RAM - a lamb. I am a dingdong?

[2Dogs]

I blame Ken Berry For all this!

In a recent post, he had the temerity to imply that I was being a cheapskate! In his words, "RAM is now almost dirt cheap!", and he suggested that I go buy some. Since he's wearing a suit in his avatar picture, I felt strangely compelled by this Sir Les Patterson-esque onslaught and went ahead and bought a matched pair of 512Mb "Rosewill" brand PC3200 sticks from NewEgg.com, for $75 inc shipping.

I admit, I am cheap! This particular RAM was advertised as being 2.5-3-3-8, however, in contrast to the 3.0-3-3-8 timings I get with the installed 2 x 256Mb Samsung units in my HP pc - so I thought that if anything, my Samsung memory would drag down the new stuff to work at the same fairly plain brown wrapper timing of 3.0-3-3-8.

When it arrived, I was encouraged to see "PQI" on the chips. Fairly reputable, then, I thought. Upon installation, however, I was not too pleased to see that the memory timings had gone to 2.5-3-3-7 and at 160mHz, instead of 200mHz! There were obviously compatibility issues between the Rosewill RAM and the Samsung.

Sure enough, when installed on its own, the Rosewill RAM got back up to 200mHz, and the timing improved to 2.5-4-4-8. It still didn't match the advertised 2.5-3-3-8, however, so I'm a bit miffed at NewEgg.com and will see what they say about it.

This all resulted in a more complicated set of combinations and permutations than I had hoped for. I thought I could simply try out combinations of RAM in single channel and dual channel mode, from a minimum of 256Mb single channel up to my maximum of 1536Mb dual channel. I hadn't anticipated the RAM configurations running at different speeds, which would affect some of the performance effects.

Still, I tried out 5 combinations. These comprised:

1) 256Mb in slot 1
2) 256Mb in slots 1 and 2
3) 256Mb in slots 1 and 3
4) 512Mb in slots 1 and 3
5) 256Mb in slots 1 and 3, 512Mb in slots 2 and 4

I tried to be systematic in all my tests. For each configuration, I first ran my washer program and then defragged all the drives, then booted up into my "Bare Bones" video editing profile, with only 13 services running, including Task Manager, Windows Explorer and of course Video Studio 9.

Booting into Windows XP took between 55 and 58 seconds - so no really noticeable benefits there in any RAM configuration.

Likewise, opening Video Studio took about 18 seconds, with perhaps a very slight improvement for the larger RAM configurations. (after closing the program, re-opening it took approximately 9 seconds)

I then tried direct to MPEG-2 capture, using a firewire cable to my MiniDV camcorder. In all RAM configurations, I was limited to the same 89% limit on the quality slider (that you get from the capture property options dialogue box if you turn off device control) that I've always had. Anything over this would result in the transcode buffer filling up and having to flush. I have to conclude that the process is CPU limited, and that mine is not quite up to it. (But my intention is not to resurrect any discussion on the merits/disbenefits of direct to MPEG-2 capture, it's just a useful benchmark, I am a believer in the recommended procedure....amen)

Next up I took a sample 3 minute AVI file of around 600Mb and tried "Smart Rendering" it. For repeatability, I did this test 5 times, erasing and washing the generated file each time before re-running the test.

Writing it from my E: drive (my slightly faster drive, a separate partition on my operating system drive) to another filename on that same drive resulted in a time of between 99 and 101 seconds in the various RAM configurations.

Writing the same file from the E: drive to my slightly slower G: drive, a separate physical hard drive, however, reduced the time in all the RAM configurations to 29 - 30 seconds. In fact the fastest 29 second write was using the 256Mb single channel RAM arrangement of case 1)!

I would guess that the process is governed largely by hard drive speed, so those fortunate Scuzzy and SATA folk may see better results. I also suspect that faster RAM might also slightly improve results. Overall, though, it seems obvious that you should write to another physical drive when you are trimming AVI clips - which this test really represents.

When I tested writing the same file from my slightly slower G: drive back to the E: drive, it took 34 to 35 seconds - with the 256Mb RAM configuration again being the quickest. This would seem to imply that it's better to read from a faster drive and write to the slower drive if you have different drive speeds. (I've unfortunately just acquired a 250Gb IDE drive with an 8Mb cache, as opposed to the paltry 2Mb on my existing drives, so I fear more tests coming up!)

Next, I tried encoding my test AVI file to MPEG-2, using the default high quality NTSC DVD profile - 8000 kbps VBR. I ran each test 5 times, erasing and washing the generated file each time for repeatability.

The 3 minute file took from 4 minute 24 seconds down to 4 minutes 16 seconds to transcode, with the better times from the two biggest RAM configurations. Writing the MPEG-2 file to the other physical hard drive gave only a 3 second benefit.

From this, I presume that the process is CPU limited, with only a slight benefit from more RAM. Task manager shows the CPU running at something over 50% - but that's with HT enabled. I know from previous tests that HT gives up to 15% or more benefit compared with using the CPU with HT disabled, even though the CPU then shows as running at 100%.

The last tests were "Smart Renders" of the MPEG-2 file. In these, RAM showed a very distinct benefit.

Using as a baseline the case 1) RAM configuration of 256Mb, and writing from my E: drive to another filename on the same drive I obtained the following results:

Case 2) (512Mb single channel) 43% speed increase.
Case 3) (512Mb dual channel) 43% speed increase
Case 4) (1024Mb dual channel) 106% increase
Case 5) (1536Mb dual channel) 106% increase

Now doing the same thing but writing the output file to the physically separate G: drive gave the following results, again related to the baseline Case 1) writing to the E: drive configuration:

Case 1) (256Mb single channel) 22% speed increase.
Case 2) (512Mb single channel) 65% speed increase.
Case 3) (512Mb dual channel) 65% speed increase.
Case 4) (1024Mb dual channel) 120% speed increase.
Case 5) (1536Mb dual channel) 120% speed increase.

It's a pity that the RAM in the case 5) configuration is running more slowly - so it masks any possible benefits that may arise from using more than 1024Mb. Unless NewEgg.com offer to replace my under-performing RAM with something up to the advertised speed, however, I suspect I'll keep my pc in the Case 4) configuration of 2 x 512Mb in dual channel. That does allow a bit more cooling of the dimms anyway, which is maybe not a bad thing.

You might also think the percentages above look a little suspicious - for eaxample the same 120% increase for both cases 4) and 5) above. Bear in mind, however, that these represent smart rendering the 3 minute MPEG-2 file in 15 seconds - and my stopwatch control is probably only accurate to half a second at best - so I rounded the output times to the nearest second.

From all of the above, therefore, I draw the following conclusions:

1) More RAM will not improve direct to MPEG-2 capture.
2) More RAM will not improve trimming AVI clips. Faster hard drives may do. You should write the trimmed clips to a separate physical drive if possible.
3) More RAM will not improve transcoding AVI to MPEG-2.
4) More RAM dramatically improves "Smart Rendering", which is relevant to the final burn phase, but does not amount to a great percentage of your overall video project time. (the best smart rendering results I achieved represented 12x real time)

Regarding item 3) above, my tests were of course simple transcoding of an AVI file to MPEG-2. I did not try rendering a test AVI with added sound, titles, effects, transitions and overlays. It may be the RAM has some effect on such transcoding. Another test for the future, perhaps.

Also, results may differ for different platforms, such as AMD machines, for example.

Overall, however, I believe there are only limited, specific benefits from having more RAM when using VS.

Thanks Ken!

[DVDDoug]

Interesting!

Ulead's System recomendations are 512MB or more. Only Ulead knows how the program utilizes extra RAM.

[2Dogs]

In order to satisfy my curiosity concerning the effect of increasing amounts of RAM on real-world avi to MPEG-2 rendering, I tried rendering a 60 second part of a clip, which featured two applied filters, an audio track, title and avi overlay track.

In the Case 1) configuration with 256Mb RAM in slot 1, it took 982 seconds to render.

That's less than 1/16 real time, so it demonstrates the effect of such additions to basic rendering of a "plain" avi file.

Increasing RAM to the Case 3) configuration of 2 x 256Mb in dual channel, the render time was reduced to 960 seconds - a 2.3% improvement.

Inthe Case 4) configuration with 2 x 512Mb in dual channel (though with slower timings as previously noted) the render time dropped only slightly further to 957 seconds. With the same RAM timing as the Samsung units, I'm sure it would have dropped a little lower than this, but in any case it was only a 2.6% improvement on the 256Mb Case 1) RAM configuration.

By contrast, overclocking my P4 2.8c in the Case 4) RAM configuration to 3.0gHz dropped the render time to 898 seconds. Overclocking to 3.2gHz rendered the file in 858 seconds. I suspect I'd need better RAM to be able to overclock any more, but it's interesting to note that the overclocked P4c still runs a lot cooler than P4e's, which are notorious "flamethrowers".