The strongest Al supercomputers you can rent right now

Jason Voorhees

Jason Voorhees

Say cheese
Joined
May 15, 2020
Posts
78,887
Reputation
229,642
Here are my previous threads on how to become an AI powered Devs.



Now the biggest roadblock you'll run into is how to train massive machine learning models. Because they require a lot of compute power much much more than whatever your current PC has. If you aren't a kid/ learning and are doing serious work that you are getting paid for you absolutely need enterprise hardware

After burning thousands of dollars across AWS, GCP, Lambda. the best cloud provider for raw power is by far in my experience is

CoreWeave


They have the newest fastest chips from Nvidia H100, H200, and apparently the brand new Blackwell ones that just came out.


1000127521

They charge per hour for the whole pod (a group of chips). It's on-demand pay as you go but you can save big with commitments. No hidden fees for data moving in/out like some other retarded services.


GPUs offered

H100- Kind for most jobs.
In the pod you get 8 chips, 80GB memory each, tons of storage. Around $49.24 per hour.

H200- An upgrade over H100 with 141GB memory. Good for handling huge files/videos.

B200- The newest and most powerful pod on offer. ~2x faster than H100 for next-gen Al but also expensive. Almost $70 per hour.

They also have a budget option for A100 that is super cheap like $20 for testing stuff

Also for networking and storage prices are super cheap like $0.015-$0.06 per GB/month and they have Zero Egress Migration program meaning moving data around is totally free no surprises. Extra IPs or fancy connections it is $4/month or $1,250+ for big links which you realistically don't need.

How to rent it without spending crazy money


1. Don’t click “pay per hour" shit. That is useless. On-demand for H100 pod (8 chips: $49.24/ hour around ~$6.15 per chip/hour. Too much if you're doing real client work or not a broke bitch testing toys


2. Pick one of these two

Plan A: Reserved Capacity

-Promise to use them for at least a few months (flexible terms).

-Price drops up to 60% like H100 pod $19.70/hour ($2.46 per chip/hour, half off).
You can still scale/cancel with notice after initial commit.


Plan B: Bigger Reserved Blocks

Buy a big bucket of hours in advance (like a gift card for compute).

Up to 60% off -H100 effective ~$1.85-$2.46 per chip/hour depending on block size.

Use anytime in the next year I grabbed a 500-hour block on H100s and it ran ~$1,200-$1,300 total way better than alternatives

3. Extra free money
When you sign up, message their sales team and say I’m new, can I get some free credits? And they generally give it to you.

Real example a qhat friend did last month:
Trained a super smart image generator (similar to Midjourney)

On CoreWeave with 60% reserved discount: ~$1,180 total.

Same job on Google Cloud Practitioner (A3 equivalent) would have been ~$5,400.

Same job on Amazon (p5) would have been $7,900.

If you ever want to play with really big AI stuff or your company does just go to coreweave.com make an account, and pick one of the cheap plans above.

That’s it. Everyone else is slower and more expensive in 2025.
 
Last edited:
  • +1
  • Love it
Reactions: SplashJuice, Mogs Me, Leo and 5 others
How much water does it waste?
 
  • +1
Reactions: BigBallsLarry, chadVSvirgin, MiserableMan and 1 other person
  • +1
Reactions: BigBallsLarry, MiserableMan and Glorious King
@Glorious King @imontheloose
 
  • +1
Reactions: imontheloose and Glorious King
imagine using an super computer just to rot on forum @Master :feelshah:
 
Last edited:
  • +1
Reactions: BigBallsLarry, MiserableMan and Jason Voorhees
btw will u be using any of these in future?
 
  • +1
Reactions: BigBallsLarry, MiserableMan and Jason Voorhees
Here are my previous threads on how to become an AI powered Devs.



Now the biggest roadblock you'll run into is how to train massive machine learning models. Because they require a lot of compute power much much more than whatever your current PC has. If you aren't a kid/ learning and are doing serious work that you are getting paid for you absolutely need enterprise hardware

After burning thousands of dollars across AWS, GCP, Azure, Lambda. the best cloud provider for raw power is by far in my experience is

CoreWeave


They have the newest fastest chips from Nvidia H100, H200, and apparently the brand new Blackwell ones that just came out.


View attachment 4388167
They charge per hour for the whole pod (a group of chips). It's on-demand pay as you go but you can save big with commitments. No hidden fees for data moving in/out like some other retarded services.


GPUs offered

H100- Kind for most jobs.
In the pod you get 8 chips, 80GB memory each, tons of storage. Around $49.24 per hour.

H200- An upgrade over H100 with 141GB memory. Good for handling huge files/videos.

B200- The newest and most powerful pod on offer. ~2x faster than H100 for next-gen Al but also expensive. Almost $70 per hour.

They also have a budget option for A100 that is super cheap like $20 for testing stuff

Also for networking and storage prices are super cheap like $0.015-$0.06 per GB/month and they have Zero Egress Migration program meaning moving data around is totally free no surprises. Extra IPs or fancy connections it is $4/month or $1,250+ for big links which you realistically don't need.

How to rent it without spending crazy money


1. Don’t click “pay per hour" shit. That is useless. On-demand for H100 pod (8 chips: $49.24/ hour around ~$6.15 per chip/hour. Too much if you're doing real client work or not a broke bitch testing toys


2. Pick one of these two

Plan A: Reserved Capacity

-Promise to use them for at least a few months (flexible terms).

-Price drops up to 60% like H100 pod $19.70/hour ($2.46 per chip/hour, half off).
You can still scale/cancel with notice after initial commit.


Plan B: Bigger Reserved Blocks

Buy a big bucket of hours in advance (like a gift card for compute).

Up to 60% off -H100 effective ~$1.85-$2.46 per chip/hour depending on block size.

Use anytime in the next year I grabbed a 500-hour block on H100s and it ran ~$1,200-$1,300 total way better than alternatives

3. Extra free money
When you sign up, message their sales team and say I’m new, can I get some free credits? And they generally give it to you.

Real example a qhat friend did last month:
Trained a super smart image generator (same power as Midjourney)
On CoreWeave with 60% reserved discount: ~$1,180 total.

Same job on Google Cloud Practitioner (A3 equivalent) would have been ~$5,400.

Same job on Amazon (p5) would have been $7,900.

If you ever want to play with really big AI stuff or your company does just go to coreweave.com make an account, and pick one of the cheap plans above.

That’s it. Everyone else is slower and more expensive in 2025.
I would like to sex you
 
  • +1
Reactions: Jason Voorhees
what's the purpose
 
  • +1
Reactions: BigBallsLarry and Jason Voorhees
Here are my previous threads on how to become an AI powered Devs.



Now the biggest roadblock you'll run into is how to train massive machine learning models. Because they require a lot of compute power much much more than whatever your current PC has. If you aren't a kid/ learning and are doing serious work that you are getting paid for you absolutely need enterprise hardware

After burning thousands of dollars across AWS, GCP, Azure, Lambda. the best cloud provider for raw power is by far in my experience is

CoreWeave


They have the newest fastest chips from Nvidia H100, H200, and apparently the brand new Blackwell ones that just came out.


View attachment 4388167
They charge per hour for the whole pod (a group of chips). It's on-demand pay as you go but you can save big with commitments. No hidden fees for data moving in/out like some other retarded services.


GPUs offered

H100- Kind for most jobs.
In the pod you get 8 chips, 80GB memory each, tons of storage. Around $49.24 per hour.

H200- An upgrade over H100 with 141GB memory. Good for handling huge files/videos.

B200- The newest and most powerful pod on offer. ~2x faster than H100 for next-gen Al but also expensive. Almost $70 per hour.

They also have a budget option for A100 that is super cheap like $20 for testing stuff

Also for networking and storage prices are super cheap like $0.015-$0.06 per GB/month and they have Zero Egress Migration program meaning moving data around is totally free no surprises. Extra IPs or fancy connections it is $4/month or $1,250+ for big links which you realistically don't need.

How to rent it without spending crazy money


1. Don’t click “pay per hour" shit. That is useless. On-demand for H100 pod (8 chips: $49.24/ hour around ~$6.15 per chip/hour. Too much if you're doing real client work or not a broke bitch testing toys


2. Pick one of these two

Plan A: Reserved Capacity

-Promise to use them for at least a few months (flexible terms).

-Price drops up to 60% like H100 pod $19.70/hour ($2.46 per chip/hour, half off).
You can still scale/cancel with notice after initial commit.


Plan B: Bigger Reserved Blocks

Buy a big bucket of hours in advance (like a gift card for compute).

Up to 60% off -H100 effective ~$1.85-$2.46 per chip/hour depending on block size.

Use anytime in the next year I grabbed a 500-hour block on H100s and it ran ~$1,200-$1,300 total way better than alternatives

3. Extra free money
When you sign up, message their sales team and say I’m new, can I get some free credits? And they generally give it to you.

Real example a qhat friend did last month:
Trained a super smart image generator (same power as Midjourney)
On CoreWeave with 60% reserved discount: ~$1,180 total.

Same job on Google Cloud Practitioner (A3 equivalent) would have been ~$5,400.

Same job on Amazon (p5) would have been $7,900.

If you ever want to play with really big AI stuff or your company does just go to coreweave.com make an account, and pick one of the cheap plans above.

That’s it. Everyone else is slower and more expensive in 2025.
Mr. Musk said none of us will have to work in 20 years. May God help us. That will usher in unprecedented violence.
 
  • +1
Reactions: BigBallsLarry and Jason Voorhees
what's the purpose
Training AI models, running protein folds, predicting weather and climate, astrophysical phenomena etc. for any complex scientific or engineering problem that would take ordinary computers years to finish.
 
  • +1
Reactions: BigBallsLarry, MiserableMan and Terrortheplug
Mr. Musk said none of us will have to work in 20 years. May God help us. That will usher in unprecedented violence.
I like the idea of universal basic income
 
  • +1
Reactions: BigBallsLarry
Training AI models, running protein folds, predicting weather and climate, astrophysical phenomena etc. for any complex scientific or engineering problem that would take ordinary computers years to finish.
how do you rent and how much does it cost
 
  • JFL
Reactions: Jason Voorhees
@imontheloose does your line of work use cloud infrastructure?
 
  • +1
Reactions: BigBallsLarry
Read the damn thread nigga. I mentioned exactly how you can keep costs minimum
boi. i work in saas it just something i do on the side I'm trying to find a better ai then yolov6 to run computer vision do you know any
 
  • +1
Reactions: Jason Voorhees and BigBallsLarry
boi. i work in saas it just something i do on the side I'm trying to find a better ai then yolov6 to run computer vision do you know any
Lol, nerd. I used to shove broomstick handles up the butts of kids like you in the janitor's closet back in highschool. You and your fancy computer talk. Have fun getting settled for by a girl who's had her guts rearranged by someone like me who wouldn't have her.
 
  • JFL
  • +1
Reactions: BigBallsLarry and Jason Voorhees
boi. i work in saas it just something i do on the side I'm trying to find a better ai then yolov6 to run computer vision do you know any
Isn't there a Yolov12 version too. I remember reading about it a few months back. You could also use swin transformer for features extraction as a support depends on your use case tho
 
Last edited:
  • +1
Reactions: Terrortheplug and BigBallsLarry
Lol, nerd. I used to shove broomstick handles up the butts of kids like you in the janitor's closet back in highschool. You and your fancy computer talk. Have fun getting settled for by a girl who's had her guts rearranged by someone like me who wouldn't have her.
I literally mog you to oblivion
 
Isn't there a Yolov12 version too. I remember reading about it a few months back. You could also use swin transformer depends on your use case tho
i didn't realize there was a yolov12 i will look into it
 
  • +1
Reactions: BigBallsLarry and Jason Voorhees
I will 100% never in my life need this information, but still thanks.

It's nice to know that no matter what advice i need, i can just search up "by @Jason Voorhees 'x problem'" and get a detailed guide on it.. It's like reddit! :soy:
 
  • +1
  • JFL
Reactions: 0 bone mass, Leo, Terrortheplug and 1 other person
I will 100% never in my life need this information, but still thanks.

It's nice to know that no matter what advice i need, i can just search up "by @Jason Voorhees 'x problem'" and get a detailed guide on it.. It's like reddit! :soy:
I try to help niggas because I like sharing my knowledge. I share all this so you don't make the same mistakes I do.
 
  • +1
Reactions: Lefty Rankin, BigBallsLarry and Terrortheplug
  • +1
  • JFL
Reactions: Jason Voorhees, Terrortheplug and BigBallsLarry
I try to help niggas because I like sharing my knowledge. I share all this so you don't make the same mistakes I do.
You're a good guy jason. You've got your life all together and yet you're still sharing precious and FREE advice to incels that will probably never make use of it - just incase there's this one person that really needs it. That's how great men are made. I appreciate you being here a lot.
 
  • Love it
  • +1
Reactions: 0 bone mass and Jason Voorhees
@Terrortheplug
I just looked it up Yolov12 is around but it's a research/experimental fork its not very stable. But YOLOv9 is. YOLOv9 is the last official and reliable one people are actually using in deployment. So yeah that and swin should be enough also use black well gpu if you have the budget. Nvidia mentions on its site that ir has a dedicated Second Generation Transformer Engine. So it will probably run much better on blackwell GPUs
 
  • +1
Reactions: Terrortheplug
@Terrortheplug
I just looked it up Yolov12 is around but it's a research/experimental fork its not very stable. But YOLOv9 is. YOLOv9 is the last official and reliable one people are actually using in deployment. So yeah that and swin should be enough also use black well gpu if you have the budget. Nvidia mentions on its site that ir has a dedicated Second Generation Transformer Engine. So it will probably run much better on blackwell GPUs
thank you
 
  • +1
Reactions: Jason Voorhees
@gooner23 @Leo @Pakicel @SplashJuice
 
  • +1
Reactions: gooner23 and Leo
Damn, we're going to need nuclear power everywhere to keep running these things.
 

Similar threads

TrueEveSlayer
Replies
38
Views
1K
DarkTriad.exe
DarkTriad.exe

Users who are viewing this thread

Back
Top