How does RAM performance effect rendering times for MPEG2?

[davpen]

I am looking at building a budget system primarily for MPEG2 compression using the MainConcept codec in MSP 6.5x. I have pretty much decided on an Athlon XP processor (probably a 2000 or 2100) but I am trying to decide what sort of platform to use. Will DDR333 make a noticable difference over DDR266 or even (gasp!) PC133? What about CAS latencies? Or dual channel DDR in the latest nForce2 platforms?

If the differences only amounts to a few percent either way I will probably just go for the cheapest but if the differences are more substantial I might consider spending the extra money.

Any suggestions appreciated.


Cheers,

David.

[Hulk]

Look at these benches:

http://www.hyperactivemusic.com/MS%20Pro%206.0.htm

The amount of RAM doesn't make much of a difference as long as the OS isn't page swapping constantly. I would say 256MB is okay if you aren't going to be multitasking, 512MB and you will definitely be safe.

FSB speed IS important. Get the fastest memory bus you can.
Dual channel nforce or SIS at 333MHz seem to be very good for MPEG encoding.

If I remember correctly, CAS latency isn't a huge factor. Sychronous memory/fsb timings and overall bandwidth are more important by far.

[Brian Ellis]

If it's just for compression, size AND speed are probably important. I would guess that 512 Mb at 266 MHz would do more for you than 256 at 333, though.

[Hulk]

I would go with 256MB of 333MHz over 512MB of 266MHz if MPEG rendering speed is of prime consideration. The extra memory will go unused during compression if multitasking is light. The extra 66MHz of bus/memory speed, on the other hand will definitely be noticed. Not really a big deal either way.

[davpen]

Thanks for the feedback. I have been convinced that it probably isn't a good idea to cut too many corners to save a bit of money. I will probably go with an nForce2 and a couple of 256M sticks of DDR333. Although I might wait a little bit and see how the new SIS746 looks. Though being only single channel DDR it probably will fall a bit short of the nForce2 for performance.

I didn't mention it earlier but I am currently using a Celeron 1200 on a VIA chipset with 512Mb of RAM and running W2K SP2. My rendering tends to be in of video lengths of between 100Minutes to 150Minutes using either 6000CBR for the shorter durations or 3000VBR for the longer durations. Compression time runs from about 6 hours to 12 hours. I would be hoping to get a 70-80% boast in rendering speed with an upgrade to an Athlon 2100XP.


Cheers,

David.

[Brian Ellis]

David

I don't think that CPU speed alone is proportional to AVI>MPG rendering speed. I've a friend who overclocked a Pentium III (I think about 500 MHz), adding extra cooling, by about 20%. Rendering speed hardly changed. I believe that it may be much more complex, involving FSB and memory speeds plus the latency. I know when I changed the m/b from P III 450 MHz to P4 1.6 GHz, both with 512 Mb RAM, the increase of rendering speed was only about 2 x, even with a FSB of ~500 MHz, not the 3.5 times that one may expect. Possibly reducing the latency in the BIOS may be effective, if you dare!

[Hulk]

Brian,

I hear what you're saying but it just doesn't sound right. I've reviewed all of the benchmark data I have and it looks as though encoding speed scales almost linearly with clockspeed, within the ranges of systems speeds tested. Of course there are losses in encoding efficiency as clock speed increases but they are not more than 10%. There may be something wrong with your friend's overclock, sometimes a overclock can slow the system down due to instability (but not enough to bsod) during operation.

P4's running MS Pro 6.5 (not 6.51), with RDRAM or DDR333 will run the test at about 175MHz to achieve 1fps MPEG-2 encoding. Note that RDRAM at PC800 is about equal (performance-wise) to DDR333 in this test.

A PIII is about the same in terms of efficiency.

A Athlon is much more efficient per clock due to the better floating point unit. About 135MHz to render 1fps. Obviously, although MS Pro is optimized for SSE and SSEII, it still relies heavily on the good old floating point unit.

The newer, faster Athlon boards do even better but I don't the numbers since these systems are running v6.51 of MS Pro.

Did you upgrade to a P4 with SDRAM? That does indeed kill P4 rendering performance, I know because I had one.

Hope that helps!

-Mark

[davpen]

Mark,

It is interesting that you should mention SSE and SSEII optimisations and the efficiency of the Athlon FPU. In addition to the 1.2G Celeron that I am currenly using I also have a 1.2G Athlon that I use as my main work system (I am a computer programmer). The Celeron has SSE but not SSEII where as the Athlon has neither. Both are running SDRAM with the athlon having a 133Mhz FSB and the Celeron a 100MHz FSB. Rendering tests on these two machines sugest that the Athlon is a comfortable 15% faster for compressing 3500VBR 720x480 under MSP6.51a. This suggests that SSE isn't playing a big role and that memory bandwidth and the efficency of the Athlon FPU is more important. I don't know, though, if the MainConcept encoder makes use of 3DNow on the Athlon. However in the same test using TMPGENC with a few options like Noise Reduction the Celeron is 35% quicker which suggests that TMPGENC is making very good use of SSE and has no 3DNow optimisations.

So from my point of view upgrading the Celeron to any Athlon is going to give me at least a 15% speed improvement. And since I would be going to a much faster Athlon with SSE and, at the very least, twice the memory bandwidth I should be able to acheive the performance increase that I would like.


Cheers,

David.

[Hulk]

The SSEI/II optimizations in the P4 (and in MS Pro) result in the P4 being "about" as efficient as a PIII in MPEG-2 compression in MS Pro. Without those optimizations, the P4 with it's relatively weak floating point unit gets creamed by the PIII, and even more so by the Athlon. Remember that the P4 must be paired with a fast memory subsystem, ie RDRAM or DDR333.

The optimizations are more useful when doing straight DV editing, there the P4 has some more "legs" on the Athlon. Still, the Athlon is a great choice for MS Pro. I can't wait to get my hands on a Hyperthreaded P4. But that's not going to happen until the prices come down to reality!

[davpen]

I have also done my timing tests on my brothers P4 2.53 that has 512M of DDR266 memory. It is about 40% faster than my 1.2G Athlon. A useful increase but not all that impressive considering the specs.

[Hulk]

I would be interested to see your brother's computer score on the MS Pro test linked in one of my posts above. I've been updating that site for a few years and have no data on a P4 2.53.

-Mark