Budget 90k+ Air-cooled RTX 3090?

Emrebel

Adept
My friend is looking into buying an RTX 3090 for AI/ML work. I wanted some suggestions about air-cool version. How reliable is it for such a power hungry card? I assume it will generate a lot of heat and the workload will also be high based on the tasks.

Also if you guys have any other suggestions please let me know
 
My friend is looking into buying an RTX 3090 for AI/ML work. I wanted some suggestions about air-cool version. How reliable is it for such a power hungry card? I assume it will generate a lot of heat and the workload will also be high based on the tasks.

Also if you guys have any other suggestions please let me know
Technically, all RTX 3090s which are not watercooled, are "air cooled", aren't they?
Did you mean passively cooled?

LTT did a video with a Monster Labo case which could work with a RTX 3080, but not a 3090. This INSANE Gaming PC Has NO MOVING PARTS - MonsterLabo The Beast
 
My friend is looking into buying an RTX 3090 for AI/ML work. I wanted some suggestions about air-cool version. How reliable is it for such a power hungry card? I assume it will generate a lot of heat and the workload will also be high based on the tasks.

Also if you guys have any other suggestions please let me know
3090 has memory chips at back side also so it outputs a lot more heat than 3080ti. most people put in after market pads to decrease the mem temps otherwise it throttles. Don't know what load does AI/ML takes but have used my friends 3080ti FE for gaming for 3-4 hrs straight and it has memory temps maxing at 84c and gpu temps at 70c with fans 50-60%. I think AIB cards will be much better at temps.
 
Technically, all RTX 3090s which are not watercooled, are "air cooled", aren't they?
Did you mean passively cooled?

LTT did a video with a Monster Labo case which could work with a RTX 3080, but not a 3090. This INSANE Gaming PC Has NO MOVING PARTS - MonsterLabo The Beast
Ya I meant passively cooled versions

3090 has memory chips at back side also so it outputs a lot more heat than 3080ti. most people put in after market pads to decrease the mem temps otherwise it throttles. Don't know what load does AI/ML takes but have used my friends 3080ti FE for gaming for 3-4 hrs straight and it has memory temps maxing at 84c and gpu temps at 70c with fans 50-60%. I think AIB cards will be much better at temps.
It is more about memory available for the AI/ML on that gpu which is double than the other cards.

I will ask for the current configuration and share here.
Just he wants to know if it can be an easy swap like other cards or it will invite any headaches.
 
My friend is looking into buying an RTX 3090 for AI/ML work. I wanted some suggestions about air-cool version. How reliable is it for such a power hungry card? I assume it will generate a lot of heat and the workload will also be high based on the tasks.

Also if you guys have any other suggestions please let me know
how many cards he is purchasing. are cooled is not that great if going for 24x7 cooling. get the aio version. like msi sea hawk or something. you can get the founders / normal version and add a 3rd party aio block also. in air cooled version top version like suprim X or asus strix version are the best to buy. super cooling.
 
how many cards he is purchasing. are cooled is not that great if going for 24x7 cooling. get the aio version. like msi sea hawk or something. you can get the founders / normal version and add a 3rd party aio block also. in air cooled version top version like suprim X or asus strix version are the best to buy. super cooling.
Single card only.
How about Zotac cards?
 
in terms of warranty you get 5 years. that's it. temps are high in comparison to other and a bit slow as well.
what I would try is to get my hands on suprim X > Asus Strix > Zotac amp holo exteme > Founders Edition > any other brand.
Task is not strictly 24x7 but maybe sometimes 2-3 day run
 
I know people who are running 2 INNO3D 3090's inside a case for AI/ML and they are doing fine.
He get around 73'C when the machine is doing AI/ML work 24/7. However we tried mining once and the card closest to the CPU would throttle. In general I would say you need not worry about heating during such jobs. And looking above I would say that the 3090 FE has an over engineered heatsink. Let's not short change it. The TDP difference between the 3080ti and the 3090 is very little, but their size difference is huge. Don't worry about it. Infact I also have first hand experience with a MSI 3090 X3 Pro which had serious issues during AI/ML. Atleast avoid this.
 
I know people who are running 2 INNO3D 3090's inside a case for AI/ML and they are doing fine.
He get around 73'C when the machine is doing AI/ML work 24/7. However we tried mining once and the card closest to the CPU would throttle. In general I would say you need not worry about heating during such jobs. And looking above I would say that the 3090 FE has an over engineered heatsink. Let's not short change it. The TDP difference between the 3080ti and the 3090 is very little, but their size difference is huge. Don't worry about it. Infact I also have first hand experience with a MSI 3090 X3 Pro which had serious issues during AI/ML. Atleast avoid this.
Thanks for sharing
 
Just to update wth a full configuration:

1) Ryzen 5950x
2) 64gb Ram
3) RTX 3090
4) PSU : 1000W

What do you think about the above config?
 
All looks good.
But you can save if you so desire.
Any 8.core CPU would do.
A 850w PSU would do.
for AI/ML but it can be subjective.
 
No need, those thermal pads work, just that they would last may be 6 to 12 months. So think about it later unless you are getting a used 3090. The market has started flooding with used cards. I personally know 4 people who got used cards that throttle. In which case go for RMA or you might end up changing the thermal pads.
 
No need, those thermal pads work, just that they would last may be 6 to 12 months. So think about it later unless you are getting a used 3090. The market has started flooding with used cards. I personally know 4 people who got used cards that throttle. In which case go for RMA or you might end up changing the thermal pads.
I was also asking for cpu
Also will it make a much of difference between 5900x and 5950x?
 
Last edited:
This are the specs from the quotation:

67101496-FE35-4CFD-A030-A7BEDE97FA9B.jpeg


I have asked for detail specs on the components but pls see if this if fine.
 
Check out this article, it is one of the detailed information on which GPUs to pick for deep learning setup.


I will suggest you to read the comments especially. You can buy R9 CPU chip but it will be overkill as it will have very least role to play when DL kicks in to action.

I myself use X99 based Xeon chip and added extra RTX GPU as it gives you 40 PCIE lanes to expand. You could get x99 chips for cheaper than R9 and still do better while running inference under DL libraries when you have extra GPU.

Note: it takes time to configure tensorflow or pytorch to run on 2x or 4x GPU.
 
Back
Top