MSI GeForce RTX 5090 Suprim SOC

[lexluthermiester]

So this card seems to be the overall champ of these first round 5090 cards. The looks could be improved but the performance is solid.

 

[Dristun]

is no one else going to comment on that poverty level PCB on a $2400 GPU? am i the only noticed that MSI is basically selling reference cards like zotac and pny usually do?

Yeah, but most people pray they won't ever have to look at it or disassemble the thing so why bother? There aren't any "enthusiasts" cards at all right now, imo, they're more like "premium" and most of the premium goes towards a beefier cooler and/or bling. You'd have to wait for Kingpin to come back with his stuff, but it's probably going to end up costing even more than Asus' insanity.

 

[Timbaloo]

The noise levels on the normal BIOS are astonishing. The second BIOS though, why call it "Performance BIOS" when it's really only just cooler (and louder) but not faster. Kinda pointless.

 

[Bugsy004]

I often prefer cooler and silent cards, despite losing 2-3 fps. I love quiet gaming, so MSI RTX 5090 Suprim SOC is a good value, but I’m waiting for Gigabyte cards to compare, Gaming OC and Aorus Master. Yes, the price is insane, $2.500+ for a GPU!!!
Let’s see next in line, RTX5080 and RTX5070 and also mobile implementations. I have friends that prefer mobile gaming.

 

[_roman_]

It's four screws, not two.

Plus if you look closely at the picture you posted, the IO plate isn't just a flat piece of metal, it's bent inwards into the cooler shroud itself which will literally hold the cooler up

If you want i can screenshot and make an arrow to point to that - even on your picture.

Where do you see that bend metal sheet. I just checked your picture and saw, ah these are two pieces.

Here you have the proof.

Only two screws on the side where the fragile connectors are for DP and HDMI. And two screws as connection point close by on another side with most likely small metall fins from the slot insert metall. In total most likely four - still a few screws too less in my point of view. They are saving money on screws - this was my thought.

--

Look what ASUS is doing - same graphic card type review from today.

A million screws on the side - not just two.

 

[Flying Fish]

If you want i can screenshot and make an arrow to point to that - even on your picture.

Where do you see that bend metal sheet. I just checked your picture and saw, ah these are two pieces.

Here you have the proof.

Only two screws on the side where the fragile connectors are for DP and HDMI. And two screws as connection point close by on another side with most likely small metall fins from the slot insert metall. In total most likely four - still a few screws too less in my point of view. They are saving money on screws - this was my thought.

--

Look what ASUS is doing - same graphic card type review from today.

A million screws on the side - not just two.

Sigh, ok.....
firstly, look at the teardown shots of the asus. the million screws aren't structural at all. In fact you wouldnt want them to be as they're mostly attached to the IO ports below them. See the first picture of the PCB below, 5 screws screwed onto the IO port surrounds.

 

[Timbaloo]

Sigh, ok.....
firstly, look at the teardown shots of the asus. the million screws aren't structural at all. In fact you wouldnt want them to be as they're mostly attached to the IO ports below them. See the first picture of the PCB below, 5 screws screwed onto the IO port surrounds.

Wow, that's not a good design

Unless you have someone at hand that can re-solder those connectors

 

[Bwaze]

"The MSI GeForce RTX 5090 Suprim SOC will sell for $2400."

I think with all the news of extremely limited stock and scalpers already rearing their ugly heads such definitive statements might be quite a bit off their mark...

 

[lexluthermiester]

The noise levels on the normal BIOS are astonishing.

Oh please. Don't exaggerate.

 

[rjc34]

Wow, that's not a good design

Unless you have someone at hand that can re-solder those connectors

Eh, that part is fine. Those connectors have a couple of thick solder bonds each, and more importantly when mounted normally the weight of everything would be pushing the connections towards the PCB.

Part of me wonders if they originally just had the two tabs sticking out from the slot support to hold the weight and maybe found that the weight of the card could sag it enough to misalign the ports with the cover.

 

[newbypcgamer]

Can someone please tell me the difference between the SOC and OC?

 

[Bee9]

I think 32GB VRAM size is decent for Flux and SDXL. 100% will buy this on launch day and tweak the PT to get some extra efficiency. Then it should be the dedicated training machine to replace my old 3090.

 

[jnv11]

is no one else going to comment on that poverty level PCB on a $2400 GPU? am i the only noticed that MSI is basically selling reference cards like zotac and pny usually do?

MSI added one or more fuses and some RGB stuff to the reference card. Out of all of the partner cards that were reviewed at TechPowerUp so far, I was able to find a fuse only on the MSI GeForce RTX 5090 Suprim SOC.

One idea for improving the article would be to point out the protection features that this video card has. For example, I was able to find a fuse on the front of the card near the PCI Express x16 connector at https://www.techpowerup.com/review/msi-geforce-rtx-5090-suprim/5.html . Pointing out the fuses in other locations on the card could really help users check the fuses in case the card shuts down so that they can check the fuses to see if one of them blew. Mentioning what the fuses protect can also help someone discover what could have shorted out and forced the fuse to blow.

 

[jnv11]

I think that I found an error at https://www.techpowerup.com/review/msi-geforce-rtx-5090-suprim/5.html .

The sentence "Powering the 16 GDDR7 memory chips is a seven-phase VRM driven by the same Monolithic MP87993 controller that controls GPU voltage." says that the memory power is controlled by an MP87993. However, the sentence above that sentence that describes the controller that controls the GPU voltage is "A massive 22-phase VRM powers the GPU, highlighted in the red rectangles above. These are controlled by a Monolithic Power Systems MP29816 controller.". MP29816 and MP87993 are not the same model numbers, so there is a contradiction. Also, this sentence in between the two previously cited sentences, "All GPU power phases use Monolithic MPS MP87993 DrMOS.", suggests that MP87993 is a DrMOS, while the first sentence that I cited calls it a controller. This is another contradiction.

I think that this could require an article correction. I understand that there is a lot of pressure to get the articles out as soon as the NDA expires, so I understand that these errors happen.

 

[Vya Domus]

You paying 20% more for better temp and less fan noise, not for the extra perf.

And all that because Nvidia chose to use a crappier cooler vs 4090 FE, amazing. Hyper ripping you off just so you can get a proper cooler for this card.

 

[W1zzard]

I think that I found an error at https://www.techpowerup.com/review/msi-geforce-rtx-5090-suprim/5.html .

The sentence "Powering the 16 GDDR7 memory chips is a seven-phase VRM driven by the same Monolithic MP87993 controller that controls GPU voltage." says that the memory power is controlled by an MP87993. However, the sentence above that sentence that describes the controller that controls the GPU voltage is "A massive 22-phase VRM powers the GPU, highlighted in the red rectangles above. These are controlled by a Monolithic Power Systems MP29816 controller.". MP29816 and MP87993 are not the same model numbers, so there is a contradiction. Also, this sentence in between the two previously cited sentences, "All GPU power phases use Monolithic MPS MP87993 DrMOS.", suggests that MP87993 is a DrMOS, while the first sentence that I cited calls it a controller. This is another contradiction.

I think that this could require an article correction. I understand that there is a lot of pressure to get the articles out as soon as the NDA expires, so I understand that these errors happen.

Fixed. Thank you!

 

[bobbybasketball]

Fixed. Thank you!

Does the SUPRIM use phase change TIM or just regular thermal paste on the GPU? The card looks great but if it only uses paste I’m concerned that it will pump out very quickly on such a large and power-hungry die. This was an issue on some 4090 cards and replacing the stock paste with PTM would solve it completely.

 

[SRS]

"The heatsink is ventilated by three of MSI's latest StormForce axial airflow fans, which aren't just webbed toward the edges, but also feature serrations."

Are these FDB or dual ball bearing? As far as I know, ball bearing fans become louder with time and can be damaged more easily if exposed to shock (e.g. dropped card). The only things ball bearing fans seem to have going for them are that they're less expensive and can be made to produce quite a bit of static pressure.

All reviews of GPUs should specify the fan type. It's actually more important information than the VRM setup, particularly for a card that is praised for being quiet.

 

[lexluthermiester]

As far as I know, ball bearing fans become louder with time and can be damaged more easily if exposed to shock

That's a myth. Ball Bearings are made of chrome steel and not easily damaged. While they can get louder, they are easily cleaned and reoiled.

 

[SRS]

That's a myth. Ball Bearings are made of chrome steel and not easily damaged. While they can get louder, they are easily cleaned and reoiled.

Good to know that the shock issue is a myth. Good to also know that it's the case that they become louder. Who is going to clean and reoil their GPU fans? This is the first time I've heard of this and I have been following tech since the 80s.

Is there any reason to substitute ball bearing fans for FDB fans in a consumer product, other than to reduce the cost of production slightly? Having the fans in a product people spend 20% more in order to get less noise get louder to save the manufacturer of the card a small sum, and potentially requiring them to futz around oiling them... Doesn't add up to me.

 

[lexluthermiester]

Who is going to clean and reoil their GPU fans?

I do. I service all of my fans on a regular basis. It's very easy once one understands the basics.

This is the first time I've heard of this and I have been following tech since the 80s.

Then perhaps you haven't been associating with people who have the right kind of know-how. I learned how to service fans from my grand-dad in the 70's & 80's and then from two jobs in the 90's. Fans are fans, they are all easily serviced and most need it on a yearly or biyearly basis. In my experience it's common knowledge.

Is there any reason to substitute ball bearing fans for FDB fans in a consumer product, other than to reduce the cost of production slightly?

That's a much more involved question. FDB VS BB is a debate that has happened ever since FDBs were created. There are pros & cons to both, but not to any serious degree. In the application of this MSI version of the 5090, it really shouldn't matter much and shouldn't be a deal-breaker either way.

EDIT:
I made a guide about this very subject some time ago.

How to relubricate a fan and/or service a troublesome/noisy fan.

I've seen a number of complaints recently about users experiencing problems with the fans in their system. Most fans are serviceable by design. Whether they're a case fan, a fan mounted on a CPU heatsink or a fan mounted on a GPU heatsink, the vast majority of fans can be serviced. Today, I'm...

www.techpowerup.com

 

[SRS]

I do. I service all of my fans on a regular basis. It's very easy once one understands the basics.

Then perhaps you haven't been associating with people who have the right kind of know-how. I learned how to service fans from my grand-dad in the 70's & 80's and then from two jobs in the 90's. Fans are fans, they are all easily serviced and most need it on a yearly or biyearly basis. In my experience it's common knowledge.

Well, it's not common knowledge nor is it a common expectation. Your experience is anecdotal.

There is also information, if one does a search, that seems to support the concern about shock damaging ball bearing fans, including Sanyo Denki's San Ace manual. How overblown that concern is I don't know. There is a lot of debate about various concerns in these products, like coil whine.

The main point here is that GPU reviews should specify the fan type so buyers, regardless if they have technical omniscience or not, can make an informed decision. There is more relevance to the fan type for a product trying to be sold for 20% above an already high price on the basis of quietness than there is in the VRM configuration.

 

[lexluthermiester]

Well, it's not common knowledge nor is it a common expectation. Your experience is anecdotal.

Your understanding seems out of touch. My experience is common knowledge.

Source?

 

[Bugsy004]

We talk discussions about some fans that can be damaged in some strange situation. RTX5090 it’s not what we expected, it is very power hungry, it is very hot, and it is very expensive. It is not worth buying unless you really need AI or you have a lot of money.
If you drop the video card it is more likely that it will damage the printed board before the fans can be damaged.

 

[SRS]

Source?

Sanyo Denki's San Ace manual.

https://www.sanyodenki.com/archive/document/product/cooling/San_Ace_M0011876.pdf

I am not going to further engage with this, unless someone can explain why most buyers looking to pay 20% more for a product on the basis of its noise profile should be more interested in the VRM configuration versus more specifics about the fan type and the pros and cons of those. That is the main point here. I am not interested, at all, in a contest about who is more of a fan expert. This happens every time fans come up and I was bored of it years ago when I first experienced this phenomenon.

If there is no substantive difference between dual ball bearing fans and FDB fans, then the review can state that, along with what type the product uses.