.

QC wise, just look at that cheap machined plate and how bad the thermal pad is.

You talk like you don't have a Strix 4090. It is whisper quiet, no coil whine, 550W all the time. And it is bought at the same price but in fact thousands miles of superior than this Giga Wateforce.
 
I've had the 2080Ti and 3090 Strix and I won't gamble on another, as they all had coil whine. My 4090 doesn't touch 400W in games I play.
 
Discussion starter · #16 ·
Coil while is a matter of lottery instead of the quality of the components. Every GPU has a chance.

You haven't maxed out the card yet. It's either the resolution or the CPU not enough to drive the card.
 
3. This is critical. The default fans of Gigabyte hit to max like a jet every now and then under load. Something triggered the annoying behavior. The BIOS update didn't fix it. When you control the fans through software like MSI afterburner. The fans go to the double speed of the set parameter. Multiple people have the same issues.
I have another suggestion for the brief and apparently random moments of maximum fan speed. This is a known issue with all nVidia cards, but it is particularly noticeable with their Linux driver. There were a couple of threads about it on the nVidia forum a few years ago, not sure if they're still there. Sometimes a card will stop communicating with the driver for a fraction of a second. To protect the GPU, the firmware will kick the fans up to 100% until that communication is restored. I've seen it personally on GTX 970s, 1080s, 1080Tis, RTX 2080s, and 3090s. I've not seen it on Quadros (yet). It appears more likely on systems with multiple GPUs, but that doesn't mean it doesn't happen on single GPU boxes. Might have something to do with PCI-E topology and load as well, as I played musical graphics cards in a quad-GPU system to see if it was a specific faulty card and the card which would always ramp up to maximum was always, always, always in the third PCI-E x16 slot from the CPU socket. Some drivers are better than others at this (i.e.: it does not happen as often).

But the other issues are pure Gigabyte.
 
Its rather odd, gigabyte actually made some of the best GTX 900 and 1000 series cards with the windforce and G1 gaming models.

But for most of the rest of their stuff, their product quality is on par with their customer service, which to say = exploding PSUs.

My windforce OC 1080Ti hits 2100 GPU and +1000 on vram with ease and is still going strong. Wouldn't buy anything other than an FE now though since they got so much nicer since with all the RTX series so far.
 
@SilenMar

What are your temperatures after repadding?

(And nice pictures I have not disassembled mine. The design with 24-phase VRM looks solid and leaving one partly covered by the copper plate is negligent and careless. I hope I can sleep because of your post otherwise I have to make a copper shim. But for now my max junction temp. has never passed 80'C and is silent. I seriously punish my PC with hours of MSFS in VR and 4K)
 
1 - 20 of 209 Posts