View Full Version : Avisynth - exploits CPU Core?
byme
5th September 2013, 12:26
I use Avisynth and Virtualdub together, but using several filters the speed of the process is very poor, and there are always so many hours of waiting
I thought then to change the CPU (then motherboard and Ram accordingly)
What I want to know is if Avisynth uses the CPU Core
The solutions caused by them may be, or an Intel Core i5 processor, or an AMD FX-8350 8-Core 4.0GHz
They told me that as benefits are equivalent (although here it seems to be more than the AMD, for the same price http://www.cpubenchmark.net/high_end_cpus.html), but that the i5 consumes less and warm. But if Avisynth exploited the Core, convenient to take the AMD, which is an 8 Core
Do you know something?
thanks
turbojet
5th September 2013, 22:06
Avisynth official builds are only single threaded which work just fine as long as there isn't cpu heavy filters being used, then avisynth MT could speed the process up a lot. The encoder matters a lot, if you are using xvid which doesn't have good multithreading the i5 will probably be faster because of the faster cores. x264 has great multithreading and makes use of all 8 cores in the AMD and probably around twice as fast as an i5 when encoding.
Asmodian
5th September 2013, 23:18
Avisynth MT is very unstable and I have found it to be more trouble than it is worth. Normal Avisynth only uses one CPU Core so an i5 4670K would be much faster than the FX-8350.
x264 should be around 33% faster on the AMD (FX-8350 vs i5 4670K). (http://www.anandtech.com/bench/CPU/28)
hello_hello
6th September 2013, 05:36
For me encoding usually keeps all cores busy but CPU intensive filtering does change that. Sometimes it'll drop CPU usage down to around 50%.
Being too lazy to upgrade to a MT version of AVISynth, I have a couple of simple workarounds. I either run two encodes simultaneously which keeps the CPU busy, or I create a copy of the AVISynth script and add Trim() so each script encodes around half the video. Once again I run them at the same time, then join the output video with MKVMergeGUI.
Blue_MiSfit
9th September 2013, 02:12
AviSynth does not multithread well. There are some filters that do multithread well, but generally AviSynth can be a huge bottleneck for modern multicore systems. I avoid it as much as possible these days, even though it's extremely powerful.
The MT builds work, but not reliably enough for me to call them "production ready".
byme
9th September 2013, 13:15
thanks for the answers
then, for the x264 is better AMD, and for Avisynth is better Intel
I'm interested especially speed Avisynth
although faster encoding with MeGUI and x264 would be a good thing
Avisynth MT is very unstable and I have found it to be more trouble than it is worth. Normal Avisynth only uses one CPU Core so an i5 4670K would be much faster than the FX-8350.
x264 should be around 33% faster on the AMD (FX-8350 vs i5 4670K). (http://www.anandtech.com/bench/CPU/28)
why the i5 with Avisynth is much faster than FX-8350?
Groucho2004
9th September 2013, 14:26
for the x264 is better AMD, and for Avisynth is better Intel
That statement is too general and under certain circumstances probably not true. The tool chain avisynth (especially MT) -> filters -> encoder is very complex.
However, it appears that Intel CPUs have the upper hand for Avisynth and x264 when you look at the performance/core. The FX8xxx is fast with x264 because it has 8 real cores.
I also consider efficiency very important (performance/watt), probably because I don't need an additional heater and I pay my own electricity bills. In this regard, Intel CPUs are much better than AMD.
I'm interested especially speed Avisynth
although faster encoding with MeGUI and x264 would be a good thing
I suggest that you try Avisynth MT if you are doing heavy filtering. Many people here have had good results with it. In some cases it simply does not work (unstable).
byme
9th September 2013, 16:44
but when the MT works, produces the same results that without the MT?.. or it may create defects in the video (that maybe you do not notice right away)?
Groucho2004
9th September 2013, 16:58
but when the MT works, produces the same results that without the MT?.. or it may create defects in the video (that maybe you do not notice right away)?
Recent MT versions appear to be pretty stable and I don't recall complaints about defects.
Still, you just have to try it yourself.
byme
9th September 2013, 18:51
ok, thanks Groucho2004... I'll try
What do you think though I bestow € 100 more and I bought the Intel Core i7 4770 (not K)?
I would have a gain with Avisynth compared to Intel Core i5 4670K?
the talk concerned also the same price... and the AMD FX-8350 costs as a Intel Core i5 3450 or 4570... however models below i5 4670K
Asmodian
9th September 2013, 20:28
My issues with Avisynth MT were crashes not defects. If it didn't crash it always worked well in my experience. I haven't used the newest MT versions, I turn down the speed settings of x264 until I get 100% CPU even with single threaded Avisynth. For SD video this does require running the veryslow preset with fast p-skip turned off on my i7 980.
With a pure Avisynth workflow I wouldn't expect much improvement from an i7 4770 vs an i5 4670K and if you overclocked the i5 (it is a "K" afterall) it would be faster. If you decided to use Avisynth MT (and your workflow ran multithreaded reasonably well) and you didn't overclock the i5 the i7 would be faster. A lot of "if" there. The i7 would be more help to x264, if you plan to use x264.
If you let us know what kind of workflow you plan to use with Avisynth some of us might know how well it runs multithreaded.
turbojet
9th September 2013, 23:37
The $130 FX8320 is 10% faster than the $350 i7 3770K with HD x264, with SD 3770K runs 10-20% faster than 8320 if intel quicksync is used for decoding to get 100% cpu. Avisynth MT slows things down quite a bit when using just ivtc/deinterlacing and built in resizing filters. I haven't seen a benchmark with x264's haswell's optimizations but that appears to give intel an edge on HD too.
If you are on a tight budget and do mainly HD encoding with x264 I'd suggest 8320 or 8350. AMD crippled my 8320 but changing boost multiplier from auto to 20x fixed it and puts it 200mhz shy of an 8350. You can build an 8320 box for the price of an intel i7 cpu.
If you encode SD and/or with xvid or mpeg2 go with intel haswell for it's igp, it makes a big difference.
If money isn't an issue go with intel haswell i7 as it should give the best speed and cheaper to operate then the amd's.
Groucho2004
10th September 2013, 09:25
The $130 FX8320 is 10% faster than the $350 i7 3770K with HD x264
Did you test that yourself? If not, where are the numbers from?
Asmodian
10th September 2013, 18:00
AnandTech's benchmarks, which I linked earlier, show ~9% performance lead for the i7 4770K vs the FX-8320 using x264 0.59.819 to encode HD video.
We still do not know if byme cares about x264 performance at all. Maybe they plan to encode to MPEG2. ;)
For Avisynth I am pretty sure an i5 or i7 would be faster but at a higher initial cost and with lower electricity bills.
Groucho2004
10th September 2013, 18:11
AnandTech's benchmarks, which I linked earlier, show ~9% performance lead for the i7 4770K vs the FX-8320 using x264 0.59.819 to encode HD video.
I don't trust the Techarp benchmark, the guys who created it are complete morons.
The FHD benchmark from Atak_Snajpera (can't find the results now) is better.
Asmodian
10th September 2013, 21:19
Interesting, sadly I cannot find any recent benchmarks using the FHD benchmark from Atak_Snajpera. None of the results I can find include anything from Intel at 22nm.
It seems the FX-8350@4Ghz is ~6.3% faster than an i7 2600K@4Ghz (http://www.x264fhdbenchmark.republika.pl/Results.txt) using an unknown version of the x264 FHD benchmark. This makes me think an i7 3770K or an i7 4770K would be significantly faster, especially if using a newer version of x264.
Groucho2004
10th September 2013, 21:58
This makes me think an i7 3770K or an i7 4770K would be significantly faster, especially if using a newer version of x264.
These benchmarks (http://techreport.com/review/24879/intel-core-i7-4770k-and-4950hq-haswell-processors-reviewed/12) (scroll down a bit) seem to confirm that.
turbojet
10th September 2013, 22:30
Did you test that yourself? If not, where are the numbers from?
With a guy who has a stock 3770K, on a 720p30 mpeg2 source to 720p30, your avisynth 2.6icl build with just dss2(), LAVfilters libav decoding, x264 --crf 19 --preset slow. Both maintained 99%+ cpu usage, 33.xx fps for the 3770K, 36.xx fps for the 8320. It's more like 9%. It was about a 5% lead for 8320 with the same source and lanczosresize(720,404) with libav decoding, 60-70% cpu usage. Switching to cuda and quicksync decoding 3770K won by 20-25% using 100% cpu, 8320+cuda only used around 80% cpu.
AnandTech's benchmarks, which I linked earlier, show ~9% performance lead for the i7 4770K vs the FX-8320 using x264 0.59.819 to encode HD video.
We still do not know if byme cares about x264 performance at all. Maybe they plan to encode to MPEG2. ;)
For Avisynth I am pretty sure an i5 or i7 would be faster but at a higher initial cost and with lower electricity bills.
More than likely that's a crippled 8320. By amd's specs an 8320 and 8350 are the same core with 200mhz difference in turboboost. My 8320 is specced at 4ghz turbo but it only ran at 3.5-3.7ghz under full load by default, 4ghz was only reached for a second or so of load. Forcing it to use 4ghz made about a 10% fps increase in x264 encoding. Bumping 8320 to 4.2ghz (8350) is fairly easy with an aftermarket hsf but I only have the hsf that came with it and the extra voltage required bumps it above 60C.
x264 is the only encoder I've used that the 8320 can keep up with i7's.
Under load intel definitely has a big edge in power usage. At the power outlet at idle/near idle, browsing web, email, etc. 8320+GTS250+2DDR3+2HDD+1SSD uses 80-90W, celeron 1610+9500GT+2DDR3+1HDD uses 60-70W. That's a very low voltage intel, 4770K is ~80W at idle according to sites. When x264 encoding with gpu decoding the 8320 uses ~350W compared to celeron's ~150W. Given the time it takes the 8320 is still more efficient in this case but apparently 4770K under load is about ~170W which would make it much more efficient.
I don't trust the Techarp benchmark, the guys who created it are complete morons.
The FHD benchmark from Atak_Snajpera (can't find the results now) is better.
I don't either, some sites have 8350 way ahead of i7's like anandtech, others have it way behind. The benchmark doesn't even work for me OOB and doesn't relate well to my typical encoding. But it does use a lot of the same parameters except bitness would be better if it used 32 bit always. I'll try the old FHD benchmark, have a feeling there's a lot of optimizations missing for the 8320 though.
EDIT: FHD benchmark didn't work either, first asks for admin priveleges then shows an info window that can only be closed.
ChiDragon
10th September 2013, 23:22
I'm running i5-4670 if anyone wants a particular benchmark run.
byme
12th September 2013, 01:21
I ordered a 4770 (not k)
as the price is a middle ground between a i5 4670K and a i7 4770k
I hope to have made the right choice
in theory the i7 4770 should have excellent results with Avisynth and good with x264
I have said that, however, the i7 4770 can reach high temperatures, also 80/90°C (I do not know if it's true)... So I was advised to add a heat sink, like this http://www.coolermaster.it/product.php?category_id=7&product_id=6777
What do you think?
well... if the only difference between the k and the non k, is the possibility of overclocking, I made the right choice... as wary of overclocking
http://www.cpu-world.com/Compare/516/Intel_Core_i7_i7-4770_vs_Intel_Core_i7_i7-4770K.html
Asmodian
12th September 2013, 02:20
The Haswell chips do run hot. It appears to be at least partially due to low quality thermal compound between the chip and the metal "lid" on the chip. Delidding the CPU and replacing the compound with something better can help a lot but that requires a razor blade and a hammer or a vice, a block of wood, and a bigger hammer. :eek:
My short experience with an i7 4770K and a Hyper 612 saw ~60°C when at stock speeds (3.9 Ghz) under full load (prime95 small fft). I didn't test the stock heatsink. I did have it set to run full turbo on all cores (a BIOS setting) even when at full load on all cores which is a kind of overclock.
I do think it would be a good idea to upgrade the heatsink for a long term 100% load system like an encoding box. Intel supplies a fairly minimal heatsink with the CPU.
I think you will be very happy with the i7 4770, that is top of the line if you do not want to overclock or go very expensive (LGA 2011).
nevcairiel
12th September 2013, 07:28
Aftermarket coolers are usually never a bad choice, however the stock cooler should manage to keep it at OK temperatures if your case is well ventilated and no heat builds up in the case itself.
byme
12th September 2013, 11:49
a technician denied the temperatures so high
told me that at maximum load reaches 75 ° C
But the Cooler Master I buy it too (lower teperature = lower consumption - less noise - more stability)
byme
12th September 2013, 11:54
The Haswell chips do run hot. It appears to be at least partially due to low quality thermal compound between the chip and the metal "lid" on the chip. Delidding the CPU and replacing the compound with something better can help a lot but that requires a razor blade and a hammer or a vice, a block of wood, and a bigger hammer. :eek:
My short experience with an i7 4770K and a Hyper 612 saw ~60°C when at stock speeds (3.9 Ghz) under full load (prime95 small fft). I didn't test the stock heatsink. I did have it set to run full turbo on all cores (a BIOS setting) even when at full load on all cores which is a kind of overclock.
I do think it would be a good idea to upgrade the heatsink for a long term 100% load system like an encoding box. Intel supplies a fairly minimal heatsink with the CPU.
I think you will be very happy with the i7 4770, that is top of the line if you do not want to overclock or go very expensive (LGA 2011).
well
thanks for the opinion
turbojet
12th September 2013, 21:43
It's a great choice if it's within your budget, generally the fastest cpu under $500, amd fx8 series only bests it in well multithreaded apps/games which is an area that keeps growing.
Intel specs (http://ark.intel.com/products/75122/) say 72.72°C, from experience the thermal shutdown is always a bit higher. For example: FX8320 max is 62°C according to amd, it shuts down around 69°C. I take the specs as more of a 'want to keep it under' instead of 'have to keep it under'.
Asmodian
16th September 2013, 01:57
The i7-3770K/4770K start down clocking to avoid going over 100°C, in my experience. :scared:
Groucho2004
17th September 2013, 09:50
If you're reporting CPU temperature you'll have to be more specific.
Modern Intel CPUs have two types of sensors. One is right under the heat spreader and the others are in the cores. With heavy load (x264), my i5 2500K @4GHz has about 55C-60C in the cores and 40C at the heat spreader.
The Intel specs actually refer to Tcase, which is the heat spreader sensor. The maximum temperature for the cores is about 100C.
Asmodian
17th September 2013, 18:32
I was referring to the core temperatures, not Tcase.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.