# Best laptop (fast) under 300$-400$

1. Sep 5, 2017

### Gjmdp

I'm about to start a simple project on machine learning and I need a laptop. I've been looking for this on several pages, but I don't really understand much of if. I think for 300$-400$ an intel i5 would fit pretty well or something like that. I don't really care about the graphics or all that sort of things, I just want it to be fast. Thank you.

2. Sep 5, 2017

### Borg

What users perceive as speed is due to processing speed and RAM. If you go with an i5, be sure to have plenty of RAM (8 - 16 GB).

3. Sep 5, 2017

### Gjmdp

Acer Swift 1, Acer Aspire E 15, Asus VivoBook W202 seem to me good computers, considering its positive reviews and relative cheap price. But they have 4GB of RAM, so I don't know if that is quite bad. Anyway, which one is the best?

Last edited: Sep 5, 2017
4. Sep 5, 2017

### Borg

An i5 with 4 GB of RAM is not going to be very fast. Because of the small amount of RAM, the computer will constantly be reloading from disk every time you look at a different app. For example, lets say that you have Excel open while you're running your machine language program. If you look at the output of the program and then switch to your already opened Excel, it will probably have to reload Excel from the hard drive which will be very slow as compared with loading it from RAM. And, depending on how much processing is going on with the program, it will take longer to do anything in either the program or Excel. You want to have an absolute minimum of 8GB RAM for any new computer. Anything less than that is little more than a brick.

5. Sep 5, 2017

### Greg Bernhardt

Could you do with a Chromebook? Regardless of price I've always been happy with Asus products.

6. Sep 5, 2017

### Gjmdp

But is that fast enough for a Machine Learning software? I was not able to find any Chromebook with speed >2.1GHz

7. Sep 5, 2017

### Gjmdp

There is "ASUS P-Series P2540UA-AB51 business standard Laptop, 7th Gen Intel Core i5, 2.5GHz (3M Cache, up to 3.1GHz), FHD Display, 8GB RAM, 1TB HDD, Windows 10 Home, Fingerprint, TPM, 9hrs battery life" , with i5 and 8GB. Is there a catch or something bad with this laptop?

Last edited: Sep 5, 2017
8. Sep 5, 2017

### fluidistic

I'm quite surprised there's no mention of "SSD" so far. If you care about speed, you cannot go without a SSD.

9. Sep 5, 2017

### newjerseyrunner

Machine learning should not require a massive amount of RAM. A neuron shouldn't take more than a few bytes of information, and you're not going to have the processing power to handle millions of them so you'll run into a CPU bottleneck before you hit the RAM one.

Make sure that you thread your application right. I'm always asked by people how to make programs faster, then I go into the source code and find a single thread or dynamically created threads.

You may want to consider the graphics card. There is a lot more parallel processing power available to you on the graphics card if you know how to tap into it. High parallelization is what you want for the fairly simple maths of machine learning.

10. Sep 5, 2017

### newjerseyrunner

SSD will only help if loading time is a problem or the OS is doing page swapping. During processing, the disk shouldn't be doing anything.

11. Sep 5, 2017

### Gjmdp

So, is an Acer Aspire E15 enough? (Intel i3, 4GB RAM), or should I go to the fancy "ASUS P-Series P2540UA-AB51 business standard Laptop, 7th Gen Intel Core i5, 2.5GHz (3M Cache, up to 3.1GHz), FHD Display, 8GB RAM, 1TB HDD, Windows 10 Home, Fingerprint, TPM, 9hrs battery life" (Intel I5, 8GB RAM, and no idea of SSD)?

12. Sep 5, 2017

### NFuller

Personally, I would go with the ASUS machine over the Acer. Every Acer I have ever used was total garbage but I have had no problems with ASUS machines.

13. Sep 5, 2017

### fluidistic

Correct me if I'm wrong but opening a program (or even a folder with many files) involves reading from disk and that's the bottleneck in speed, not the cpu speed, at least generally.
This is why Firefox, Libreoffice or any other "big" programs tend to open extremely quickly on my desktop pc (with an ssd) compared to work and university where I only have a HDD. All CPUs are either core i7 or intel xeon above 3 GHz.
From what I had been told, I can expect a speed increase even on an Intel Celeron N2830 where both the CPU and HD are bottlenecks. On a core i3 or better, I don't think the CPU is the bottleneck in most cases.

Another good point of SSDs is that they use less power than mechanical HDDs. It might be worth considering on a laptop.

And about during processing that the disk shouldn't be doing anything, I don't know if this holds true for machine learning but that doesn't necessarily hold true when solving PDEs with finite elements and saving the solution to disk. If you follow the FEniCS tutorial, you will write like over 60 GB of data for the Navier Stokes example, at the same time the CPU is doing the calculations.

14. Sep 5, 2017

### Gjmdp

I agree, Acer is worst than ASUS. The thing is that there is a difference of 200\$ in price (Acer being cheaper). So despite that, can the Acer support a Machine Learning code/ Software?

15. Sep 5, 2017

### NFuller

Yes, but it will likely be slower. I think the issue with the Acer machines is the PCH in these machines, which controls the information throughput between the cpu, ram, and peripherals, is very cheap. Acers also tend to not last very long before the hardware starts falling apart, i.e. keys popping off, power jack breaking, screen breaking, etc.

Yes, there is a price difference, but you get what you pay for!

16. Sep 5, 2017

### phinds

After all that has been said, you don't seem to be listening. 4GB of ram will make for a computer that is slower than mud. Personally I wouldn't touch anything less than 16GB but for the average user 8GB MIGHT be enough (if you're not running more than a program or two at any one time).

17. Sep 5, 2017

### Gjmdp

I think it is premature and unnecesary to say I was not "listening" at all. Correct me if I'm wrong, but I was told here not to have a 4GB RAM provided I would have i5 too (whereas I clearly say that the Acer has an Intel i3, not an i5). In the case of i3, I thought the necessary RAM may be more flexible (but it's not to be the case).

Anyway, thank you NFuller, I will take the ASUS. And thank you all for your responses :)

Last edited: Sep 5, 2017
18. Sep 5, 2017

### Willelm

I agree, that was unnecesary and somewhat rude.

19. Sep 6, 2017

### newjerseyrunner

Opening a program yes, but once it's opened, it's no longer reading from the disk. For long-running data crunches, we don't really care how long it takes to load. I don't care that one of my Unity scenes takes ten minutes to load if I'm going to spend the next eight hours baking it's lights.

20. Sep 6, 2017

### Staff: Mentor

@newjerseyrunner - some of what you describe with graphics programs like Unity is algorithmic slowness, iterating over large numbers of data points. Parallelism available in graphics cards - lots of gpgpu's - offloads (or potentially should) most of that kind of load to an ancillary set of cpus that run floating point well. FP can be another performance bugaboo. A lot of smart people have worked for a long time on both of these issues.

Example of one the leading software engineers from Silicon Graphics working on the issue of square roots. Fun read.

http://www.azillionmonkeys.com/qed/sqroot.html

21. Sep 6, 2017

### newjerseyrunner

Yes, I'm aware it's algorithm-based. I was just saying that for long running processes like the one the original poster explained, loading time is meaningless. I use the graphics card whenever I can. That square root article was interesting, I notice he's actually using the "Carmack Inverse Square Root," which is famous in graphics circles because of it's use in Quake.