First and foremost, hashtag first world problem.
I’ve recently rented a gaming laptop. I am not a gamer and I usually don’t play 3D computer game. But I need a computer with a GPU and (decent) gaming laptops usually have one. I considered buying one but my own calculus suggests that renting one is cheaper for my research. Also, I don’t like the feeling of owning something because ownership is a responsibility, or even a liability, if I need to move to another country.
In Germany, there are several firms offering renting service of electronic equipment. I believe their customers usually rent things that can be easily “outdated”, let’s say a smart phone, a smart watch, or a pair of noise-canceling headphones.
I thought I can also take this as a trial of one of these renting services. I don’t want to own a computer with a GPU; I just want to use one. It is probably an ideal use case of these renting services and so I became a hipster-like customer. I also wanted to say that the customer service department of that firm is quite good. The officer in charge was very attentive to my questions. My questions were very technical and specific such as “Could I access the BIOS/UEFI and select the boot sequence from there? I need that to install Linux”.
After the CS department cleared up all of my concerns, I spent around 50 Euros per month and the firm sent me a brand new gaming laptop, which was unexpected as I expected a used one. It is a current generation, entry-grade gaming laptop. It has a GTX 3080Ti in it, not bad. I mean, the internal of the machine is superb. For my research purpose, it is more than enough.
However, I recommend researchers NOT renting (and of course NOT BUYING) a gaming laptop for conducting computational experiments. I am very glad that I got the laptop through renting, not buying. Actually, the firm does provide an option for their customers to own the rented device, if they have rented it for long enough. But surely, I will return the rented gaming laptop. I am pretty sure that I don’t want to own this crazy laptop.
As I said, the internal of the machine is superb. Why so negative?
Well, I always wonder why all gaming laptops must have that similar design aesthetics. In Asia, the aesthetics would be called Chūnibyō (中二病), roughly translated to English as “eighth-grader syndrome”. To summarize, the industrial design is usually like a Batmobile (Clooney’s, not Keaton’s) collided with Michael Bay’s Megatron to form a train wreak, but still bumped with some neon-colors-emitting motorcycles from the original Disney’s TRON. The design gives the gamers an illusion that they are invincible, strong, powerful, or anything in-between. I wondered who on earth would think this kind of design is practical. After using one myself, I came to a conclusion: It is entirely impractical to a point that the computer hardware is challenging to use.
One design that bugs me a lot is the hand rest. Hand rest, as the name implies, should allow one to rest hands or the upper part of the wrist on it. On a practical laptop, it is usually just a flat surface. Nothing fancy. There are larger ones (usually on “business” laptops, like my Lifebook U) or smaller ones (usually on those so-called Ultrabook, such as Macbook Air). But I think the spirit of the design is to make a hand rest comfortable to rest our hands on it.
On the gaming laptop, there is a relatively large hand rest because the computer is kind of large (and bulky). However, the designer(s) thought that the hand rest should be made of two types of plastic (a matte one and a shiny one). Therefore, I can always feel the difference in texture when I need to rest my hands on it. This is actually a minor issue. The bigger issue is next.
The designer(s) thought it would be cool not to make the laptop a boring rectangle. Instead, they literally cut corners. Instead of one corner which sane manufacturers usually round it, the designer(s) instead made this gaming laptop have SIX SHARP corners. It is more like Michael Bay’s Megatron: To have a lot of meaningless sharp edges to project a sense of fierceness.
But man, it’s super uncomfortable to type on that keyboard! Well, “super uncomfortable” is an understatement: It is painful! It creates an unnecessary occupational hazard.
I call this “Nitro wrist”.
Luckily, I have choices to mitigate this occupational hazard: I can use the “business” machine provided by my employer. I can also plug in an external keyboard. And ultimately, I can also return the laptop… later. I still need to rent it for a few months. Apart from this occupational hazard, the design makes it quite embarrassing to use the laptop in public for a person in his forties.
And that summarizes my experience of using a gaming laptop. But it raises an interesting question.
Sure. The most obvious alternative is not to use a GPU, just use the CPU. CPU is not as powerful and the time difference between computing by a GPU and a CPU is from 10x to 200x, depending on tasks.
I need CUDA for my computational research. It is rather unfortunate that a NVIDIA graphics card is essential for that. And computer manufacturers usually put one in these gaming machines, because, well, those cards were designed for improving the graphics of video games. My use case is “off-label use”, applying the pharmaceutical analogy. There are other graphic cards, e.g. those from AMD (previously ATI) and the custom silicon by Apple. All of them claim to have greater computational power than NVIDIA’s. For computation using OpenCL, Core ML or whatever, yes. But none of them supports CUDA. Computations through ROCEm (an “open” standard, but actually AMD’s API) or Metal (Apple’s API) are catching up. But at the moment, both are just in their experimental stage and not as mature as CUDA (as of writing, it has been developed for 15 years and has 11 versions).
So, if I need a non-gaming machine with a NVIDIA card, I can either build a rig my own or use an external graphics card. I have thought about trying external graphics cards. But those enclosure (you need an enclosure to house the graphics cards and connect to the computer through ports such as Thunderbolt) is expensive and there is no way to rent one.
Some super high-end business laptops by generic brands also have a good NVIDIA graphics card, e.g. Dell XPS 15. But these machines usually have a >2500 price tag. Some are also available for rent too, but usually for >150 a month. For my little project, it doesn’t worth it. There are also brands specifically produce laptops for machine learning, e.g. Lambda’s Tensorbook. And these are in the range of >3500 and not available for rent.
Finally, cloud. And this is actually my original option, which I thought a gaming laptop can replace. There are cheap but unconfigurable ones such as Google Colab; and expensive but configurable ones such as Google Compute Engine. A company offered me a 1-USD-per-hour service, which is extremely cheap, configurable, and speedy. But even with this cheap service, I can’t have unlimited computational time like I have right now for around 50. However, if your analysis is only occasional, it is cheaper to use cloud-based GPUs. The free Google Colab is super attractive and I did several projects with it.
With the current chip crunch, I don’t think a GPU would get cheaper in the next few years. With all these alternatives, you can see that there are cheap, fast, and configurable alternatives. But one can usually take two at a time, not three.
And it also highlights one problem.
The hardware cost of doing computation up to a certain point is almost like free (in free beer). But it is not that free when you need to do some crazy shit such as current-gen machine learning. Coupling with the knowledge cost and human cost (e.g. annotation of data for training and validation), the hardware cost make computational methods a hobby of the rich. It’s just nobody wanna talk about it.
Once again, I can just hashtag first world problem. But it is also a third world problem. There is environmental cost of harvesting minerals in third world countries. The energy used to power my laptop in Germany is probably from a non-renewable source and is still financing the Russian war machine in Ukraine.