[VIDEO] CPU Performance vs. Real-Time Performance in Your DAW

OP
rgames

rgames

Collapsing the Wavefunction
Great work Richard! One minor suggestion: a lot of what you talk about with the Bios, various video cards, are mostly PC only issues as Macs are designed to include components that work together optimally and higher quality than some PC makers or DIY guys choose. So perhaps the tile should read:

CPU Performance vs. Real-Time Performance in Your PC DAW
Maybe I don't completely follow but it sounds like you're saying Mac users don't have latency problems. That's clearly not the case, hence the reason for my confusion. I almost never have latency problems with my PCs - the last 3-4 machines I've built run just fine at 128 samples with no tweaks to anything (except the laptop shown in the video, but I haven't seen latency measurements on a Mac booting from a PCIe SSD either). But clearly there are folks who do have problems with PCs. The only thing PC specific in the video was the mention of the latency checker tools. I assume there are similar tools for Mac but I'm not sure what they are.

Regardless, Macs use all the same hardware that you get from HP or Dell or VisionDAW or whomever. Macs still have a BIOS. Macs still have video cards and hard drives and USB hubs that interrupt the processor for processing. And all of those devices are the same as on PCs. It is true, of course, that you have fewer options to tweak those devices on a Mac. But it's not because they're different or non-existent.

So I think the video applies perfectly well to both Mac and PC.
 
OP
rgames

rgames

Collapsing the Wavefunction
And to the hardware purists out there - someone pointed out that EFI is used in place of BIOS these days. Yes, it is - but I'll probably be calling it BIOS for a long time to come :)
 

EastWest Lurker

Senior Member
Maybe I don't completely follow but it sounds like you're saying Mac users don't have latency problems. That's clearly not the case, hence the reason for my confusion. I almost never have latency problems with my PCs - the last 3-4 machines I've built run just fine at 128 samples with no tweaks to anything (except thNo e laptop shown in the video, but I haven't seen latency measurements on a Mac booting from a PCIe SSD either). But clearly there are folks who do have problems with PCs. The only thing PC specific in the video was the mention of the latency checker tools. I assume there are similar tools for Mac but I'm not sure what they are.

Regardless, Macs use all the same hardware that you get from HP or Dell or VisionDAW or whomever. Macs still have a BIOS. Macs still have video cards and hard drives and USB hubs that interrupt the processor for processing. And all of those devices are the same as on PCs. It is true, of course, that you have fewer options to tweak those devices on a Mac. But it's not because they're different or non-existent.

So I think the video applies perfectly well to both Mac and PC.
Richard, all I am saying is that Macs come universally with optimal choices for their hardware and there are no Bios settings to tweak for audio. while PCs run the gamut, although as you say, the best ones use high quality stuff. There has to be a reason however why when I was looking to get a slave PC, so many people who are knowledgeable told me essentially "stay away from PCs you buy in a store from HP or Dell, etc. they are not good for audio work. DIY or get one from a company like VisionDaw."

Unless you think they all were wrong?
 
Thanks Richard for your contribution to the community.

But after watching your video, I am more confuse about my next server built.
I am deciding between 4790K, 5820K and 6700K.

My instincts tells me to go with the 5820K, but according to Richard speed is more important than cores (for sample streaming).
If some one can help me with the following questions I really appreciated.

. Those VEPRO take full advantage of multi-cores and multi-threads or it´s a urban myth?
. Is DDR3 a better real-time performance than DDR4 (because of CAS)?
. For sample streaming only, those the i5 6600K benchmark similar to i7 6700k?

Thanks
_Gabriel
 

kdm

Active Member
Very nicely done Richard. Just came across a question on this on another forum and referred them back here.

Gabriel2013 - I think VEPro may be somewhat of a deviation from the general cores vs. speed rule since it does split processes. I don't know for sure, so this is just my hunch that both speed and cores will help VEPro (perhaps speed having a slight edge in importance), but speed would be more useful for VIs within a DAW. This would be a good question for the folks at VSL.
 

ptsmith

Member
Thanks Richard. I've implemented some of your suggestions including lowering my RAM speed and seeing about a 30% increase in DAW performance. :thumbsup: Oddly I'm getting even worse CPU performance. It went from 40-45% usage to 32-35%.

On my DAW LatencyMon shows exactly what you said about the majority/all of the interrupts going to one core.

CPU.jpg
 

brett

Active Member
Lowering RAM speed? Was that in the video? I must've missed that

Have you done this in the bios? XMP profiles?
 

bcarwell

Senior Member
THANKS Richard for taking the time. A few folks may carp about a few nits, but you have performed an amazing service. Not to mention how professionally well-produced with graphics and animation, clear, and entertaining your talk is. You have a very real talent for addressing a highly technical issue in an extremely articulate, clear, thoughtful, organized and informative manner. Not bad for an aero. (BTW your prior system design video was equally first rate). So many folks are techies but cannot formulate an English sentence or communicate or they are good communicators and musicians but weak and often ill-informed in informational content. You've got them both in spades. I hope you will continue with these, and I noted that recently this has now been made a well-deserved stickie on vi-control.net. I'd sure love to see future videos on how your produce your music, e.g. template design, workflow, library selection, reverb, etc. Excellent work from a big fan awaiting more.
 

snattack

Senior Member
http://www.soundonsound.com/sos/jul15/articles/cpu-performance-0715.htm

More cores = better performance.

Multi-CPU = worse performance than one processor with the same total amount of cores.

Other factors with different stuff hogging the realtime chain is another thing of course. But the same motherboard with more cores equals better RTP.

Macs don't have a BIOS, they have EFI. And OSX is programmed differently than Windows when it comes to RTP, with driver priority and such.
 

neblix

Music, Math, Cats
There has to be a reason however why when I was looking to get a slave PC, so many people who are knowledgeable told me essentially "stay away from PCs you buy in a store from HP or Dell, etc. they are not good for audio work. DIY or get one from a company like VisionDaw."

Unless you think they all were wrong?
No one is trying to assert that HP/Dell computers are optimal for finely tuned workstation performance.

You are correct however that PC's require more manual tuning than Macs. Though this is not because of quality of parts, on the contrary, good DIY computers have way higher quality parts (such as power supplies and extensive motherboard technologies) than anything Apple or any other computer brand will ever offer to consumers because those latter components are designed with size and power constraints (so they often must compromise on the full capabilities of DIY hardware you'll find).

It's mainly because Apple and its programmers are allowed to make many assumptions about the hardware their software will run on; as such, many hard optimizations can be made that normally aren't possible in normal software development practices (which cater to maximum portability and scalability at the expense of hard optimizations). Because they have a history of using specific technologies that they never deviate from, they only have had to solve their issues like compatibility a long time ago, whereas in PC-based systems it's a recurring challenge since the hardware market is bigger, with more players, and everyone is given a fair chance.

You must realize though that in saying this, it's functionally equivalent to observing that a heavily modified monster of a race car made from various high end parts and assembly techniques requires more work than buying a Lamborghini. It is more work, but also way more reward in both flexibility and performance.

Of course this doesn't factor that you lose OSX on a PC, which is of course important to consider. Lamborghinis are stylish and fast, nothing wrong with them, I'm just clearing up that PC's do in fact have sizable benefits in multiple areas that make them a viable counterchoice to Macs.

PC's can be every bit just as if not more stable, fast, powerful, etc. than Macs if you dig deep enough. And it will cost less money, too. :)
 

Alatar

Active Member
Very pedagogical video. Thanks.

Here is a more technical video, which higlights the locking/latency problems of audio real-time programming (in C++):
 

peksi

Active Member
If I were to increase audio buffer a lot would I be able to reap the benefits from multi cpu system? That way the real time demand would be easier to fullfill - comparable to the audio mixdown process in Cubase which seems to utilize all cores to the max.

I have no need for live performance as I only do studio composing. I would gladly trade a slight lag for more power.
 
OP
rgames

rgames

Collapsing the Wavefunction
If I were to increase audio buffer a lot would I be able to reap the benefits from multi cpu system?
Up to a point, yes. But once the buffer is large enough then it won't make a difference if you increase it further. Where that point falls depends on your system and what you're trying to run. In my experience, it tends to be around 1024 samples. I think most audio interfaces don't offer buffers much higher than that for that reason - it doesn't make any difference.

Pretty much all DAWs take advantage of multiple CPUs, though, so that's not really a multi-CPU problem. Regardless of the number of CPUs, the total system performance will be better with a larger buffer *up to some point* because you're reducing the likelihood of real-time-audio limitations. But then, at that point, other limitations start to dominate.

In general, even audio mixdown can't use all available CPU power because it still has to wait on other subsystems to do their things (like the disk). Rendering video is different because it is working mostly in RAM and transfer to/from RAM is much faster than transfers to disk or the audio interface. The calculations take the majority of the time, so video rendering tends to be much more CPU limited.

Cheers,

rgames
 

peksi

Active Member
What is beyond my understanding is that why most DAW manufacturers have dual Xeon models at the top of their line? Based on the knowledge in this thread those DAWs perform RT-wise worse than their single CPU counterparts. Or do they know something we not?
 

David Hall

its difficult to learn but not impossible
Perhaps of interest to the folks here - I've found that the difference between real-time performance and CPU performance is one of the most-misunderstood elements of DAW performance. I've discussed it a bunch of times with a bunch of folks here and elsewhere, so I thought I'd put a video together to explain the difference so I can save myself some time the next time it comes up :)

Hopefully you'll find some useful info in this video.

wow... thanks Richard.

quick question is there a way to find out what part of your computer is causing the lock-up? like a troubleshooting way to figure this out?