Advice for Purchasing laptop

Thanks for the suggestion.

Actually, My Father was getting some grant from his office so he had generously pitched in the amount for making the 3400$, We did pay the MRP but some amount was covered by the grant, so after much thought that seemed to be a better investment than going for Cloud vendors.
The seconday reason was that my internet is really slow at times so even SSH-ing doesn’t work (It caps out at 50Kbps-this is the best option available in my area-it’s an semi-rural area)

I still have been begging them for a refund-they’ve simply been raising the issues and putting it on priority.

More importantly, any eGPU box provides just four PCIe3.0 lanes (providing more would have little sense, since the bandwidth would exceed thunderbolt 3 maximum bandwidth). You will bottleneck your GPU no matter how powerful your laptop could be.

More interesting would be building a custom eGPU external case using a PCIe 3.0 extension cable.

2 Likes

A few more tips for (any) laptop Practitioners:

  • Using a laptop is like teaching a baby, it’s more fragile, less powerful and you will have to take care of it now and then.
    I experience temperatures of 70 deg Celsius at medium train loads and if you plan to keep the machine at optimum performance for a few years,

Here are some things that you might find helpful:

  • Repasting: Reapplying the Thermal paste helps keeping down the temperature down.
  • Vacuum fans: Vacuum fans are designed to suck away air and helps dip down the temperature by a bit. The areas around the laptop get concentrated with heat and add to the heating so the fans help.
  • UnderVolting: [Advanced] Changing the CPU voltages helps cut down the temp by a HUGE Factor.
  • Air Coditioning: I’m still suprised that this even makes a difference but I can experience a decent difference and almost no throttling when using the system in a decently cooled room.
  • Cooling Pads: These may/may not work for you. In my case, the vents are at strange angles so it doesn’t work for me, but these too help keep down the temperature.

If you apply all these things, first you’ll be able to hold the laptop in your lap without burning clothes. Secondly this will definitely keep the temperatures down by a factor of 5-10degrees (Celsius), keep down throttling and increase the lifespan.

Sanyam

2 Likes

Razer also just unveiled their wicked cool GTX 1070 and 1080 Laptops
(Really portable compared with other contenders) These are around 2cm or so and very very light- 2kgs or so.
Most gaming laptops are around 5k+ or so.

  • Razer Blade
  • Nvme are solid
  • 8th Gen Processor
  • Max Q which means efficient cooling
  • A “Vapor” chamber cooling system=You probably don’t have to worry about heating issues
  • RAM is up-gradable.

Request: Could any admin please edit the original post as being a laptop suggestions thread (The Original body is a discussion of just my situation)?
Since I’ll keep updating this thread continuously with news and good options-given the overpriced warning and issues that come with laptops for anyone who might be forced to using a laptop.

1 Like

Another Fact:

Forget doing DL on the Move if you can’t find a socket to plug in.
The GPUs on laptops are limited when on Battery. There isn’t a work around this, the Battery in my case can’t handle the maximum requirements of the 1070. I’m not sure but in most cases it should be the same.

It indeed is better than using a CPU. But it doesn’t help with, “Hm let me train this huge model while I sip on my coffee before a lecture”. You could probably do the Mini-Experiments that @radek has mentioned Here.

But beware, you won’t have full Power access when on the Move.

Hello everyone!

I found this threat as I am looking for the best option to buy a laptop for ML/DL.

I would prefer a desktop PC but it has to be a laptop due to portability and I want to be able to carry out proper experimentations before using the cloud.

My budget could get until around 2500$.

Thanks in advance!!!

@frodesc One thing I’ve learnt and can say after talking to people with Laptops as their daily driver.
It's a portable CPU, much less of a laptop.

2500$ could get you a really good deal if you’re ordering from the US.

@init_27 you mean that I should save some money for the cloud instead? Is something I was thinking about. Maybe 2500$ is too much.

By the way, I’m ordering from Spain. Sadly. Thanks!

@frodesc there are many trade offs
Performance Vs Portability.
If you can get both then it would be more expensive.

I’m in the same boat as you. I bought a laptop for 2x because of country pricing.

You could go for 2 years with 2500$ when using a cloud instance.
But there is a comfort of having your own machine.

Could you share the prices of the laptops you’re looking at?
If you can get one with a 1070 or 1080, that could be great too.

It’s upto you to make a choice

  1. Laptop+Cloud
  2. Expensive gaming laptop.

@init_27 Sure (The prices and specifications are from a different link in spanish):

These are the ones I was looking at as the best laptops but I might have gone too far and I should go for something cheaper and save money for the cloud or a future desktop PC. What do you think? Is 32Gb RAM a must?

@frodesc
The Tensorbook and Aero have an okay reputation.

I believe an 8th gen processor would be a better choice.
I went with a 16Gb ram that can be upgraded later, allocated it 64gb SWAP to compensate that.
(I have an HP OMEN X Ap045-TX)

It’s up to you. This indeed is expensive and for this much amount you could assemble a PC with a 1080Ti+32 or 64GB RAM.
The thing about PCs are that you can always swap in and out components.
Need a new GPU? Just change it.
Need more RAM? Just put it in.

Cloud is generally slower. Paperspace and Vectordash are good options.

You could do a comparison.
Estimate your cloud usage bill (always keep a higher number).
If this bill<The price of the laptop for the duration you want to use the laptop and the cloud service is as good as the Laptop then you could go with that. Otherwise get a Laptop.

Sanyam

You can also check this guide for best gaming laptops

#1) EVGA SC17 1070 17.3" 4K Gaming Laptop (~$1400 brand new on eBay) original price ~$2800 on EVGA website

#2) Lenovo Flex 5 15 2-in-1 Touchscreen Laptop - Intel Core i7 - GeForce MX130 - 4K Ultra HD - Active Stylus (~$1000 on Costco after $200 off)

#3) Dell XPS 15 4K GTX (starting at $1000)

Anyone had reviews about tensorbook ?
I am currently sourcing for a laptop for my work. I am looking at anything with at least a GTX 1070 and above and I am looking at a budget of lower then 3K USD if possible. I am trying to weigh the price vs my needs.

Budget: Below 3K USD, best is between 2000 - 2500 USD

My needs for work:

  • Mobile, as I will be meeting client
  • I have limited access to cloud resources
  • I do not have a physical office as the nature of the job is remote. (However, it might change in the future and the external GPU would definitely be the top of the list)

I need a laptop to at least allow me to deliver work that deals with images and text data. I expect to have a external storage to work off from there so in terms on storage requirement, its not that strict.

Preferable specs: gtx1070, 32ram, 1TB ssd, i7 8750 processor.

Here are some of my findings:

For the price point of $3081 USD (excluding shipping) for the premium option, its does seem to pack a serious punch with the preloaded DL libraries. However, I am still a little hesitant to go for this because the money I put in is just to ease my time with setting up the machine. I can easily get another laptop that cost less and do the settup myself.

Other alternative that is cheaper but require a little effort in setup is the MSI prestige series.
Specifically, MSI P65 Creator 8RF which cost around $2272.21 USD including shipping and the option to just upgrade the RAM and SSD yourself. Note P65 has 2 SSD slot in total. Similar model includes the MSI GS65 ($2509.32 USD). Another model from MSI is the GS75 which even have 3 SSD slot at the price of $2509.32 USD. However from my previous experience with MSI laptop, there build quality are sub par and I had visited the support centre several times to fix hardware issue (keyboard, screen, web cam connection etc). Plus, they dont come cheap with each repair. Upgrading hardware on my own is a little troublesome as the layout of the laptop hardware is not the best out there. (In my opinion, Dell has the best laptop layout for upgrades)

Looking at models from Dell, Asus and HP, I find that the GPU are usually packages with 16GB ram and 512GB ssd which might be good enough but at the same price point as MSI, u can get much more (32GB ram, 1TB ssd).

External GPU is an option

Right now I am a little bit lost on how to go ahead with this. Anyone with good suggestion that they had tried would be much appreciated.

@AlanK If you train on the laptop, be sure to evaluate how the battery performs.

re the tensorbook:

I bought one, and I regret it.

What I expected was a well designed system that would save me the time of installing and maintaining linux, and I was willing to pay a little bit of a premium to save the time.

The problem/realities are these:

  1. a lot of the supposed value proposition is that tensorflow/pytorch etc come already installed. I never use the host-installed versions of these things, I’m always in practice using a conda env or docker

  2. It can’t drive a second display. I tried connecting with both HDMI and minidisplay. I went to Best Buy and tried like 10 different monitors. I went as far as contacting their support and they were not particularly helpful.

  3. You can’t connect an eGPU because its usbc ports lack thunderbolt support. Sure, the laptop has a built in gpu, but I have use cases where I need 2 gpus–specifically, I’m working on an app that requires two docker containers/microservices and each needs its own gpu. I understand that “most users just use the single gpu that comes in the laptop,” as I was told by lambda support, but why would you sell a super-premium machine-learning $4000 laptop in 2018/2019 and not include thunderbolt/make it impossible to connect an egpu?

Hi Larry thanks for the feedback. I actually didnt go ahead with the tensorbook as my budget really didnt allow me to purchase something that premium.

re: tensorbook
Laptop preferable specs: gtx1070, 32ram, 1TB ssd, i7 8750 processor.
Hi all, I returned after some research and a final purchase.
After much pondering, I decided not to go with the tensorbook as it was just to expensive and the lead time to get the laptop was a little long (I am from Singapore)

A bit more research and I managed to find these laptop that fit my criteria.

  1. MSI P65 Creator 8RF – SGD3308 ~ USD2442
    Specs: i7-8750, 32GB ram, 1TB Samsung 970 Evo Plus sdd, gtx 1070, battery 82whr
  2. MSI GS65 – SGD3295 ~ USD2432
    Specs: i7-8750, 32GB ram, 1TB ssd (2 x 512 Super Raid 4) , gtx 1070, battery 82whr
  3. Dell G7 15 Gaming Laptop (7590) – SGD2899 ~ USD2140 (on offer)
    Specs: i7-8750, 16GB ram, 512GB ssd, rtx 2070, battery 90whr
  4. Alienware 15 r4, (older version with gtx 1070) – SGD2799 ~ USD2066
    Specs: i7-8750, 16GB ram, 128gb ssd + 1TB hdd, gtx 1070, battery 68whr
  5. Aftershock Apex 17 – SGD2699 ~ USD1992
    Specs:i7-8750, RTX 2070, 8GB, 512 nvme ssd, battery 68whr (components are customizable)

In the end, I went with Aftershock Apex 17 as it was the cheapest and the company is also a home grown customized laptop dealers (supporting my own country hahah). Some comments for the others are:

  1. MSI laptop had all the specs that I wanted with reasonable price. However, I had also found online that there are a lot of hiccups during installation of ubuntu on the msi laptop, somehow it does not play well with msi hardware so I decided to skip this even though they are very attractive with the prices and specs offer. Also, I was a owner of a slim msi gaming laptop before and it has brought me hell with the sub par build quality and frequent visit to warranty to get those parts change. However, these recent design seems good and people had pretty good review. There is also a github page that tells us how to install ubuntu on msi laptop so I guess it might be worth a try.

  2. Dell and Alienware was a great option and I was actually having a hard time choosing this and aftershock. Reason been that 1. Dell has better ubuntu support. You know your dell hardware will play well with ubuntu and there is some support. 2. Dell laptop had an amazing build quality and installing your own ram / storage for future upgrade is very easy with their design.3. Excellent thermal design. during the time i was looking, there was an IT show in Singapore and much of the price I had stated was after offer discount so its looks extremely affordable to get it. The only issue I had was the weight. The Dell G7 was weighting at almost 3kg ! (no surprise due to the battery cell it is carrying). If you are not so mobile, then Dell G7 might make some sense. Otherwise, its a great laptop for deep learning.

  3. Lastly we come to Aftershock laptops. They are a relatively small customised laptop builder based here in Singapore. Was actually looking into customized laptop builder and lucky enough found one here. The price they provide seems very competitive and they allow customization of components. However battery is not fantastic in this machine. I opt for the final specs of i7-8750, RTX 2070, 16GB x 1, 512 nvme ssd, battery 68whr (components are customizable) without windows 10 and comes up to around SGD 2655 = 1960, allowing me to have some cash to upgrade it further where rams and storage are cheaper in the future. FYI aftershock laptops are actually clevo / tong fang / eluktronics / wallmart overppowered laptop. All are from the same manufacturer tong fang.

I had been using the Aftershock Apex 17 with ubuntu 18.04 installed for the past 2 weeks and I must say, its performing well with not much of difficulty setting up. all hardware working, Only issue is the backlight customizable is not available in ubuntu but hopefully in the future there will be.

Highly recommend for deep learning is to look into customized laptop that allow you to pick components and not waste money on getting a default windows installation unless you need it. Also pick a laptop with additional slot for upgrade.

Hope this helps in anyone trying to buy a laptop for deep learning.

2 Likes

Hi fellas,
i noticed this is quite the old thread but my problem is quite similar.
I am currently a comp. sci masters student from europe and I am going to china (Chengdu) for an internship and my masters thesis. This might be up to a year. I started getting into ML in my free time, since my university does not offer specialized degrees in this field. So far I did Andrews DL course, fast.ai, a few university projects on self driving and weather prediction and I am currently working with the carla simulator on intelligent agents. At home i have my TR / 32gb ram / 1080ti workstation, but i definitely want to be able to continue learning/training and then starting to focus on kaggle challenges while I am abroad. Now my budget for hardware is somewhere around 2000±300 Euros.

This leaves me with 2 options: (ultra)book + eGPU or a gaming Laptop:

(Iam excluding cloud solutions here because Iam not sure how this is gonna work out in china)

For the first option Iam looking at a Laptop from Tuxedo Computers:

for around 1250 euros I am getting:
-i7 8565U
-32 GB DDR4 (2*16)
-full 4 PCIe lanes TB3
as eGPU iam currently looking at the Aorus Gaming Box with an RTX 2070 for around 750 euros.

As for a gaming Laptop in this price-range the specs are somewhere around this:

Intel Core i7-9 9750H
32GB 2666MHz DDR4
RTX 2070 8GB max-q

-What Iam mostly worried about is that those gaming Laptop wont cope with training DL models for hours and hours, whereas in the gaming box the graphics card has its own enclosure, therefore better cooling

  • I feel like I pay a lot of money for useless tech on gaming laptops like 144+ Hz, G-sync, useless RGB. Gsync even makes one lose optimus, so worse battery life when in uni or on the go in general.

-Using eGPU gives somewhat more flexibility.

-performancewise I imagine both setups to be somewhat equal (when cpu is no bottleneck) because laptop 2070 (assuming non maxq) is somehwat slower than desktop, but eGPU is limited by TB3, has anyone experience with that?

-I have a 13" xiaomi air laptop atm which i can use for any uni related stuff where i need portability, or if i need to train models for longer and need a second device

Has anyone been in a similar situation or regularly training models on Laptops? Iam happy for any input and discussion :slight_smile:

EDIT: good thing, when i come back home i could plug the 2070 from egpu enclosure into myworkstation. the gaming laptop would not serve too much purpose then, only if i sell the thing completly!?

@larrykirschner Sorry for the late follow up. Has your experience improved? Did you purchase the one with the RTX 2080? I have quite some funds left and I was seriously considering ordering some tensorbooks until I read your post … Short notice would be highly appreciated. Cheers.