Dual-Core CPUs

Is it worth the cost?

I’m about to purchase an Athlon 64 CPU for my home-built DAW.

Assuming a similar cost, would I see more performance gains from a higher speed CPU or a lower-speed dual core CPU when using N-Track?

I’ve read that I’ll only see a significant advantage from dual-cores when running multiple applications. And, running a single application may see less performance with a dual core.

Would the use of plug-in effects be considered additional applications when running N-Track?

Thanks - TJ

According to the build notes:

<!–QuoteBegin>

Quote
Build 1911 - Wednesday, July 13, 2005 3:31:02 PM EDT
Increased audio processing multithreading to take advantage of hyperthreading, dual-core and multiple CPU systems, removed “multithreaded DirectX plug-ins” option and added “Multithreaded audio processing” option. Increase in
performance should be ~10% with hyperthreading CPUs and ~50% with dual CPU


It’s not the amount of applications that you are running, but the amount of simultaneous threads that the applications are running which is important.

That being said, when testing these things I’ve never seen a massive change in n-track recording/playback when I turn hyperthreading on and off (though mixdown does show up a 10% difference).

I haven’t got a dual core CPU, but perhaps someone with one could run a test (with a few VSts running etc) with “muti-threaded audio processing” turned on and off and post the differences? Counter intuitively, the absolute cpu usage should actually increase - see below comment from Flavio:

<!–QuoteBegin>
Quote
The fact that the absolute CPU is higher is a good thing, as it might mean that some work is assigned to the extra (virtual in the case of Hyperthreading) Cpu. When not using multithreading the absolute cpu never goes above 50%, and that means that the extra cpu is idle. The important thing is that the time it takes n-Track to prepare a buffer is small enough.
Looking at cpu times during the playback of a song is not usually a very reliable measurement. Try to mixdown the song and check the difference in the time it takes for the mixdown to complete (as shown in the mixdown window).
You may have to repeat it a couple of times, usually by the 2nd time most of the .wav files of the song are loaded in the Windows cache.
Multithreaded processing does adds some overhead, but the benefits should compensate the overhead, although it might not be true for any song, as the degree of parallel processing depends on the structure of the song.
I’ve been testing mostly with the Sometimes sample song that can be downloaded from the download page at www.ntrack.com. On my hyperthreaded 3.2 Ghz P4 I’ve noticed a gain in mixdown time of slightly more than 10%.


Thanks John

Here are a couple of videos to sell the power of 64 bit applications.

64 bit videos

I’ll bet the 64 bit n-Track shows the same abilities

Thanks for the info!

The video is interesting on a number of levels. There are factual errors due to the obvious Intel sponsorship (AMD was the one who introduced 64-bit computing for the x86, not Intel who played catch-up). Also all the arguments of improved performance came from using a 64-bit address space to allow the use of 4 G of memory (which would be necessary to get the benefits). Nothing they demonstrated can be directly attributable to dual-core processors (although it is possible that it does make a difference, they did not demonstrate it and in fact carefully stepped around it).

What they really showed was that you can get an advantage from using massive amounts of memory for plug-ins that require massive amounts of memory. Duh!

Overall I view it as PR fluff aimed at selling Cakewalk and Intel products. Look elsewhere for the real story.

Jim

Thanks, Jim. I pretty much caught the Intel-Cakewalk PR angle as well.

This is what I have gathered from others I have talked to or inquired on line: The primary benefit of dual-core is running multiple apps concurrently or software apps specifically developed to run mulitple tasks concurrently. Currently, the latter is fairly limited, but will become more common as dual/muliple-core systems become more common. In fact, many applications may actually run slower due to the management overhead required for dual-core processing.

So for the present time, I’m going with the faster single-core AMD CPU and may upgrade to a dual-core in the future when I can see a significant benefit to do so.

I believe that in general, threaded apps should benefit from multi-CPU cores. Many apps are threaded already, though many aren’t. Apps like n-Track are excellent candidates for multi-CPU operation because so much of the work is parallel: results from plugins on one track don’t affect results from plugins on another, so it’s reasonable to run track plugins in parallel.

Whether it’s worth the money is another call altogether. With great workarounds like freezing, it’s definitely not necessary. But it’s certainly convenient and nice. It also depends on whether you use a lot of CPU-hog plugins. (Without plugins, n-Track doesn’t need a lot of horsepower. It’s the plugins that suck the juice.)

I think you would see more performance gains from a higher speed CPU than a dual core CPU because nTrack is probably not multi-threaded to any great degree, however I haven’t tried this yet.

My next DAW will be a dual-core. Why do you think they’re expensive? An Intel dual-core 3.0 GHz is around $260. The AMD’s are more expensive - yes.

<!–QuoteBegin>

Quote
Why do you think they’re expensive?

Read my post again: I assumed only that they cost more than single cores at the same speed.

n-Ttrack should be multithreaded pretty well. It should be relatively straightforward: one thread per track, aux, and group. This should go a long way towards sharing the load if you have plugins on more than one track, more than one aux, or more than one group. One thread for the master channel too, of course, but that won’t help much in terms of sharing the load.

Flavio says expect a 50% increase in speed, which sounds reasonable. A 50% faster CPU probably costs more than a dual core at the same speed. That’s speculation where facts are available, so forgive me for not googling the prices.

In any case, whether something is worth the price depends on budget and values.

To see the “mulithreadedness” (have I invented a new noun?) of the app you just have to run task manager and select View/Select Columns/Thread Count. Try this with and without “Multithreaded audio processing” and see the difference.

Ta
John

I asked Flavio the different benefits (as far as n-track goes) between dual core and hyperthreaded CPUs.
Here’s his response:

n-Track takes advantage of dual core cpus. Depending on the song structure and the buffer size the performance advantage over a single core might be up to 1.7x (70%), while with hyperthreading the advantage is around 1.1x (10%).

Thanks for the pointer, jm – I’d forgotten about that option.

James, those numbers don’t surprise me. The theoretical limit to the benefits of dual core is obviously 2 (100% faster). The costs include the thread-switching overhead (but with a threaded program you pay that whether or not you have dual cores, and it should be relatively small) and CPU scheduling overhead (which I have no idea about).

It also depends on how lucky you get about things like plugins: if you have a plugin on one track that takes way more CPU than all the other plugins put together, the dual core helps less than it otherwise would (but it still helps).

Also, I have a sneaky suspicion that we could find a case where we get better than 70% improvement, due to using plugins that are so CPU intensive they dwarf all other CPU usage.

Finally, bigger buffers will get more advantage with either hyperthreading or dual core, because there’s less context switching. Of course, bigger buffers causes less CPU usage regardless of dual core. As I said above, you’re paying for the thread switching overhead (or at least, some of it) whether you have the hardware features or not. The difference in CPU usage due to buffer size might not be very significant, though. For songs with few FX and lots of tracks, probably. For songs with few tracks and lots of FX, probably not because the thread switching CPU overhead is dwarfed by the FX CPU usage.

This is based on my understanding of scheduling on real-time operating systems; I don’t have any specific knowledge about Windows. The same general principles apply but the gory details can cause actual results to vary considerably. (In other words, theory only goes so far …)

<!–QuoteBegin>

Quote
I’ve read that I’ll only see a significant advantage from dual-cores when running multiple applications. And, running a single application may see less performance with a dual core.


Well, AMD are putting out an SLI type app/driver for their dual core cpu’s that lets it appear and run as a single core - Which will be huge for games - mainly because it’s widely considered that tehir next cpu won’t be up to the next intel release.

JM put that bit of fiction to rest in the first response.

You only get much of a benefit if there are multiple apps OR multiple threads. And keep in mind that the OS itself runs multithreaded and can take advantage of the dual core even if the app itself can’t.

But if you’re just running a single app and it’s single threaded, the advantage of a dual-core CPU is marginal.