🕵️‍♂️ Engage in discussions ANONYMOUSLY. NO REGISTRATION is required. 💬

✅ Registered accounts earn up to $25 per thread via PayPal and BTC. 🤑 Read more in News and Announcments! Unlock the ability to chat, share, send private messages and earn money per post in our community. Just register an account. It's Free!

SignUp Now!
  • 🕵️‍♂️ Engage in discussions ANONYMOUSLY. NO REGISTRATION is required. 💬 ✅ Registered accounts earn up to $25 per thread via PayPal and BTC. 🤑 Read more... Unlock the ability to chat, share, send private messages and earn money per post in our community. Just register an account. It's Free!

Question Nvidia Tesla k80 vs p40, M40 for deep learning

Choose this one if you want to ask a question
New member
Joined
Aug 18, 2023
Messages
4
Credits
130
I'm choosing a graphics card to start my journey in deep learning. Figured out that Nvidia Teslas would be the best budget-performance option out there. Any thoughts on the mentioned cards - k80, p40 or m80? What would you go for if you were buying a single card?
 
New member
Joined
Aug 20, 2023
Messages
4
Credits
27
I would go with the Nvidia Tesla P40 - it has newer architecture, more VRAM, and better CUDA cores. K80 is older and is basically two K40s stuck together on one board, each having 12GBs of VRAM, which means you won't be able to use all the VRAM simultaneously. Also, the P40 offers improved performance, newer architecture, and better compatibility with current software.
 
  • Great post
Reactions: w_G
New member
Joined
Aug 19, 2023
Messages
3
Credits
57
You must have 10 posts to see the links. Currently you had 0 posts.
:
  • Architecture: Based on Kepler architecture.
  • GPU Cores: Equipped with two GPU cores.
  • Memory: Features 24 GB of GDDR5 memory.
  • Double Precision Performance: Offers up to 2.91 Teraflops with NVIDIA GPU Boost.
  • Single Precision Performance: Provides up to 8.73 Teraflops with NVIDIA GPU Boost.
  • Memory Bandwidth: Boasts 480 GB/s aggregate memory bandwidth.
  • Purpose: Designed for demanding computational tasks requiring high compute performance and data throughput.
  • Suitability: Suitable for workloads needing high data throughput and precision.
You must have 10 posts to see the links. Currently you had 0 posts.
:
  • Architecture: Built on the newer Maxwell architecture.
  • GPU Cores: Includes a single GPU core.
  • Memory: Comes with memory suitable for the architecture.
  • Double Precision Performance: Offers lower double precision performance due to design trade-offs in Maxwell architecture.
  • Single Precision Performance: Delivers up to 7 Teraflops with NVIDIA GPU Boost.
  • Purpose: Specifically designed for deep learning training tasks.
  • Suitability: Perfect for deep learning training tasks but not optimized for applications relying on high double precision operations.
If your primary goal is to accelerate deep learning training tasks, the M40, with Maxwell architecture and single precision performance, should be the choice!

The K80s are good dual GPUs, but they usually use them for physics calculations and deep learning not so much.
 
Moderator
Joined
Aug 18, 2023
Messages
24
Credits
294
Right now, I'm looking at Nvidia Tesla p40s, but the fact they need active cooling, and there are no custom water blocks or cooling solutions apart from those 3D-printed exhausts that they sell on eBay is just sad. I think I'll wait a bit more for consumer GPU hardware. I plan to use that card for video editing and Davinci Resolve since it has plenty of RAM.
 
New member
Joined
Aug 20, 2023
Messages
1
Credits
7
I bought a Tesla P40 and modified the Kraken G12 water cooler to cool it off. Just remember if you are doing this, you have to make sure the VRAM is also cooled. This was a very budget-friendly solution, and I can currently play games on it, and stream. It's a perfect low-budget solution for a single card.
 
Moderator
Joined
Aug 18, 2023
Messages
24
Credits
294
Just a quick heads-up, folks! I've recently made a tech upgrade, opting for the AMD RX 7800 XTX 24GB. Tired of dealing with Nvidia's recurrent issues like their GPUs causing Molex connectors to overheat and risking system fires, despite having a robust 1000W Platinum PSU. It's not just a one-off problem; newer Nvidia models seem to exacerbate these issues. Additionally, Nvidia's penchant for locking down their hardware and pushing proprietary software has left me seeking refuge in the open embrace of Linux, fully powered by the AMD RX 7800 XTX 24GB. Here's to smoother, safer computing ahead!
 
Top