• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

MSI GeForce RTX 5090 Suprim Liquid SOC

Hey @W1zzard Thanks for your hard work!
Is the Memory OC @ 34Gbps really stable? If yes then that's a huge Memory Bandwidth uplift vs Stock !! +21% is no slouch!

It seems Gigabyte will also have an AIO version: https://www.gigabyte.com/Graphics-Card/GV-N5090AORUSX-W-32GD#kf

I wonder how it compares to MSI's. Looks like Nvidia's block covers the entire thing, and isn't using a fan on the card itself.

Do we know when Gigabyte AIB benchmarks are expected? Apparently Gigabyte has some delays with sending out review samples?
2655MHz Boost Clock is literally 10% more than the FE ! That's pretty impressive...
 
I'm gonna buy the new Alienware Area 51 so I get their AIO cooler for the CPU and the GPU.
Otherwise, this is the 5090 I was waiting for.
 
Is the Memory OC @ 34Gbps really stable?
Unfortunately I couldn't OC higher due to the driver limiting the OC, so I can't tell you where it becomes unstable.

I didn't notice any issues in performance benchmarks, but feel free to subtract a few Gbps safety margin, still an impressive number
 
It seems Gigabyte will also have an AIO version: https://www.gigabyte.com/Graphics-Card/GV-N5090AORUSX-W-32GD#kf

I wonder how it compares to MSI's. Looks like Nvidia's block covers the entire thing, and isn't using a fan on the card itself.

Do we know when Gigabyte AIB benchmarks are expected? Apparently Gigabyte has some delays with sending out review samples?
And it seems Gigabyte is also using liquid metal (unlike MSI): "Liquid metal thermal grease for GPU".

That should be an extra 2 degrees cooler or so, compared to regular thermal paste. No?

I really want reviews for that :(
 
It seems Gigabyte will also have an AIO version: https://www.gigabyte.com/Graphics-Card/GV-N5090AORUSX-W-32GD#kf

I wonder how it compares to MSI's. Looks like Nvidia's block covers the entire thing, and isn't using a fan on the card itself.

Do we know when Gigabyte AIB benchmarks are expected? Apparently Gigabyte has some delays with sending out review samples?
Asus will have one as well https://rog.asus.com/graphics-cards/graphics-cards/rog-astral/rog-astral-lc-rtx5090-o32g-gaming/

I got the MSI Suprim Liquid 4090, I haven't had any complaints. I did only buy EVGA hybrid cards before, but now that they are no more, I guess I'll stick with the MSI AIO card. This review sealed it for me.
 
All reviewers are saying that the radiator is 360 type, but it is 394 mm in length, about 400 mm.

 
1737799471226.png

What's happening to this website's grammar?
 
All reviewers are saying that the radiator is 360 type, but it is 394 mm in length, about 400 mm.

360 because it uses 3x120mm fans. If the entirety of the radiator length was covered by the fans, where would the tubes go? Hence it's longer than the sum of the fans.
 
I would expect 370 mm, but it is 400 mm, so the buyer should check the case, if it fits inside.
 
View attachment 381587
What's happening to this website's grammar?
When websites are under competitive pressure to launch articles as soon as NDAs expire, things like this will initially slip past copy editing. With the way ads pay little, I don't know if many sites can afford copy desks or not. Many smaller newspapers which used to be able to afford copy desks no longer can afford to hire the professional copy editors required to staff copy desks these days.
 
I found a similar error in https://www.techpowerup.com/review/msi-geforce-rtx-5090-suprim-liquid/5.html to the one that I found in https://www.techpowerup.com/review/msi-geforce-rtx-5090-suprim/5.html where a VRM controller is mistakenly described with a DrMOS's model number instead of a VRM controller's model number. This should also result in this article getting a correction.
The sentence "Powering the 16 GDDR7 memory chips is a seven-phase VRM driven by the same Monolithic MP87993 controller that controls GPU voltage." should probably be edited to replace "MP87993" with "MP29816". That sentence is in https://www.techpowerup.com/review/msi-geforce-rtx-5090-suprim-liquid/5.html .
 
I'll start by saying that I think $2500 is an indecent price but probably no more and no less than the 5090FE at $2000...

At this point, having money availability, a little bit of madness and considering the following premises:
- greater performance
- excellent temperatures
I think it should be considered instead of FE!!!
 
Unfortunately I couldn't OC higher due to the driver limiting the OC, so I can't tell you where it becomes unstable.

I didn't notice any issues in performance benchmarks, but feel free to subtract a few Gbps safety margin, still an impressive number
I wonder if even 32Gbps (aka 2TB/s) is totally stable for 24/7 use! If so I might do that if I get a 5090.

I was able to get my GDDR6X running at 25Gbps on my 4090 SUPRIM LIQUID X.
 
I found a similar error in https://www.techpowerup.com/review/msi-geforce-rtx-5090-suprim-liquid/5.html to the one that I found in https://www.techpowerup.com/review/msi-geforce-rtx-5090-suprim/5.html where a VRM controller is mistakenly described with a DrMOS's model number instead of a VRM controller's model number. This should also result in this article getting a correction.
Fixed

View attachment 381587
What's happening to this website's grammar?
When websites are under competitive pressure to launch articles as soon as NDAs expire, things like this will initially slip past copy editing. With the way ads pay little, I don't know if many sites can afford copy desks or not. Many smaller newspapers which used to be able to afford copy desks no longer can afford to hire the professional copy editors required to staff copy desks these days.
Whoops, I missed that in the teaser, and I read it like 3 times, but it was already 30 sec past NDA. Thanks for pointing this out, fixed now
 
Fixed



Whoops, I missed that in the teaser, and I read it like 3 times, but it was already 30 sec past NDA. Thanks for pointing this out, fixed now
That's some high pressure, releasing reviews right when embargo lifts! Thanks for the hard work.

I believe there are some proof-reading software out there that should be able to catch these fairly easy. With the AI craze I won't be surprised if one of those tools is much smarter than the good old ones, and can probably catch even the easy-to-miss mistakes.
 
That's some high pressure, releasing reviews right when embargo lifts! Thanks for the hard work.

I believe there are some proof-reading software out there that should be able to catch these fairly easy. With the AI craze I won't be surprised if one of those tools is much smarter than the good old ones, and can probably catch even the easy-to-miss mistakes.
Yeah we have some AI tools integration, and a human proofreader, it's just that he has to go through so many reviews, too, so it was missed. Next week will be even more fun :)
 
Can someone explain to me what is going on with the power figures? It seems Nvidia has broken something and should be called out on it a lot for a sizeable regression (not on gaming workload).

Is there an error in the data or is that right for the raytracing power usage? Somehow the FE is using 99w more than the MSI 5090 to achieve LESS performance, how is that possible? I know you can claw back some efficiency with lower temperatures but I don't think it's possible to that degree and it would be visible on the other workloads, just seems like a rather large outlier for no logical reason.
 
raytracing power usage?
I looked into this earlier today, and it seems that in Cyberpunk enabling a certain RT setting randomly drags power consumption way down

Not sure why this is happening, probably some game/driver/engine bug
 
Yeah we have some AI tools integration, and a human proofreader, it's just that he has to go through so many reviews, too, so it was missed. Next week will be even more fun :)
Are you also doing 5080 and ever newer Games testing too like FF VII Remake? And also a TLOU w/ FSR 3.1 comparison? :D
 
There will be seemingly RTX 5090 Ti (with possibly 3 slot tick) with Shader Count of 192, that can be seen by the diagram. I wonder about RTX 4080 Ti, 96 or +100?

Nvidia had a full AD102 (144SM, 96MB L2 Cache, 24Gbps GDDR6X, etc). in their lab but they never released it because they had no competition. So unless UDNA in 2026 is really kicking ass don't get your hopes up.
Imo they might release a TITAN with 64 or 96GB GDDR7 or a QUADRO GPU so they can charge $10,000 for it...
 
The test setup picture does not show this card, but some other air cooled card. I would have been interested in how you installed the 2 AIOs? Mainly was each of them intake or exhaust?
 
Back
Top