Does AMD Graphics such Radeon VII or RX5700XT support to acclerate TensorFlow?

[Spirit`yL]

Well,guys.Now i am a new Junior College student that i am studying BigData Anylize for major and my teacher says we will gonna use Tensor Flow as a AI Architecture for Data Anylizing and seems like now my laptop wich is a GTX1650 laptop doesnt run it fluently.So i need to expand a New graphics card with a grpahics extander.

I know,i know,my classmate always says AMD YES!but here is a thing that some fellow students says AMD doesnt has SDK to acclerate Tensor Flow on MainStream card like Radeon VII or new RDNA Based card

and our dormitort just support for a whole room with 500w power consumprtion.so there is two choice for me:Go RTX2070 or Go Radeon VII

If you are me,how should you chose?

[Zagna]

AMD has ROCm & ROCm Tensorflow. No support for Navi/RDNA yet.

So Radeon VII would be supported with those it seems like.

Whether its faster than nVidia, depends on workload.

[Spirit`yL]

34 minutes ago, Zagna said:

AMD has ROCm & ROCm Tensorflow. No support for Navi/RDNA yet.

So Radeon VII would be supported with those it seems like.

Whether its faster than nVidia, depends on workload.

But seems like AMD says that only radeon instinct would be fit on ROCm,doesn it?

[Zagna]

2 minutes ago, Spirit`yL said:

But seems like AMD says that only radeon instinct would be fit on ROCm,doesn it?

Quote

ROCm officially supports AMD GPUs that use following chips:

• GFX8 GPUs
  • “Fiji” chips, such as on the AMD Radeon R9 Fury X and Radeon Instinct MI8
  • “Polaris 10” chips, such as on the AMD Radeon RX 580 and Radeon Instinct MI6
• GFX9 GPUs
  • “Vega 10” chips, such as on the AMD Radeon RX Vega 64 and Radeon Instinct MI25
  • “Vega 7nm” chips, such as on the Radeon Instinct MI50, Radeon Instinct MI60 or AMD Radeon VII

 

[martward]

If you're buying a card for tensorflow I'd really stick with Nvidia, mostly because it's supported and will be in the future. They also support some other things like half precision on Nvidia rtx cards. Personally I'd get a 2060 super or 2070 (super) since these have 8 GB of vram. 

[Spirit`yL]

3 hours ago, Zagna said:

 

OK,Got it,but i wanna ask is AMD more efficency or Nvidia`s?

[Spirit`yL]

3 hours ago, martward said:

If you're buying a card for tensorflow I'd really stick with Nvidia, mostly because it's supported and will be in the future. They also support some other things like half precision on Nvidia rtx cards. Personally I'd get a 2060 super or 2070 (super) since these have 8 GB of vram. 

Well 2070S is good buta my dormitory only could run 500w as total(because my school is limit),and my laptop processor which is Intel Core i5 9300H is 45w,i mean will i break the powerwall if my other two roommates are both charging there phones

[martward]

1 hour ago, Spirit`yL said:

Well 2070S is good buta my dormitory only could run 500w as total(because my school is limit),and my laptop processor which is Intel Core i5 9300H is 45w,i mean will i break the powerwall if my other two roommates are both charging there phones

Probably not, but you can always get a 2060 super which consumer less. You can also get a gtx card, but those lack some of the features of an rtx card. I used a 1060 during my thesis, it worked fine and was already way faster than using the CPU. Just pick an Nvidia card with at least 6gb of vram, preferably 8gb and if possible an rtx card.

[Spirit`yL]

22 hours ago, martward said:

Probably not, but you can always get a 2060 super which consumer less. You can also get a gtx card, but those lack some of the features of an rtx card. I used a 1060 during my thesis, it worked fine and was already way faster than using the CPU. Just pick an Nvidia card with at least 6gb of vram, preferably 8gb and if possible an rtx card.

Ok got it

[tech.guru]

Telsa cards are designed for AI and machine learning with Tensor flow.

 

If you can't afford you can rent a virtual machine from azure. It probably be most cost effective and fastest for your needs.

[gwj99]

I was unable to get Rocm running on my vega 64 card, so I was using Tensorflowcpu with my Threadripper 1920x. Since selling the Vega 64 and buying a RTX 2070 - using Tensorflow Gpu and my time/epoch is approximately 20x faster.