Transcript:
Okay, guys, in today’s video. I’ll do something different. I’ll just do a very fast walkthrough to how to install tensorflow using your AMD graphics card. I struggled with that because I moved from Nvidia and I was using cuda cool dnn and all of these stuff with tensorflow and then I bought the new RX 6800 and then I was stuck with the various errors in tensorflow, including this one which I’ll show you if I type Conda, activate EF2, which is my Tensorflow 2 version, which used to work perfectly fine with my nvt GPU, and if I go into Python import tensorflow right off the bat, you can see something is not really good because it’s loading cudart, but we ignore that, and we do tensorflowtestgpu. Oops, available and what you get is false. That’s probably what you’ll be seeing. If you just pipping installed Tensorflow GPU, it doesn’t really help if you just reinstall Tensorflow GPU. But the good news is that there is an easy fix. The easy fix is if you want, you can drop a new environment, so that kind of deactivate and we will do Conda create dashing and we’ll give it a random name like tf Radeon. I will do Python 3.6 for example. We’ll just wait for this to be created. Then what we’ll do is well activate DF radeon and then extremely easily as Microsoft has explained it in their site, which I will show on the screen right now Is that what we have to do is quite simply type in the following command. So what we do is Pip install. I see like the guest tensorflow then we will use direct ML so direct ML press enter, and we just wait for a couple of minutes for everything to be downloaded. I’ve already downloaded it. That’s why it should be a little bit faster, but we’ll see, and then if we do python and we do import, then server flow will get this precise error, which at first it’s very annoying. Um, but what you have to do is simply do either Peep or I like to use Conda for this condesto numpy. Yes, please, very fast. They need the python import 10 sir flow then serve flow as you see? The error is now gone dot test. DOT is GPU available and, uh, kaboom. We have true. And now we can use tensorflow with this one. The only thing to have in mind which I’m not sure if that’s not even updated is is that I have to double check it now. So that’s what I’ll do is. I believe it’s 1.15 Um, but we’ll check in a second. Is your version of the tensorflow and yes, it’s 1.15.3 Unfortunately, that’s the maximum available weight, Uh, direct ML. But still, you have tensorflow. It’s not 2.0 Yes, but it’s 1.15 which is quite good and it is working on your AMD card as you can see here. AMD Radeon rx 6800 It’s basically able to run any tensorflow script. You had any Kera’s script? You had, so I found this really annoying because when you search for and now I’ll just do an overlay of this. Um, if you search for AMD Windows, Tensorflow or tensorflow windows AMD, however, you want to spin the wording on Google. You get this hugely impractical tutorials where they, uh, they install 100 dependencies. They use some virtual machines just to get it to work with virtualization with whatnot. It’s it’s ridiculous, it’s ridiculous, and it’s so easy, and and I’m not discovering the hot water here or anything. It’s just a matter of going into the official Microsoft site that that’s why it’s bugging me so hard. There are tutorials where they explain. Oh, no, you can’t make it with pure tensorflow you have to use, uh, apply a played ML or something like this. Only for Keras. It’s one medium article. I read where it’s it’s again very long, and I don’t get it. It’s it’s so simple. It’s just literally one command and okay. Okay, I get it. Numpy version. Numpy’s version is buggy with this package. Okay, it’s two commands. It’s not one, but still two commands. Come on. Yeah, sorry, that’s, uh, just one what I wanted to speak about. After, um, my horrible experience. So I googled it for, like, an hour or two before coming up with this. I was quite depressed that I got the new AMD card. I was so excited that I got 16 gigs of Vram, and then I couldn’t basically use it for machine learning, but now I can, so I’m happy. I hope you’re happy too, and if this worked for you, please, like the video or dislike the video. If you don’t like me, gibber jabbering or my explanation, um, or anything else and please put in the comments. What either you would like to see more or if that was helpful or if this worked for you because if it didn’t, um, please let me know. I’ll try to help you and that’s it for today. Thank you very much, guys.