z170 or x99 better for h264 encoding?

[JimmyBarnes]

My current rig is: Q9550 Intel Core 2 Quad 2.83 GHz, 4 GB RAM on Windows XP 32-bit.

While it is still quite serviceable, it is 10 yo and I figure it is time to move on.

Anyone have any suggestions on what hardware to get for h264 encoding from blu-ray?

Obviously the faster the encoding rate the better but with a budget around $USD1000...

TIA

[Groucho2004]

If you're planning on keeping WinXP, your options are very limited. Drivers for mainboards that fit newer Intel CPUs (i5/i7 6xxx/7xxx) are basically non-existent.
Some AMD boards (AMD970 chipset for example) for 8 core CPUs (fx8xxx) may have drivers for XP.

As for encoding speed - Intel CPUs are faster and much more power efficient than AMD CPUs.

So, if you stick with XP your options for faster encoding would be:
AMD FX8xxx CPU on AMD970
- or, much better -
Intel Sandy/Ivy Bridge (i7 3xxx) on Intel P67, etc.

The latter CPU might be difficult to get these days.

Best option is to upgrade your OS and use Intel i7 6xxx/7xxx.

[JimmyBarnes]

If you're planning on keeping WinXP, your options are very limited.

Thanks for your reply. I should have mentioned I will be upgrading to Windows 10 Pro 64-bit.

[mariush]

AMD Ryzen processors will show up in stores towards the end of February, maybe sooner.
The 8 core / 16 thread Ryzen processors are about the same performance or faster than 1000$ processors from Intel. May be worth waiting for them to see the retail price and reviews with actual retail processors.

You'd need an AM4 socket motherboard but unlike Intel, the AM4 motherboard should last a longer time.. they're not changing sockets as often as Intel or forcing you to upgrade to another motherboard because the newest processor on same socket only works with some newer chipset, or crap like that.

[Groucho2004]

AMD Ryzen processors will show up in stores towards the end of February, maybe sooner.
The 8 core / 16 thread Ryzen processors are about the same performance or faster than 1000$ processors from Intel. May be worth waiting for them to see the retail price and reviews with actual retail processors.
The benchmarks I've seen so far are very vague, I usually pay no attention to these "leaked" numbers.
However, all the hype and chatter seems to indicate that AMD have finally produced something capable to break the Intel dominance in the desktop market. Let's wait and see...

[JimmyBarnes]

AMD Ryzen processors will show up in stores towards the end of February, maybe sooner.
The 8 core / 16 thread Ryzen processors are about the same performance or faster than 1000$ processors from Intel. May be worth waiting for them to see the retail price and reviews with actual retail processors.

Providing encoding software/codec can use multi-cores, for a given clock speed will more cores ALWAYS encode faster than less cores?

[mariush]

It depends.

You can only use so many threads until the quality of the image actually starts to decrease, or the threads are not used efficiently.

From memory, I think h264 works best (without losing quality) with one thread for at least 40 pixels vertically, or something like that. So for example for 720p content, it probably wouldn't help (or the image quality would suffer in a noticeable way) to use more than around 16 threads.
Here's some very old message from one of the x264 developers that explains why lots of threads may not help : https://forum.doom9.org/showthread.php?p=1646307#post1646307

x264 would automatically use up to 1.5 x number of cores so with a 8 core / 16 threads cpu like Ryzen it may use up to 24-30 threads

Another thing... Sometimes the actual encoding speed is restricted by the decoder speed which may be single threaded or has poor multithreading, or there's some filter in the processing chain which doesn't like lots of cores.
Most processors these days also have turbo boost modes, basically they have a stock frequency and a turbo frequency which is only active when just a few cores are "loaded". Other processors boost the frequency of individual cores to a higher degree, depending on how many are used..
Here's an example : http://hexus.net/media/uploaded/2016/3/1c347159-fbda-499b-925e-3b99a2969e9f.PNG <- you can see there that for example, the 22 core 2699 v4 cpu has base frequency of 1.8 ghz but a single core will boost to 3.6 ghz (1.8 + 1800 mhz) if all the other cores are not active.

So a cpu with less cores/threads may actually perform better than a cpu with more cores/threads but running at lower frequencies.

[JimmyBarnes]

It depends.

Doh!

I had in mind to get an Intel Core i7-6700K (4-core, 4.0 GHz, 8 MB, 14 nm, 91 W. DDR3L 1600/DDR4 2133. Max.RAM 32 GB.) Thoughts on this?

Which do "serious encoders" tend to use, AMD or Intel?

Where does one go to compare CPUs for x264 encoding performance, Doom9 doesn't seem to have much.

[GMJCZP]

Only for curiosity:

Here (http://www.win-raid.com/t22f23-Guide-Integration-of-Intels-AHCI-RAID-drivers-into-a-Windows-XP-W-k-W-k-CD.html)

A benchmark review, read the h264 section:

Here (http://www.cpu-world.com/Compare/492/Intel_Core_i3_i3-6100_vs_Intel_Core_i7_i7-6700K.html)

PD: It's interesting the new AMD processors.

[Groucho2004]

Where does one go to compare CPUs for x264 encoding performance, Doom9 doesn't seem to have much.
Here's (http://www.legitreviews.com/intel-core-i7-7700k-processor-review_189491/6) a comparison using the Techarp x264 HD benchmark (scroll down to "X264 HD Encoding"). A bit dated but still OK.

[JimmyBarnes]

Here's (http://www.legitreviews.com/intel-core-i7-7700k-processor-review_189491/6) a comparison using the Techarp x264 HD benchmark (scroll down to "X264 HD Encoding").

Cool. According to that clearly i7 6950X is king but at a price some 5x higher than i7 6700K it would be hard to justify unless one were being paid for encodes (I wish!)..

i7 6700K isn't too shabby tho.

I almost always use CRF single pass for encoding (unless I am trying to fit a series on a disc, when I use 2-pass), presumably this would equate reasonably well with 1st pass results in the x264 table?

A bit dated but still OK.

? Posted Jan3, 2017..

[Groucho2004]

I almost always use CRF single pass for encoding (unless I am trying to fit a series on a disc, when I use 2-pass), presumably this would equate reasonably well with 1st pass results in the x264 table?No, second pass.

[JimmyBarnes]

No, second pass.

Glad I asked!

This might be a toughie: Is there any way to compare x264 encoding performance on i7 6700K with that of my current CPU Core 2 Quad Q9550 2.83 GHz? I would be a tad disappointed to find after doing the upgrade that I was getting say, less than 25 % faster encoding.

Does GPU have any relevance to x264 encoding speed?

[Groucho2004]

This might be a toughie: Is there any way to compare x264 encoding performance on i7 6700K with that of my current CPU Core 2 Quad Q9550 2.83 GHz? I would be a tad disappointed to find after doing the upgrade that I was getting say, less than 25 % faster encoding.It'll be difficult to find a direct comparison for this particular application but I'd say it'll be at least twice as fast, probably more.

Does GPU have any relevance to x264 encoding speed?x264 can use OpenCL for some calculations but the speed-up is barely noticable.
However, since the encoding chain includes a decoder and possibly some Avisynth filters, off-loading these tasks to the GPU will improve encoding speed. DGDecNV for decoding and KNLMeansCL for de-noising are two good examples.

[JimmyBarnes]

It'll be difficult to find a direct comparison for this particular application but I'd say it'll be at least twice as fast, probably more.

I found a way:

https://www.mediafire.com/convkey/c177/akzsccq5xzgkayt6g.jpg

using your table and http://www.anandtech.com/bench/CPU/54 and http://www.anandtech.com/bench/CPU/28, x264 HD Benchmark versions 3.03 and 0.59.819.

Latter lists my ancient Core 2 Quad Q9550 and can be used with the table for v3.03 to indirectly link to i7-6700K. I compared the 3 tables for interest.

In the x264 CPU Encoding Comparison, q9550 encoding rate is set to 1.0, then results for other CPUs scaled to this (in the "x" column" (the formula bar fx contains an example). Figures in bold orange are reference "linking" points (nominated, not calculated) and used as a basepoint to scale other CPU results for that version.

The bottom line is, if this comparison between x264 HD Benchmark versions is valid, the "x" values between versions for a given CPU should agree. And while there is some "slop", they do, IMO remarkably well. Remember that values in "x" columns are how fast that CPU would encode relative to Q9550.

So the above indicates that i7-6770K should encode 3x as fast as Q9550, which agrees with your intuitive guess.

[hello_hello]

So the above indicates that i7-6770K should encode 3x as fast as Q9550, which agrees with your intuitive guess.

Only 3x ????

I'm about to build a new PC as my Q9450 is getting a bit long in the tooth. Well.... it still works perfectly, but I figured it was time for a speed increase and maybe a new GPU.... plus apparently XP is now the work of Satan so it's an excuse to install Win7, but only 3x ?????.

At least I know to lower my expectations.... I'd figured a 6770k would be so fast it'd finish encoding before it started.... at least relative to my 9yo Q9450. Nine years later and it's really only 3x faster...... ?? :(

[mariush]

I suspect it's much more than 3x.
You can't compare results of hd benchmark 3.0 and hd benchmark 4.0 and 5.0. They all use very different revisions of x264 and - i'm not sure but probably - I think only the 5.0 version uses a 1080p sample...yeah, 4.0 uses a 720p mpeg2 encoded sample for input.

That old benchmark I think from memory uses a short 720p sample to run a test encode at a specific bitrate and measure the fps. With higher resolutions and possibly different encoding settings (for example CRF instead of 2 pass vbr), you may see some speedups due to cpu instructions new processors have, and due to more cores or more parallel threads.

Also Windows XP ... that means you probably have the 32 bit version of Windows XP, which means you're using the 32bit of x264 and used less than 2-3 GB of memory - with some high quality presets and 1080p encodings, you'd go over this amount of memory, forcing you to use the 64 bit versions of x264. These may also be slightly faster than the 32 bit versions in such situations.

[JimmyBarnes]

I suspect it's much more than 3x.
You can't compare results of hd benchmark 3.0 and hd benchmark 4.0 and 5.0. They all use very different revisions of x264 and - i'm not sure but probably - I think only the 5.0 version uses a 1080p sample...yeah, 4.0 uses a 720p mpeg2 encoded sample for input.

To my mind the comparison I made (the table) is internally consistent, but we'll see..

I committed to i7-6770K / DDR-2133 16GB / Windows 10 and am in process of setting up the whole thing. Unfortunately TMPGEnc 4 Xpress which I was using doesn't work under 10 so I will probably get Video Mastering Works 6. It will take awhile but eventually I should be able to do the same encode on both PCs and compare the times taken.

[kuchikirukia]

I suspect it's much more than 3x.

It won't be.

http://www.anandtech.com/bench/CPU/1052

The Phenom II's and early APUs had about the IPC of a C2Q, so the Athlon X4 750K @ 7.89fps is probably pretty close to his C2Q. 6700k scores 27.82.

[hello_hello]

If I get off my backside (I've developed a loathing for upgrading anything over the years) I'll be able to test it and report back next week. I might even try to get the the local PC shop and buy the bits tomorrow.

Edit: Nah..... I did a bit of reading and might hang out for Zen first, to see if it's a better option, or at least better value for money. Hopefully it'll only be only a month away.

[JimmyBarnes]

I suspect it's much more than 3x.


Well I have now done a full BD encode (source files ~23 GB, 167,424 frames) to x264, 1080p using CRF18.0 with same parameters on both PCs.

My old PC with Q9550 Intel Core 2 Quad 2.83 GHz, 4 GB RAM, WinXP Pro SP3.

My new PC with Intel Core i7 6700K 4 GHz, 16 GB RAM, Win10 Pro (buggy crap!) - appeared to be running at 4.16 GHz (turbo mode, not user overclocked).

New PC took 1:46:18 (106.3 min)
Old PC took 5:44:45 (344.75 min)

New is 3.243x faster than Old.

I'm reasonably happy with that, hope other movies show a similar improvement.

Later: Did 2 other full BD encodes on both PCs: #2 was 3.02x faster #3 was 2.94x faster

[hydra3333]

x264 can use OpenCL for some calculations but the speed-up is barely noticable.
However, since the encoding chain includes a decoder and possibly some Avisynth filters, off-loading these tasks to the GPU will improve encoding speed. DGDecNV for decoding and KNLMeansCL for de-noising are two good examples.

I suppose it also depends what final quality you're happy with - using GPU encoding (nvenc eg with ffmpeg) gave me ~500fps .

Depending on your need, those kinds of speeds are kinda hard not to take seriously for "ordinary quality" h264 encoding .. especially since it can be combined with other GPU things like DGDecNV and KNLMeansCL.

A good GPU ... just a thought. It doesn't avoid the need for SSDs and a fast multi-core multi-lane processor/board, though.