What's the Best Platform for Transcoding Media? Find Out NOW!! Linus Tech Tips

Linus Tech Tips ·Linus Tech Tips ·2012-05-07 · 1,788 words · ~8 min read
Floatplane YouTube

Transcript

JSON SRT VTT 132
0:07 well today I have what I believe is going to be a fascinating episode for
0:11 you guys so let's start with video okay
0:15 so videos everybody likes to watch videos there's a few people these days
0:19 who own tablets there's a Motorola Zoom or smartphones there's an iPhone 4 and
0:23 want to watch video on them but not all these devices and not all the media
0:26 players are compatible with every kind of media that you that you download or
0:30 rip or whatever however else you get your media so you often have to
0:35 transcode or convert the media from one format to another so I'm going to be
0:39 using media converter 7 for my testing today this is actually a really cool
0:43 little piece of software it was $29.99 with a coupon code that I found very
0:47 easily on their website okay so I'm going to be using a 200 Meg it is a 3
0:54 minute 1080p file that was actually taken with oh here you guys can actually
0:58 see me this camera the sx1 is 1080p I'm
1:02 going to be outputting it to an iPad compatible format and I'm going to be
1:05 doing it on a number of different platforms a number of different ways so
1:10 I have three platforms I'm going to use and I'm going to be using kind of a best
1:13 of Intel 1155 so that's the z68 chipset
1:17 with a 2600k hyperthreaded processor that is the most expensive platform out
1:21 of what I'll be using today I have an AMD am3 platform I have a Crosshair 4
1:26 formula with a 6 core 1100t CPU bear in
1:30 mind of course that the motherboard doesn't really have any impact on the performance in this case it's a CPU
1:35 limited thing and finally I have AMD socket fm1 so this is their new Apu
1:40 socket so I've got a 3850 quad core Apu
1:44 here on an a75 m55 motherboard I'm going
1:47 to be using all the same Hardware across these different boards and platforms so
1:52 I'm going to be using 8 gigs of DDR3 memory I've got same power supply and
1:56 Intel 510 series SSD on my test platform
1:59 then on each platform I'm going to en uh transcode the video a different way so
2:05 I'm going to use a CPU then I'm going to use a GTX 580 then I'm going to use a
2:10 radon 6970 and finally I'm going to use the
2:14 integrated GPU and we're going to compare how all these Solutions perform
2:18 against each other whether we're using CPU Cuda direct compute or uh in Intel
2:24 1155 case virtue or the Apu on AMD fm1
2:28 let's see how it goes okay so my first two runs are done with
2:32 the CPU as well as with the GTX 580 using Cuda on the 1155 platform and the
2:37 one that I'm about to run and the reason I'm stopping to tell you guys about it
2:41 is with Lucid virtue so all I have to do
2:44 is turn virtue on there we go the
2:47 processor Graphics are now enabled although I am plugged into the dedicated
2:52 graphics card so the igp is working now
2:55 and I'm going to run it with this one and see how it goes so I'll share
2:59 results with you as I go here and as I discover what works better and what
3:04 doesn't so here's a bit of an update I have the AMD fm1 system running right
3:09 now so I've got most of my results done for that and I'm actually running it
3:13 with the 6970 right now so you can see that the GTX 580 due to its very high
3:20 performance actually outperforms the CPU by itself which we expect I mean the CPU
3:25 was the slowest out of all of our results even on the 1155 platform where
3:30 it's a significantly faster CPU with hyperthreading um the igp did beat the
3:36 CPU but uh not maybe not by as much as I
3:39 would like to see and uh actually here's something interesting I also added on
3:44 the 1155 platform I tested a GTX 550 to
3:47 see how it would compare to the 580 so it's almost as slow as a CPU so and
3:52 definitely slower than the igp so if you're using 1155 and a lower end
3:57 graphics card well anything but a super high-end graphics card you might
4:00 actually be better off with uh with
4:04 virtue or using the CPU versus using
4:07 Cuda so uh here we go we've got our result for the
4:11 6970 so here it is it is
4:15 133 so you can see the 6970 once again
4:18 doesn't perform quite as well as the 580 but it's close and then the igp trails
4:24 significantly behind those two due to it well being substantially slower so I'm
4:29 pretty much much finished my data at this point only to discover that it was
4:34 all pretty much for nothing so what I accidentally did was I used a video file
4:41 this one which had already been converted from its uh large High bit
4:46 rate self into a Windows media audio
4:49 file so what I'm going to do Second Time Around is I'll be using this one this is
4:54 a 1.64 gig file whereas this is a 200 Meg file by
4:58 aent okay so this is a OV file that is
5:02 what is natively recorded by my camera
5:05 then I'm going to convert this into an iPad format so you can see the data I've
5:10 got here with the other file so the uh
5:14 Intel Quicks sync is a good performer uh Cuda also performs quite well open CL is
5:20 a little bit behind Cuda and then CPU is
5:23 uh inv variably uh the slowest solution
5:26 however I just ran quicksync on the new
5:30 file which is the Intel Core i7 extreme
5:33 Edition 990x or maybe it's a 980x unboxing and I it was done in 56 seconds
5:40 I am now running it on opencl and you can see the difference in performance
5:44 between quicksync and opencl on that already transcoded once file this ran in
5:49 yeah 56 seconds and it's already taken well over 2 minutes with open CL when
5:54 we're doing a far more complicated file so I'm going to just rerun everything
5:59 which is good going to take me look at the time it is now already which is
6:02 going to take me another you know hour or hour and a half or whatever and then
6:06 I will be back with proper results for you guys okay well I'm finally done took
6:10 me another couple hours to uh finish up here but I have the results so I've just
6:16 uh used my old chart but all of the
6:19 bottom results are the ones using the
6:22 larger file and converting it to an iPad compatible format with media converter 7
6:28 so let's have a look at the chart here and let's go we'll go through it together so this is the
6:33 2600k with the CPU alone it clocks in at
6:36 1 minute 40 seconds which is better than the 1100t better than the 3850 and
6:41 better than the I 32105 so that makes a lot of sense cuz
6:47 this is more expensive than this which is more expensive than this which is
6:50 almost the same as this so this one you can see due to its four cores this is
6:54 two cores with hyperthreading this is six cores four hyperthreaded cores so
6:58 this guy with this four cores performs better than the equivalently priced
7:02 Intel CPU with only two cores all right
7:05 now let's factor in Cuda so we've got a GTX 580 now so they actually all perform
7:11 fairly similarly this was a bit of a glitch but I ran it again and I don't
7:15 know why but the 1100t did not perform quite as well with my Cuda GTX 580 as
7:21 the uh the other cards did with the 6950
7:25 we seem to see a lot more benefit from additional CPU power so you can see that
7:29 the 6950 is still an improvement over
7:33 the CPU but it doesn't like Cuda was very
7:38 similar for with a GTX 580 across the
7:41 board whereas the 6970 it seems like if you have a more powerful CPU to feed it
7:46 it is really able to ramp up the performance so with the I
7:50 32105 it clocked in at 246 with the 3850
7:54 226 209 with the 1100t and a whopping 1
7:58 minute 36 seconds seconds with the
8:01 2600k now igp performance so the
8:04 1155 the 2600k and the 2105 were both
8:09 within spitting distance of each other and in spite of this being a quad core
8:15 CPU that is clocked higher this being a dual core CPU so you can see that the
8:19 igp is doing all the work here it's not getting a whole lot of help from the CPU
8:23 side like we saw with the 6950 as for the fm1 solution remember
8:28 unlike the Intel igp with using Lucid
8:33 virtue on the fm1 solution you can only use it if you are running Onboard video
8:38 this can be used if you're using a z68 or h67 chipset here's my z68 board this
8:45 can be used even if you are running a discret graphics card so that's kind of
8:49 a cool feature as well so here you can see that the igp did not perform as well
8:54 as the 6970 but was still better than the CPU on its own so hopefully this was
8:59 helpful thanks for checking out my video on the uh different platforms and their
9:04 video transcoding performance Baron ma here's another quick thing to bear in
9:08 mind you guys so you can see here that the with the h67 or z68 chipset the 1155
9:14 chips whether it's an i7 or an I3 just smoke everything else but only a few
9:20 software applications actually support all of these different Hardware
9:24 accelerations so if you're using anything that doesn't in terms of share
9:28 CPU perform performance you can see that the fm1 3850 due to its additional cores
9:34 blows away the equivalently priced uh
9:37 product from Intel and then you have to spend quite a bit more to get better
9:40 performance on the Intel side don't forget to subscribe guys