3 ms·

Language performance benchmarks are generally pointless. The much more important benchmark is how easy is the code to write / change. Write it quick, get the cu

by fleitz 16y ago

Language performance benchmarks are generally pointless. The much more important benchmark is how easy is the code to write / change. Write it quick, get the customer, and then when you have that customers money if somehow it doesn't perform well enough use that money to profile and improve the app or just buy more hardware. In 18 months a CPU that costs $200 will perform twice as fast as the current CPU for $200, I've never seen bang for the buck like doubling the performance of your code for $200. (Most apps are IO bound anyway, so you want to add disks not CPU) Yes, for SOME companies like google it's much better to pay more for programmers, but most companies don't have > 100,000 servers. If your company is a start up that is doubling every year then all you have to do is let Moore's law scale your performance as you add servers. Also, don't build a business where your profit margin is dependent on getting the last 5% performance out of a server.

barrkel 16y ago
It very much depends on your application domain. For code that runs on client devices, speed is often very important. You can't (usually) buy hardware for your customers.
Legion 16y ago
I think "generally pointless" is overstating things a bit. That said, I think it is definitely fair to say that these sort of benchmarks fall more into the category of "interesting" than "useful". Usually fun to read, but just not something to confuse with information to base business decisions on.
runT1ME 16y ago
>In 18 months a CPU that costs $200 will perform twice as fast as the current CPU for $200, I've never seen bang for the buck like doubling the performance of your code for $200. (Most apps are IO bound anyway, so you want to add disks not CPU) 2001 called, they want their answer back. Ok, that's a little harsh, but really, you're answer was perfect for about six years ago, but if you've taken a look at chip performance recently, it has certainly not been doubling every year. What you can get for cheaper is more cores, but that certainly isn't a free lunch, as many have pointed out. Suddenly having a good language that can be per formant and scale matters (python anyone?). Go is going to have a lot of catching up to do, as writing good multi threaded libraries are hard, and a good parallel GC is even harder. I don't mean to pick on your opinion, because its one shared by many, but things are going to get very interesting in the next 10 years as we really will see the end of moore's law in terms of linear CPU performance and instead get many-core processors to play with.
- fleitz 16y ago
  If you write parallel code you'll get a lot closer to free lunch. To me the transition is almost complete, you're starting to see that even the cheapest laptops these days have two cores.
  - runT1ME 16y ago
    >If you write parallel code you'll get a lot closer to free lunch. That's like saying "if you spend lots of money on good food, you can get a pretty good free lunch". Good parallel code is hard, to varying degrees. I'm in agreement, its going to be the future, but development is going to get a lot harder in the next few years.

by fleitz 16y ago

barrkel 16y ago

It very much depends on your application domain. For code that runs on client devices, speed is often very important. You can't (usually) buy hardware for your customers.

Legion 16y ago

I think "generally pointless" is overstating things a bit. That said, I think it is definitely fair to say that these sort of benchmarks fall more into the category of "interesting" than "useful". Usually fun to read, but just not something to confuse with information to base business decisions on.

runT1ME 16y ago

>In 18 months a CPU that costs $200 will perform twice as fast as the current CPU for $200, I've never seen bang for the buck like doubling the performance of your code for $200. (Most apps are IO bound anyway, so you want to add disks not CPU) 2001 called, they want their answer back. Ok, that's a little harsh, but really, you're answer was perfect for about six years ago, but if you've taken a look at chip performance recently, it has certainly not been doubling every year. What you can get for cheaper is more cores, but that certainly isn't a free lunch, as many have pointed out. Suddenly having a good language that can be per formant and scale matters (python anyone?). Go is going to have a lot of catching up to do, as writing good multi threaded libraries are hard, and a good parallel GC is even harder. I don't mean to pick on your opinion, because its one shared by many, but things are going to get very interesting in the next 10 years as we really will see the end of moore's law in terms of linear CPU performance and instead get many-core processors to play with.

fleitz 16y ago
If you write parallel code you'll get a lot closer to free lunch. To me the transition is almost complete, you're starting to see that even the cheapest laptops these days have two cores.
- runT1ME 16y ago
  >If you write parallel code you'll get a lot closer to free lunch. That's like saying "if you spend lots of money on good food, you can get a pretty good free lunch". Good parallel code is hard, to varying degrees. I'm in agreement, its going to be the future, but development is going to get a lot harder in the next few years.