Fractal VP4 Processors versus Axe-FX II, FM9, FM3 ... and implications for future products?

[likitlikeyoulikeit]

 

[DLC86]

PS: iirc recently someone on the forum mentioned there's a page on the fm9 UI that shows the percentage for cpu2, should be easy to find out with that.

I just checked and this CPU2 meter appears on the screen when editing delay and reverb blocks, when you approach 80%.
With all 4 of those blocks in the grid and the ultra-high quality capricorn type selected in both reverb blocks cpu2 was at 79.8, as soon as I deleted a delay block it went down to 72.8.
So they definitely run on the same core.

PS: if I select the diffused delay on both blocks cpu2 goes up to 89% but still works fine (fm9 turbo here)

 

[AlbertA]

IIRC the axe fx III standard has two 1 GHz cores and the turbo 1.25 GHz.

ah yes, just saw the bootup message on my "normal" for 1GHz

 

[unix-guy]

Are you sure? Afaik it's 2 cores each dedicated to one amp block, 1 core for reverbs and delays and 1 for everything else.

Imho what you said just doesn't make sense, if that was the case then an fm3 would be perfectly able to run 2 amp blocks cuz it would have the same core dedicated to amps (yeah, I know that core on the fm3 runs delays too, but no way 2 delays = 1 amp in terms of cpu).

PS: iirc recently someone on the forum mentioned there's a page on the fm9 UI that shows the percentage for cpu2, should be easy to find out with that.

PPS: another combination that imho makes sense could be:
Core 1: Amp1 + Delay 1
Core 2: Amp2 + Delay 2
Core 3: Reverbs 1 and 2
Core 4: all other FX

Yes, I am sure.

And the FM3 uses a dedicated core for both Amps and Delay.

Here Cliff references Amp modeling on a dedicated core:

FM9 Latency - Around 5ms?

Thanks for the details (mostly over my head). I suppose the option to choose lower latency vs better accuracy could be useful for those who want it but I defer to your judgement. I think a key question raised in this thread that you didn't talk about is why the FM3 latency is more in line with...

forum.fractalaudio.com

From the wiki:

CPU usage - Fractal Audio Wiki

wiki.fractalaudio.com

The Delay resp. the Reverb blocks on the FM9 run in dedicated DSP cores. When viewing a Delay or Reverb effect editing GUI screen on the FM9 hardware, a CPU meter will appear on the display if the CPU usage moves past 75%. As with the primary effects core, the recommended maximum CPU use value is 80% to ensure best performance.

Impulse responses in the Cab block are processed in an accelerator apart from the Sharc+ cores.

Announcing the New FM9 Amp Modeler/FX Processor

the SHARC dsp chips aren't the only processors in the unit. As well, the OS is going to take up CPU cycles by itself. If Cliff says it's about twice as powerful I would tend to believe him.

forum.fractalaudio.com

And finally, from the FM9 manual:

Different blocks have a different impact on the CPU. Some blocks such as Amp, Delay, and Reverb have hardly any effect on the meter at all because they run on dedicated CPU cores.

I have posted the same details many, many times on the Fractal forums.

If I was wrong, I would hope @FractalAudio and/or @Fractal M@ would tell me to stop spreading misinformation... Especially given that I'm an FM9 private beta tester.

 

[unix-guy]

I just checked and this CPU2 meter appears on the screen when editing delay and reverb blocks, when you approach 80%.
With all 4 of those blocks in the grid and the ultra-high quality capricorn type selected in both reverb blocks cpu2 was at 79.8, as soon as I deleted a delay block it went down to 72.8.
So they definitely run on the same core.

PS: if I select the diffused delay on both blocks cpu2 goes up to 89% but still works fine (fm9 turbo here)

Both CPUs have 2 cores - 4 cores, 2 CPUs

So most likely the Reverb is on one core of CPU2 and Delay is on the other core.

Core != CPU

 

[DLC86]

Hmmm... I'm still not convinced honestly.
"Both products use a dedicated core for amp processing" doesn't necessarily mean "both products use ONE core for all amp blocks".

The cpu meters on all other products always refer to a single core. Axe fx II, III, ax8 and fm3 all call "cpu" the one core dedicated to fx.

And also the manual reference is quite ambiguous and it's just a simplification which doesn't reveal any details imho.

As I said previously it wouldn't make any sense to me distributing the blocks among cores how you suppose they are distributed, considering how much cpu they use.

E.g. from the cpu usage of the fm3 (which has the same cores as the fm9, just 2 instead of 4) we know that a high quality reverb uses roughly 30% (depending on the type), and from the cpu usage of the axe fx and previous gen products we know that a delay uses way less cpu than a reverb.

That pretty much aligns with the CPU2 percentages I see on my fm9: 2 reverbs and delays = ~80%. One delay is 7-8% so with just the two reverb it would be around 65%.
If those blocks were split on two different cores and the CPU2 referred to a whole chip (aka 2 cores) I would expect to read about half of those values, a high quality reverb should use around 15% in that case, half of what we see on the fm3.

Anyway, I agree it would be nice if someone from Fractal chimed in to clear this up once and for all.

 

[unix-guy]

The cpu meters on all other products always refer to a single core. Axe fx II, III, ax8 and fm3 all call "cpu" the one core dedicated to fx.

Yep, I think all products report the CPU only for the "shared" CPU and not the dedicated ones.

I don't believe anything but the FM9 has a CPU2 meter and I was surprised recently about a quote from Cliff where he did say both Reverb and Delay could lead to the CPU2 Warning. But that also makes sense if they each use a core of one dual-core DSP (called CPU2).

In IT land, a CPU is a chip, which may have 1 or more processor cores. The cores are not referred to as CPUs. But this ain't IT

Something else I just thought about regarding Amp modeling: the FM3 runs Amps and Delays on 1 core. The FM9 is 2x the processing power, but that's by basically doubling the number of processors... So then possibly it does use 2 cores for Amp modeling.

Specs for anyone who really cares :)

Axe Fx III:

Processors

• Two 1.0 GHz floating-point “Keystone” DSPs (TMS320C66x) (2.8 times faster than the TigerSHARC DSPs in the Axe-Fx II). The Turbo module has a 1.25 GHz processor

FM9:

• Two dual-core SHARC+ DSPs (original: 2x 450Mhz, Turbo: 2x 500Mhz)

FM3:

• Processors: SC587, a 3-Core “Griffin” DSP with one ARM and two SHARC+ cores. Dedicated GUI processor. Cabinet modeling runs in a CPU accelerator

Although some quotes from Cliff seem to confuse the matter:

The III uses (1) dual-core Texas Instruments DSP. The FM3 uses (1) dual-core Analog Devices DSP. The FM9 uses (2) dual-core Analog Devices DSPs

Axe Fx II:

• Two TigerSHARCs. One dedicated to amp and cab modeling, the other to everything else

AX8:

• Processors: two dual-core 450 MHz ADSP-21469s, two microcontrollers (one is dedicated to amp modeling, the other to effects and housekeeping). Cab processing runs in an accelerator

FX8:

Can't find the processor specs.

And also the manual reference is quite ambiguous and it's just a simplification which doesn't reveal any details imho.

Agreed.

Anyway, I agree it would be nice if someone from Fractal chimed in to clear this up once and for all.

Yes - that way if I'm wrong I stop telling people bad info!

Edit:

Turns out I was wrong and @FractalAudio posted a clarifying comment in a thread on the Fractal forums.

@DLC86 for the win!

 

[DLC86]

@unix-guy we finally had our answer: https://forum.fractalaudio.com/thre...-dsp-allocations-wasteful.209303/post-2614814

EDIT: I just saw you already participated in that thread, nevermind

 

[unix-guy]

@unix-guy we finally had our answer: https://forum.fractalaudio.com/thre...-dsp-allocations-wasteful.209303/post-2614814

EDIT: I just saw you already participated in that thread, nevermind

Damn it - I was searching all over the Fractal forum to let you know and totally forgot it was on TGF!

Just searched for the thread and see you already know :)

I've added it to the Wiki!