Silicon Audio

Hi Noel, 
 
The AVX testing in question shows that Copy Int24toFloat64 gets almost as much accelaration on the Bulldozer as it does on the newer Intel chips.  However, the Copy Float32toFloat64 showed a 77% loss on the AMD.
 
Can you explain in laymans terms when Copy Float32toFloat64 would be used and what the real-world disadvantage would be for a Sonar user?  I think this is quite important for us as users to be able to make an informed choice when building a new DAW.
 
Thanks!

CopyInt24ToFloat64 gets more use typically than the other function. Both these functions get used a lot during mixing using the double precision engine, since we upconvert 24 bit audio in tracks to 64 bit before it hits the mixer. The 32 bit case gets exercised when streaming 32 bit track data (typically from bounces/freezes) and also when upconverting the output of plugins that do not support 64 bit double precision.
These functions are also used when bouncing audio to 32 or 64 bit files.


I wouldn't use just AVX performance to pick between Intel and AMD though. The above was only one common hotspot that we optimized for. 

Talking about processor level optimizations X2 has some new opts for sample rate conversion. This gets used whenever you export to a different sample rate and also when you import audio into a project at a different sample rate. We initially optimized for AVX but it turned out that the AVX version of the SRC was actually slower than the standard version!
So we ended up doing an SSE2 optimized version of the same which was much faster. In our sandbox this performs about 27% faster than the standard version. So you should see a nice boost when exporting or importing wave files with SRC.
AVX doesn't necessarily apply well to all contexts - it depends on the kind of processing loops and how you are accessing the data. In this case SSE2 was faster because the overhead of setting up for AVX outweighed the benefits.  

Thanks so much for taking the time to reply, Noel.  This is really useful information and personally helps me a lot.

Talking about processor level optimizations X2 has some new opts for sample rate conversion...

Very cool, Noel.  I guess this is the kind of stuff you guys beaver away at that doesn't necessarily make the glossy flyers, but bit by bit makes Sonar better every version.


Thanks again for the peek behind the curtain.


Bill.

Noel Borthwick [Cakewalk
]

Silicon Audio

Hi Noel, 

The AVX testing in question shows that Copy Int24toFloat64 gets almost as much accelaration on the Bulldozer as it does on the newer Intel chips.  However, the Copy Float32toFloat64 showed a 77% loss on the AMD.

Can you explain in laymans terms when Copy Float32toFloat64 would be used and what the real-world disadvantage would be for a Sonar user?  I think this is quite important for us as users to be able to make an informed choice when building a new DAW.

Thanks!

CopyInt24ToFloat64 gets more use typically than the other function. Both these functions get used a lot during mixing using the double precision engine, since we upconvert 24 bit audio in tracks to 64 bit before it hits the mixer. The 32 bit case gets exercised when streaming 32 bit track data (typically from bounces/freezes) and also when upconverting the output of plugins that do not support 64 bit double precision.
These functions are also used when bouncing audio to 32 or 64 bit files.


I wouldn't use just AVX performance to pick between Intel and AMD though. The above was only one common hotspot that we optimized for. 

Talking about processor level optimizations X2 has some new opts for sample rate conversion. This gets used whenever you export to a different sample rate and also when you import audio into a project at a different sample rate. We initially optimized for AVX but it turned out that the AVX version of the SRC was actually slower than the standard version!
So we ended up doing an SSE2 optimized version of the same which was much faster. In our sandbox this performs about 27% faster than the standard version. So you should see a nice boost when exporting or importing wave files with SRC.
AVX doesn't necessarily apply well to all contexts - it depends on the kind of processing loops and how you are accessing the data. In this case SSE2 was faster because the overhead of setting up for AVX outweighed the benefits.  

.........Hmm..Sooo...does this mean my guitar will be even phatter in X2??
 
Edited for Smiley face

By some freak chance of clicking about while investigating possibility of upgrading my AMD 4 core to an AMD 8 core, I wound up finding this thread.  Made me think.  When I first upgraded to X1, I was running an Intel Pentium D940 dual core under XP with 4 gigs of RAM, only 3.2G of which were recognized by XP.  This set up didn't even meet the minimum standards for X1 ... and ... it worked fine ... running maybe 40 audio tracks, a few soft synths, and a wad of effects.   Did stuff I'm still proud of.  Point being is that I don't think a lot of this stuff really matters as much as being able to muster up some creativity.  I could be running on a multi-cpu server designed by aliens from Planet X, and if I sucked, I'd still suck.  The ability of Intel or AMD to solve the real problems we musicians all face is nigh unto zero.  Now ... would I go back to that old machine?  Hell no.  I'm crazy.  Not stupid.  Now would I build another DAW based on the Xeon and 64 gigs of ram?  If I had the money, probably. 

I'm running an AMD 8-core (fx-8350) and it absolutely flies in X3. I did have some issues at one point after upgrading to Win8.1 with not being able to get latency down low on my interface, but as soon as I set the Power Management to Maximum Performance and manually disabled Core parking, it flies. I can live monitor my drums (all 15 channels) while playing, plus a multitude of other audio/VST tracks at fairly low latency (192 buffer) with all effects on. All this with relatively low CPU usage and no noticeable latency (something that's very easy to pick out when you are live monitoring drums).
 
I may be able to get even better performance with an I7, but the CPU (and overall DAW) that I have is definitely no slouch.

Good thread.  Just chiming in to stay notified. 
Nothing to see here. Carry on.