I just bought a tensorbook today. I got the Max version, which comes with RTX 2080. I ordered it this morning and it arrived the same day! (I live near SF.)
I like it so far - it seems fast, and is not as heavy as I feared. Keyboard and screen are good. Unlike @larrykirschner I was able to connect an external monitor (standard LG model) via HDMI just fine. (I also connected my mac keyboard via USB-C and my mouse via bluetooth.) And it’s great to have a real unix machine, instead of the somewhat bloated macbook I normally use.
Tensorflow and pytorch work out of the box, but the TF version is v1 not v2. (They told me they will update to TF2.0 in 2 weeks.) Also sklearn was not totally up to date, and the system python is 2.7 (it has python3 installed too but it’s not the default). So I made a miniconda3 environment and installed the latest versions of all the stuff I wanted (Spyder, TF, PyTorch, Jax, texlive-full, etc). It took a few hours but was fairly easy.
IMHO, the main advantage of tensorbook is that most gaming laptops come with windows, and installing ubuntu and cuda on them is non trivial. The tensorbook Just Works. But if you’re a hacker, there may be cheaper solutions.
@jeremy is GTX 1660 Ti 6GB good enough? also, any1 what’s the difference b/w ti, Q-max, RTX, GTX ?? Like which one is needed for Deep Learning?
Also, what kind of training is a laptop actually able to handle. I have seen most people say that get a decent laptop and cloud GPU, and I have > $500 GCP credit. So, those who have done work on laptop, how does it exactly help in deep learning? Like does it help in the initial data exploration phase, or maybe training as well when dataset is small , or Kaggle? Anecdotes would be nice. @init_27 How’s been your experience doing ML/DL on laptop?
The current trend now is to get a ok laptop with some min requirement + a cloud GPU instance such as colab , GCP or AWS. Take note that if you are using a GPU instance, you will quickly blow through your free credits. The thing to consider is how much of a processing you need for experimentation before pushing to cloud to train fully. For image related application, I am currently using RTX2070 with 8GB locally before pushing anything to cloud for more heavy duty training. Most laptop with GPU will definitely encounter thermal issue during training hence some here suggested using a eGPU instead.
with the GTX1660 Ti you should be able to implement some training with small batch size though convergence will take longer due to the small batch size. if you dataset is not that huge, it should work fine.
Laptops are better in cases wherein one is using pretrained model/weights ,i am guessing. May i know, which laptop model you use?
I mostly saw MSI laptops with 8 gb/16gb VRAM(gpu).
I’ll be mainly using it for Kaggle, hackathon projects, personal projects, fastai courses.
I use an aftershock apex 17. This is a rebrand of the cleo/tongfang laptop, similar in specs to MSI and Dell. I think for image related stuff, I dont recommend going below 8Gb if your budget allow. MSI laptop are definitely cheaper although getting it to working will take a bit of time. I prefer anything that works nicely out of the box. With these laptops you wouldnt want to carry it everywhere as they might weigh a little so do take that into consideration. lastly is the battery. that been said, this laptop of yours should usually be pluck into the power at most time.
I am considering an Oryx Pro from System 76 in the $1.5-3k price range, which supports both the hardware and software (with their Ubuntu-like Pop!_OS). I have dealt with many headaches getting Ubuntu to work nicely with hardware and it seems like System 76 is in a good position to eliminate/mitigate these issues. I haven’t been able to find many reviews, but those that I have found are overwhelmingly positive.
Does anyone here have any experience with System 76’s computers or with their Pop!_OS?
Hi everyone,
as a quick followup I want to say what I decided on:
I decided to get the Gigabyte Aero 15 v8, initially this one was way out of my price range but suddenly a new version came out and the price dropped from 2800 euros to 1999. I had to get it. Specs are:
CPU: 8750H 6 cores / 12 Threads
GPU: RTX 2070 max-q with 8 GB VRAM
RAM: 32 GB 2667 MHz Samsung RAM
Harddrive: 1000 GB Intel NVME
battery: 94 Whrs
weight: 2.1 kg
Initially this machine had 16 GB RAM, but due to ongoing promo they offered a free 16 GB stick on top. Also I like that they are generally selling single channel RAM configs, meaning when u get 16 GB you have only 1 RAM slot used and can easily upgrade to 32, whereas with most laptops you get 16 GB through 2x8 GB sticks and then you have to buy another 2x16 GB for upgrading. Plus upgrading is super easy, only a few screws for RAM and NVME (aero has 2 NVME slots) upgrade. Battery is also massive. There are also other custom cheap systems with the same specs but often rly lack the battery life for when you are on the go like the tensorbook for example.
As a quick note, I was also looking at the GS65 and the Acer Predator Triton 500. They are quite similar and weigh 200-300 grams less. If u want to upgrade any of those though (ram or ssd for example) you have to do a complete disassembly because the motherboard is upside down.
FINALLY, CAN IT RUN UBUNTU?
It does, i just got it working yesterday, not rly a big hassle. I can also switch between internal and external graphics depending on whether I need more power or battery life. I still have to do some testing with battery life but so far everything seems to be just fine on ubuntu. For installations quick guide:
Download latest Ubuntu 18.04
Make bootable USB with RUFUS
Change to boot from USB drive in BIOS on startup
Highlight the option ‘Try Ubuntu without installing’ and press the E key.
Add nouveau.modeset=0 just behind quiet splash like this ..... quiet splash nouveau.modeset=0 -- and press F10 to boot.
U will be asked to create a MOK key while installing Ubuntu that u have to reenter once ur restarting the Laptop to confirm the driver changes. Lastly install the Nvidia driver by opening Software&Drivers, go to additional drivers tab and select Nvidia proprietary driver. Then restart again.
I just saw this new laptop that also comes similar in specs and at a relative cheap price. (around 2.9k SGD for 1 TB and 32 GB ram on Amazon)
what is important is that its thermal cooling is really good and in a relative small form factor. The laptop has a rtx2070 max q in it (not sure if this fits the requirement for some.). It packs a 93 Wh battery and quite similar to the AERO 15.
So far there is only 1 review but do check this out! (Sadly it hasn’t start shipping to Singapore but its available for pre order in the US)
If anyone can get the hands on it, please give a review here!
I think I’ve had a good amount of experience with my laptop and a rig as well do derive some thoughts (I’ve used both for >6 months now).
Would anyone be interested in doing a discussion about Laptops for DL for a podcast conversation to create essentially an audio/video version of all the great advices here?
Hi @MJPansa!
I am considering also the Gigabyte Aero. Any comments after this time using it? Issues installing DL related software? Did you try the thunderbolt port for eGPU? Battery life is more or less what expected?
Thanks!
@Joan
Iam very pleased with it so far running dual boot, doing everything DL related on Ubuntu works great. I have also let it run epochs over night without running too hot, that was one of the fears i had generally with laptops. One downside though, but that is valid for almost all laptops I think is the fans. There is no way of changing fanspeed manually in Ubuntu. You can only change fan modes with the gigabyte software in windows and when u reboot and log into ubuntu it seems to keep this profile. So for longer training i would recommend the gaming profile for example where fans ramp up quite significantly. Theres also a quite mode, which can be quite useful when onl;y doing light programming or little testing, then fans are rly silent even when load is sometimes at 100%.
One more useful thing is CPU undervolting, can also be done easily in windows with Intel XTU. Keeps the CPU from throttling down, but since cpu and gpu heatpipes are connected it has also a cooling effect on ur gpu.
@init_27
Cool idea! I would be happy to help since I spent a lot of hours researching every laptop there is on the market. However Iam busy with exams until next month. Let me know what ur timeframe for this is.
hey Sanyam,
may i know which laptop have you been using ? I myself will be purchasing one in 2 weeks( Flipkart Big Billion Days). I think it’s basically going to come down to some Acer model. Personally, as it’ll be a powerful laptop, heat/cooling/fan speed is the main issue for me.
And the GPU, of course. 6 GB to 8GB = lot of price diff.
Also, have you written any blogpost about your rig build, cost , experience?
EDIT: never mind, I found out that you purchased HP Omen. and about the heating situation too.
But 1 question remains. is it better to get a GPU with 8 GB VRAM or a slightly better GPU with 6 GB VRAM? I am convinced that making a rig is the best option now, but, for now I am buying a laptop so, gonna make sure it balances all variables, in case I don’t buy a rig.
My work type: Kaggle, personal projects, fastai.
I guess, when I see 6GB vs 8GB, i basically want to know what kind of tasks 6GB won’t be able to do, or would be slow at.
Also, is there a specific preference in terms of RTX vs GTX in Deep Learning? I think the GPU models 1080,2080 have messed up my mind , I’ll start looking them up(benchmarks etc) with a fresh mind.
You’re really going to be limited to what kind of work you’ll be able to do with a laptop when it comes to training neural networks. Even a modest size training dataset will eat up 8GB of gpu memory (the most you’ll probably be able to get in a laptop) and slow things down. The reality is that you’re probably better off building a desktop and accessing it with a remote desktop client or ssh (Chrome remote desktop works pretty well). Gaming laptops are generally not upgradeable and you’re stuck with what you buy. Desktops can be upgraded over the years, getting you a lot more value for your money.
Also, you can log into it with any old laptop that you already have.
Why not setup dynamic DNS on that system and ssh into it remotely from anywhere in the world? That’s assuming you can leave it powered on and connected to the internet when you’re away
@srishti1 overall it works nice. I used to struggle doing dual boot on the laptop (having countless flickering screen issue), but with 20.04 Ubuntu LTS released, I can dual boot it in my laptop very nicely.
Note that the GPU power is incomparable to desktop machine as it is not optimized for performance. The laptop can output a power of ~80W for its GPU at max (RTX 2070 Max Q), while desktop machine with the same GPU can easily get to 120W+.
One drawback of this laptop is short battery duration. Without charger, it could only operate for ~2 hrs at max and the charger is as big as a brick so it is kind of inconvenient
Hi all - I’m going to treat myself to a new laptop dedicated to Zwifting this holiday season. There are many options online. What are your recommendations? My budget is around $1,300 so hoping to get a good Black Friday deal on something with high-end CPU & GPU specs.
I’ve had an iPad and an older HP laptop that has caused a lot of pain for me. Please don’t recommend Apple TV, a TV isn’t an option for my Zwift cave right now.
, I’ll start looking them up(benchmarks etc) with a fresh mind.