What is Hyper Threading Technology as Fast As Possible

Techquickie ·Techquickie ·2014-05-07 · 953 words · ~4 min read
Floatplane YouTube

Transcript

JSON SRT VTT 68
0:00 Hyperthreading technology from Intel has been on their computer processors for
0:04 over 10 years. And yet, much like the
0:07 elusive fox, most people know very little about what it says. Let's start
0:12 with an analogy. Let's say I am a CPU and I'm trying to process food or eat as
0:19 some people call it. I can only do as much eating as one mouth can do. If I
0:23 could add more mouths, then I could process more. This is what multi-core
0:28 processors do. But due to cost constraints among other things, it's not
0:32 always possible to take that approach. So, I've got just one mouth. I can use
0:37 my hand to pick up the food, bring it to my mouth, then grab another bite while
0:41 my mouth is busy. If I finish chewing before my hand is ready to deliver me
0:46 more food, however, then my mouth is just sitting there doing nothing. If
0:50 only I could use two hands to prepare food for my mouth. Then even though my
0:54 mouth can't actually work any faster, I wouldn't waste any time. Oh, wait. I can
1:00 do that. That's hyperthreading. One processor and intelligent scheduling to
1:05 make sure that it's always working. Pretty awesome, right? Well, usually
1:10 hyperthreading can't do much for single threaded workloads where you can only
1:14 work on one thing at a time. For example, if you wanted to eat a two
1:18 scoop ice cream cone, you can't work on the bottom scoop until the top scoop has
1:23 already been consumed. And heaven help you if you try to eat the cone first.
1:28 In this example, there is no benefit to adding more mouths or more hands. The
1:33 opposite of this would be eating from a bowl of candies. Grabbing and eating a
1:37 red candy does not in any way depend on finishing the blue candy that's already
1:41 in my mouth. In fact, if I had some friends to help me eat it, we could
1:46 consume the candy very quickly by adding more hands and more mouths. This is an
1:50 example of a multi-threaded workload, and hyperthreading can definitely help
1:55 with this. All right, so let's bring it away from the eating analogy for a
1:58 minute and bring it back to the real world. Computing tasks that benefit from
2:02 hyperthreading and multiple processing cores are video editing, 3D rendering,
2:07 and heavy multitasking on your PC. Video editing is a great example because one
2:12 frame of a video can be processed while the next one is queued up because the
2:16 video is already shot. The PC doesn't have to guess what's going to be in the
2:19 next frame. It already knows. Here's another multitasking example. One
2:24 processor can be handling your light tasks like Skype and music playback,
2:28 antivirus, and whatever else you have running in the background, while another
2:31 one can focus all its attention on running a power- hungry video game in
2:36 the foreground. So, with that out of the way, how do you actually choose a
2:40 processor for your PC? The first thing I hope you got out of this is that
2:44 whatever task manager might say, hyperthreading is not the same as
2:49 doubling your processing course. It's basically a clever trick for more
2:53 efficient scheduling of the work that is done by the processing cores that you
2:57 have. It also increases power consumption and heat output a little
3:02 bit, but the benefits usually outweigh this drawback. In the real world, you
3:07 can get huge double-digit performance improvements all the way to no
3:11 performance improvement at all and in very rare cases even very slightly
3:15 decreased performance depending on how the software is optimized. So, it's
3:19 important to do your research about the software you'll be running. If the work
3:23 you do isn't heavily multi-threaded, then you might do just as well to save
3:26 your money and buy a processor that doesn't have hyperthreading. Most games
3:30 right now, for example, can't take advantage of more than a couple of
3:33 threads. However, if the work you do is heavily multi-threaded, then remember
3:38 this general rule. Hyperthreading is better than no hyperthreading, but is
3:42 not nearly as good as adding more physical processors. In an ideal world,
3:47 lots of physical processors, each with hyperthreading, is where it's at.
3:51 Speaking of where it's at, audible.com gives you a free audiobook just for
3:55 setting up a new account. And if you use the audible.com/Linus
3:58 link below me, then it makes me look pretty good. They've got over 150,000
4:02 books with everything from fiction to self-help to business. I checked out The
4:06 Four Hour Work Week by Timothy Ferris, which is why I am in Hawaii right now.
4:10 That's right, suckers. I recorded this before I left and now I'm on vacation.
4:15 Actually, I'm I'm not on vacation. I'm in Hawaii working, but um you know, I'm
4:20 taking a few days to not work while I'm there. So, you know, yeah, YouTube's a
4:24 tough gig, but you can help by heading over to audible.com/Linus
4:28 and getting your free audio book today. This has been another episode of Fast as
4:32 Possible here on TechQuicki. Don't forget to subscribe for more videos like
4:35 this. And if you check out the other link in the video description, you can
4:38 suggest future topics for fast as possible. And that is exactly why we're
4:42 doing this one about hyperthreading. So many people requested it. So we do read
4:46 them and we do