Predicted speed increase

Post your testing results with HandBrake.
Post Reply
colineddy
Posts: 3
Joined: Thu Sep 15, 2011 1:19 pm

Predicted speed increase

Post by colineddy » Thu Sep 15, 2011 1:25 pm

Hi, I have AMD athln x2 5200 right now on win7 64bit with 4 gig of ddr2 800 (400mhz) ram. Satan 1 drives.
I get 30fps encoding DVD mpeg2 to iPad h264 preset.

I would like to go much faster but want to spend m. Money wisely.

I was planning phenom II X6 1055 system also with 4 gig ram but of course DDR3 at 1066 speed.

Any opinions on what fps I might realise wit this new kind of rig ? 10 extra won't be worth the hassle.....

Cheers,

Colin

User avatar
Rodeo
HandBrake Team
Posts: 12195
Joined: Tue Mar 03, 2009 8:55 pm

Re: Predicted speed increase

Post by Rodeo » Thu Sep 15, 2011 3:13 pm

You're tripling the number of processor cores, and x264 scales quite well up to 6-8 cores.
At some point there will be bottlenecks, but I think you will probably double your encoding speed, perhaps more.
The difference may be less for HD sources like Blu-ray/HD-DVD.

Edit: if encoding video with x264 is your primary goal, the best processors would be:

- Intel 6-core (i7)

- Intel 4-core (i7, i5), AMD 6-core

- AMD 4-core

- the rest

An Intel 4-core i7 (or a high-end 4-core i5) might be just as fast or faster with x264 than some AMD 6-cores.
Prices may be similar, not counting other components (motherboard etc.).

colineddy
Posts: 3
Joined: Thu Sep 15, 2011 1:19 pm

Re: Predicted speed increase

Post by colineddy » Thu Sep 15, 2011 3:50 pm

Ok thanks for that... Will go for the AMD - intel always seems expensive to me..

firewing1
Posts: 21
Joined: Sat May 23, 2009 8:18 pm

Re: Predicted speed increase

Post by firewing1 » Fri Sep 16, 2011 8:37 pm

For what it's worth, I would recommend an i5-2500k if you're looking for a cost-effective transcoding CPU. I recently purchased one and it over clocked very easily to 4.0GHz on air cooling (used a cheap $20 Cooler Master Hyper 212+ aftermarket cooler) and it rips through transcodes. I'll post more detailed benchmarks when I get back home, but I get about 85-100FPS for 480p (i.e. DVD) media and 45-60FPS for 1080p (i.e. Blu-Ray) media on the High Profile preset. The quality is amazing for the output video size.

Edit: Anandtech's CPU comparison page may be of help when choosing your new processor: http://www.anandtech.com/bench/Product/203?vs=288
The i5 performs nearly identically to the latest AMD 6-core processor in transcoding-type workloads, but it beats it by a large margin in other places.

Update 2011/09/17: I tested with a few sources, 480p transcodes on High Profile are indeed usually between 85 and 100FPS, often even higher than that (around 110 or 125FPS was common). I forgot to mention in my original post above that for 1080p, I was downscaling the file to 720p. Compression of a 1080p without downscaling will be slower of course... In my quick test I got between 30-60FPS when transcoding 1080p->720p depending on the scene.

kosobai
Posts: 15
Joined: Sun Apr 26, 2009 12:08 pm

Re: Predicted speed increase

Post by kosobai » Fri Oct 07, 2011 4:08 am

Take these benchmarks with a pinch of salt; it's well known that Intel's compiler cripples AMD CPUs on Windows. If you're running Linux, you'll get pure, unadulterated speed on any 1xxxt X6 cpu.

While we're on the subject of benchmarks, my 1065t system achieves a Geekbench score of 9812 (Ubuntu). That puts it ahead of the i7 2600, if the Geekbench PC benchmarks page is to be believed, and for 100 bucks less, too. Stick that in your pipe and smoke it, Intel! LOL

Maybe it would get over 10000 if I were running the 32bit version of Ubuntu, and not the 32bit libraries on 64bit Ubuntu. But that's just conjecture. ;)

tlindgren
Bright Spark User
Posts: 237
Joined: Sun May 03, 2009 2:14 pm

Re: Predicted speed increase

Post by tlindgren » Fri Oct 07, 2011 12:15 pm

kosobai wrote:Take these benchmarks with a pinch of salt; it's well known that Intel's compiler cripples AMD CPUs on Windows. If you're running Linux, you'll get pure, unadulterated speed on any 1xxxt X6 cpu.
It only affects a small number of programs that are compiled on (IIRC older?) Intel compilers, though some of them can be important for specific peoples, and this happens on both Windows and Linux (Intel compilers for both, and I expect mostly the same programs to be affected).

Handbrake is compiled using gcc on all platforms (not Intel) and the really performance critical parts is mostly hand-coded assembler. So performance difference for Handbrake between OS is basically a test of how good/bad the OS scheduler is for this special case, for a long while Linux got trounced by Windows but it's been fixed.
kosobai wrote:While we're on the subject of benchmarks, my 1065t system achieves a Geekbench score of 9812 (Ubuntu). That puts it ahead of the i7 2600, if the Geekbench PC benchmarks page is to be believed, and for 100 bucks less, too. Stick that in your pipe and smoke it, Intel! LOL
1065T doesn't seem to have an official Geekbench score, but the faster 1090T Extreme officially has a lower score than that! It's rated at 8074, while the i7-2600 clocks in at 9500, presumably under similar circumstances. But the benchmark browser also shows that the Geekbench numbers are highly unreliable, with very different scores for the same CPU and frequency! Seriously, with that kind of variability it's not worth anything as "benchmark", if anything it's actively misleading.

But several people have noted earlier that AMD's 6-core desktop chips are generally good value for video encoding, similar performance to Intel's 4-core models at somewhat lower "system price" (CPU and motherboard). However, since very few things can use that many cores the same Intel 4-core models generally stomps the X6 models at almost everything else, even at the same cost level, before video decoding was multi-threaded this sometimes happened even when doing video encoding!

Basically, if you want a machine that's not optimized only for video encoding AMD has good alternatives up to the mid-range X4's when looking at the whole system price, then it's Intel only. AMD's Bulldozer derived chips might still change that but the tests on the server versions looks much less promising than people had hoped, but they may well be able to fix it (early Phenom's also had issues that they had to fix).

kosobai
Posts: 15
Joined: Sun Apr 26, 2009 12:08 pm

Re: Predicted speed increase

Post by kosobai » Fri Oct 07, 2011 1:43 pm

It only affects a small number of programs that are compiled on (IIRC older?) Intel compilers,
Apparently, it's still an issue,even now:

http://www.amdzone.com/phpbb3/viewtopic ... l+compiler
1065T doesn't seem to have an official Geekbench score, but the faster 1090T Extreme officially has a lower score than that!
Here you go:

http://browse.geekbench.ca/geekbench2/view/492013

9653. My system, uploaded as of 10/7. If the 'faster' 1090t is actually slower, well it means I've purchased well!
But the benchmark browser also shows that the Geekbench numbers are highly unreliable, with very different scores for the same CPU and frequency! Seriously, with that kind of variability it's not worth anything as "benchmark", if anything it's actively misleading.
Agreed, but you can look at a handful of similar systems and disregard the flaky results for a 'ball park' figure.
However, since very few things can use that many cores the same Intel 4-core models generally stomps the X6 models at almost everything else, even at the same cost level, before video decoding was multi-threaded this sometimes happened even when doing video encoding!
There's just no pleasing the Intel crowd. :) There's no doubt that Intel has the faster chips as of writing, but AMD's offerings aren't as weak as the Intel FUD & BS machine would like us to believe.

Bling2Ming
Regular User
Posts: 104
Joined: Mon Dec 29, 2008 10:03 pm

Re: Predicted speed increase

Post by Bling2Ming » Wed Oct 12, 2011 4:42 pm


Post Reply