Multicores Cpus Really Used In Creator 2009?

[Xyzzy]

Does C2009 uses mutiple CPU cores to speed up video rendering?

[jeanrosenfeld]

This is on my one computer with specs in signature.

 

I loaded two AVC/H.264 files into Vodeowave, duration just over 24 minutes. Rendered to mpeg2 for Video best quality.

 

Test 1: in task manager, set affinity for Videowave11.,exe to just 1 CPU. Rendered in 18 minutes.

Test 2, set affinity to all 4 CPUs, Rendered in 9 minutes 30 seconds.

 

As you can see from the performance tab, with 1 CPU it is using it at 100%, so speed limited by CPU.

At 4 cCPU, it is using only around 60% of all four cores, so speed limited by L2 cache (which is shared by the 4 CPUs.

[Big_Dave]

Jean,

 

That's very good results on the value of multi-cores for the test that you executed. There are probably a number of factors other than cache that are limiting the four processors from reaching 100%. In both of your test cases, page frames were minimal and available cache was significantly high. Memory throughput and I/O throughput limitations are most likey inhibiting the four processors from reaching 100%.

 

It's a vicious circle: memory, cpu and I/O. If you relieve one bottleneck then the others will be the bottlenecks. We have to remember that cache is nothing more than speed matching buffers. Internal path speeds and memory thruputs rates are not fast enough to keep the processors busy 100%. Faster pipe line and faster caches will allow higher procesor utilization.

[jeanrosenfeld]

Good example that proves you aren't getting 4X the speed when using 4 core. You only get twice the speed. Go figure!

 

Now if there was a such thing as a single core running at 10Ghz (2.4Ghz X4), you would probably get TRUE 4X the speed.

 

Here is another set:

1 core: 14 min 33 sec

2 core: 8 min 32 sec

3 core: 7 min 27 sec

4 core: 7 min 12 sec

 

I agree, other limiting factors come into play and additional cores have diminishing effects. But still happy to take the factor of 2 :-)

[ggrussell]

Sorry, nobody seems to know.

Yes, Videowave/MyDVD are multi-threaded. Now whether or not that SPEEDS up rendering is the discussion. I tend to agree more wth JamesTW.

 

Multi-threading came long with single core Pentium 4 which 'fooled' the OS into thinking there were two processors. Although the threads could be executed separately, one thread would have to wait until another thread was executed, etc. Now with real multi-core processors, some threads can be processed in 'parallel' on separate cores. But the question now is the program code 'optimized' for this parallel processing? I'm afraid only the Sonic/Roxio programmers could answer that.

 

 

Personally, I would still rather have a single core running at 6Ghz than a dual core running at 3Ghz.

[Chappy]

Hey Folks...isn't this "Digital Life" fun!!!!!!!

Big Dave, you've definitely done your homework on Multicore processing and you understand the dynamics of I/O and bottlenecks well. Also well explained. I remember very well how many times I had to stop doing stuff on my 3.4GHz (OC'd fro 2.6) P4 Northwood, simply because my AV scanner started doing it's thing. That's a thing of the past on my Core 2 Duo E6750 and I certainly don't miss it.

Jean, excellent reports!

 

One other thing is...do we want a rendering process to consume 100% of a multi-core machine just to make it "that much faster"? I think not because then we lose one of the REAL aspects of multicore processors...multitasking. I would rather the render process stay at 75-85% and allow the rest of the power to continue other tasks. I have noticed a Marked increas on rendering a movie to DVD even with EMC10, at least twice as fast than my P4 was able to accomplish the same task. Software today is written more and more to take advantage of multicore threading so we'll see better performance in more programs soon. 2009 is written to take advantage of this and in certain areas we are bound to see improvement over other versions, but to think that 2 cores will double, and 4 cores will quadruple these improvements is simply showing a lack of understanding in what multicore processing is and how it works.

 

Gary, I believe you just set yourself up for disappointment in thinking how you did when you switched to your newer processor. Like others, I don't notice any real increases in how Fast my regular programs work now, but I sure do notice how many more things I can do at one time now over my P4, which was "theoretically" faster than my current processor but P4 had a much smaller cache and didn't properly access the L2 cache either (thru FSB).

[gi7omy]

It makes not one scrap of difference to rendering time to be honest

 

What does make a difference is the amount of Level 2 Cache on the CPU - when I changed from the AMD 64 to the Intel e6750, there was a marked increase but that was down to the 2 MB L2 cache per core

[JamesTw]

I think most people are under the impression that multi-core would make everthing faster. It doesn't really. As Daithi points out, what really makes a difference is shear CPU speed and the L2 cache.

 

If the software is written to use multi cores then you will probably see a difference but here's an explanation which is taken from wikipedia.

 

The amount of performance gained by the use of a multicore processor depends on the problem being solved and the algorithms used, as well as their implementation in software (Amdahl's law). For so-called "embarrassingly parallel" problems, a dual-core processor with two cores at 2GHz may perform very nearly as fast as a single core of 4GHz.[1] Other problems though may not yield so much speedup.

This all assumes however that the software has been designed to take advantage of available parallelism.

If it hasn't, there will not be any speedup at all. However, the processor will multitask better since it can run two programs at once, one on each core.

 

There you have it........... Multitasking is normally faster as each window will run under a separate core but video rendering and other usage purely depends on how the software is set to use multiple cores. If its not the raw power of the cpu and size of cache will determine the speed.

 

Jim

 

 

[oneeng]

Multi-core does appear to work well in VideoWave (I haven't tried checking for the other applications, nor do I think it would make much difference in them).

 

Multi-threading is not a gimmick, and while cache size does indeed help rendering substantially, the ability to render in parallel has a much greater potential.

 

In benchmarks on the web, VideoWave has been shown to see substantial improvements when moving from 1 core to 2 cores; however, it does not scale to 4 cores at the time of the latest benchmarks.

 

Video is an IDEAL application for multiple cores. An "n" thread application would be quite easy to apply to video rendering since each frame of rendering is independent from the next frame rendering.

 

Cinibench is a good example of how even a single picture can be divided and rendered in parallel. If you have 4 cores, the picture renders in 4 parts simultaneously (while you watch).

 

It is my belief that VideoWave is currently limited to 2 threads. If this is true, then the use of a 3 or 4 core processor would not make much sense (which is why I currently use only a core 2 duo with 4 Mb L2 shared cache).

 

As for the cache issue, Intel processors (Core 2 family) have a shared L2 between 2 cores. This allows a single core to have access to a very large cache when it needs it AND for the 2 cores to share the cache as well. In addition, the integer computing unit on the Core 2 is amazingly fast. This is the predominant part of the CPU that is exercised when rendering.

 

My personal recommendation for a processor suited for Roxio (best performance without wasting money on cores or features that will not help you), is the E8500 which runs at 3.16Ghz, has 2 cores and 6Mb of L2 cache: http://www.pricewatch.com/cpu/core_2_duo_e8500.htm

 

For ~$180.00 you can't beat it (at this time).

[ggrussell]

Video is an IDEAL application for multiple cores. An "n" thread application would be quite easy to apply to video rendering since each frame of rendering is independent from the next frame rendering.

For rendering MPEG 2, that is not true. In fact with MPEG 2, there are three different type of frames, I, B & P.

Each Group of Pictures (GOP) starts with an independant frame. The following frames are then created depending on the next following frame(s) and sometimes the previous frame(s). How much compression also depends on on how much motion occurs during those frames.

 

I'm not familiar with MPEG4 or AVCHD, but I'm sure it's just as complex or even more. This would seem to me that video rendering would not be 'ideal' to parallel processing at all.

 

As far as 'benchmarks', I've never trusted them from any source. Just like polititians, the results can be scewed or interpreted to make any hardware look good or bad.

[oneeng]

From here: http://www.anandtech.com/printarticle.aspx?i=2866

 

If video encoding is well threaded, good scaling per core can be found. I did a bit of testing on my C2D 2.33Ghz 4Mb L2 and found that it made little difference with Videowave if I used 1 core or 2. In both cases I used about 50% of the available CPU resources.

 

In the case of a single core, I found that it maintained ~100% of the single core. In dual core, it maintained about 50% of each of the 2 cores.

 

I also have only a single channel of memory (because of a bad DIMM), so I may well be memory bandwidth bound.

 

In the case of DivX 6.1, it appears that moving from 2 cores to 4 cores scales from 44 down to 30. Not perfect scaling by any means, but still significant.

 

In the case of Windows Media Encoder 9, it scales from 50.4 to 78.9. Again, not bad, but not perfect either.

 

Video encoding SHOULD be a near ideal case for multi-threading. Even if all you did was divide the video into "n" parts and rendered each part in parallel .... then sew the "n" parts together at the end, you would experience near perfect scaling with the number of cores as long as you were not memory bound or disk I/O bound.

 

I am seriously now considering ordering a set of RAM DIMM's to restore my dual channel memory setup

[Jim_Hardin]

Here is a trusted 'benchmark' I always use… Open MS Hearts and play a few games. It really doesn't matter how "fast" the PC is supposed to be or what benchmarks it set. If the programs I use don't run any faster, all of the tests are meaningless.

 

Those that have multi-core and run editing programs on them don't see any significant difference.

 

Gary has a more current version of Sony Vegas than I have. Seen any improvement in that rendering slug?

 

[Big_Dave]

The article referenced above at http://www.anandtech.com/printarticle.aspx?i=2866 is excellent but dated a bit.

 

The extra overhead in OS in support of multi-cores is going to be with us. Processor dispatcher code, resource locks (queues) and alike will get installed with the modern OS. Until Intel and MS improve on use of multi-cores, will we continue to see diminshing returns with adding more cores. Eight cores (oh my gosh) will not do eight times the work of a single core. "Large scale processor effects by IBM" shows the diminishing returns by adding more cores. Manufacturers are hitting the wall so to speak with heat, ghz and internal path lengths, hence it's cheaper to produce multi-cores than a single 10 ghz processor.

[Big_Dave]

Depending on the Duo Core platform and the process being executed, you might notice a big difference. I did some testing with Videowave EMC 9.1 on two different PCs and noticed 45-54% improvement. All processes available with Videowave might not enjoy such an improvement. It all depends on what you are doing.

[ggrussell]

None LOL I'll need 8 core to render hidef. (tongue in cheek) I'm getting about .5 fps rendering or less at full 1920X1080i. Rendering to HDV or AVCHD is really slow. I'm still waiting for those 6GHz processors.

[gi7omy]

Never notice Gary that CPUs seem to have levelled off at round the 3 GHz mark?

 

It looks like the only future developments will come in the way of multi-core and larger L2 cache.

[cdanteek]

Here is a trusted 'benchmark' I always use… Open MS Hearts and play a few games. It really doesn't matter how "fast" the PC is supposed to be or what benchmarks it set. If the programs I use don't run any faster, all of the tests are meaningless.

 

Those that have multi-core and run editing programs on them don't see any significant difference.

 

Gary has a more current version of Sony Vegas than I have. Seen any improvement in that rendering slug?

 

I did my own testing.

 

One machine a P4 Northood 2.8 ghz 512kb L2 cache, machine two C2D 3.16 ghz 8500 6mb L2 cache, file was a 7.52gb mpg2 video file with ac3 audio.

 

I analyzed and compressd the file on both machines to a single layer DVD. Machine one took 12 min analysing and 31 min to re-encode it, at 6,000 to 7,000kb/s and 209 frames a sec. Machine two took 2 min analysing and 5 min to re-encode it, at 29,000 to 30,000kb/s at 940 frames a sec.

 

I agree with the L2 cache but I can't buy a P4 today with 6mb L2 cache to run a actual test! I would think FSB speed and other factors enter in, at the end of all the arguing the Multicore Cpus with larger L2 cache are waaaaaaaay faster, aren't they?

 

I think Big_Dave and I are seeing the same results.

 

cd

 

[gi7omy]

Part of the problem is that Intel et al are trying to implement Moore's Law - what is happening is that, as dies get more and more semiconductors jammed on to them, they are running into a barrier of quantum levels.

 

To get round this and keep increasing the 'head count' they are pushing multi-core technology. Speeds aren't likely to rise past the 3 GHz level using gallium arsenide although there are some developments in doping zinc oxide with magnesium (the so-called 'spintronic' semiconductors) which may (or again, may not) push the semiconductor count per core even higher, giving increased speed.

 

All very well in theory, but until higher speed RAM and drives come along to complement this, it's all still in the air. There are possibilities that the existing solid state (pen drives) can be increased in size to come up to hard drive capacities and there was work being done some time back on organic based RAM (but that hstill in the laboratory stage and still under development)

 

There is a series of articles here on the CPU and Moore's Law.

 

For organic semiconductors, Sanjay K. Ram's blog has some information and links to papers

 

 

[ggrussell]

Never notice Gary that CPUs seem to have levelled off at round the 3 GHz mark?

 

It looks like the only future developments will come in the way of multi-core and larger L2 cache.

Yes, and that was reached over two years ago. Then Intel released Core Duo and started back with 1.8Ghz. They are just now getting back up to 3GHz level. Most of that was because of heat and watts used with the older technology. Now that both companies have 45nm dies, perhaps we will see an slight increase in processor speed sometime soon. Lower wattage and less heat at 45nm.

 

I've never thought about overclocking, but it's tempting just to get a faster CPU. I've seen some CPU coolers that looked like something from a car. LOL

[Big_Dave]

For those of you that don't agree that multi-cores are good for Creator 2009, then how about providing some benchmarks of Creator 2009 with multi-cores enabled and then with just one core enabled. Benchmark 2-3 different processes in Videowave and then report back.

 

VISTA with "superfetch" will use all the level 2 cache and memory that is available. However, there is a point where more memory will not return performance benefits. Effective cache hit ratios above 90% will not show much of an improvement on the wall clock.

 

Internal bus speeds, memory throughput rates (not just mhz), L1 cache speeds, pre-fetch routine effectiveness, OS dispatcher effectiveness (internal processor inference minimized), I/O throughput rates, I/O instruction path length, video card thrughput rates and many other factors will come into play with "system" performance.

 

Bottom line is that multiple cores are not going away. Expect to see higher throughput rates as Intel and Microsoft improve the effectiveness of multiple core processing.

[grandpabruce]

[snip]

 

I've never thought about overclocking, but it's tempting just to get a faster CPU. I've seen some CPU coolers that looked like something from a car. LOL

 

We may be too old to overclock anything.

 

[cdanteek]

This is on my one computer with specs in signature.

 

I loaded two AVC/H.264 files into Vodeowave, duration just over 24 minutes. Rendered to mpeg2 for Video best quality.

 

Test 1: in task manager, set affinity for Videowave11.,exe to just 1 CPU. Rendered in 18 minutes.

Test 2, set affinity to all 4 CPUs, Rendered in 9 minutes 30 seconds.

 

As you can see from the performance tab, with 1 CPU it is using it at 100%, so speed limited by CPU.

At 4 cCPU, it is using only around 60% of all four cores, so speed limited by L2 cache (which is shared by the 4 CPUs.

 

 

Very good demonstration Jean, thanks..

 

cd

[Big_Dave]

Bruce,

 

Do you mean "overtalk"?

[ggrussell]

For those of you that don't agree that multi-cores are good for Creator 2009, then how about providing some benchmarks of Creator 2009 with multi-cores enabled and then with just one core enabled. Benchmark 2-3 different processes in Videowave and then report back.

Already have done my testing. My previous CPU was an AMD single core 4800+. When I purchased my latest Dual core 3GHz machine, I was expecting TWICE the speed. Two cores Right? Nope, not even close. What meager increase I got was due to the difference in processor speed and updated FSB, etc. Was I happy? Sure, but I certainly didn't think I got my money's worth.

 

I stick with my statement. A single core running at 6GHz will run circles around a Dual Core at 3GHz regardless of cache levels. Nothing beats raw power!

[ggrussell]

Test 1: in task manager, set affinity for Videowave11.,exe to just 1 CPU. Rendered in 18 minutes.

Test 2, set affinity to all 4 CPUs, Rendered in 9 minutes 30 seconds.

Good example that proves you aren't getting 4X the speed when using 4 core. You only get twice the speed. Go figure!

 

Now if there was a such thing as a single core running at 10Ghz (2.4Ghz X4), you would probably get TRUE 4X the speed.

[smfx]

Switching from a single 3Mhz system to a dual processor dual core (four separate execution paths) system didn't significantly decrease rendering time. However, what did change was the amount of other things that I could do while the rendering was occurring - major improvement there. Which is what mult-processor with the current code environment is really good at - doing separate things faster, not one thing much faster. For it to do one thing much faster the software would have to be written specifically for multi-processor/multi-core if the task at hand even allows it (a surprising amount of tasks don't really adapt to it well or at all). That software would then preform miserably on a single core system as it would have the considerable overhead of parallel software without the benefit of the hardware.