New PC Computer Build

[ElmerJFudd]

Thanks, guys. Helpful.

Eventually Ill swap the motherboard and processor in one go.

But its running so well now with just the case fans and stock intel heat sync that came with processor. Very quiet.

[Throbert]

I bought 4 x i7 4790k CPUs last Spring.

2 x spares and 2 x separate live rigs.

Im good for years to come as all other parts are stocked.

 

You can buy the i7 4790k at most MicroCenter stores but only walk in only.

There might be some still. I paid 250 each.

The newer CPUs are nice but single core performance has not really improved much.

And I do not want Windows 10 yet.

 

I also use the i7 5775C CPU that would fit your Gigabyte Mobo.

Those are pricier but the best audio CPU Ihave.

Its got 128MB Cache for GFX.

Use a discrete GFX Card then that Cache becomes usable for your audio.

It really kicks ass for synths locked to a single core, and can take all polyphony you throw at it.

 

If you can find one new for under 400 its worth every penny.

Ha, bought my 4790K at MicroCenter as well, best deal at the time. Saw the 5775C

listed when looking earlier but the low base Ghz seemed disappointing. didn't pickup

on the Cache but thought that was why the low Ghz, Intel seems to always give and

take that way.

[hardware]

Its OCD to 3.8 with the Cache (RingBus) underclocked (or it BSODs) so its as fast as my 4.4 k series, but runs cooler.

After my outdoor experience with 100% humidity & 110 degrees I took a builders advice on Supermicro/workstation rack design.

Smooth sailing at all events last summer.

 

[Markyboard]

I'm finally done - YAY! Put my previously problematic soft-synths/VSTs to the test - here are some performance details for anyone interested:

 

Accoustic Samples Wurlie plug (UVI/Reaper) : previously giving occasional audio pops at buffer size of 32 and 64. Now absolutely nothing at 64 but still occasional pop at 32 if I pedal a whole lot of notes.

 

Scarbee Classic EP-88s plug (Kontact/Reaper): Previously unusable to me due to pedal release audio pops even at buffer setting of 256. Still get them at 32 but absolutely none at 64. Also found that running Kontact stand-alone at 32 works much better than through Reaper (makes sense), although with a pedal sustained glissando I can still get the pedal pop.

 

Diva plug (Reaper): l haven't been able to break it with buffer size 32. Even with the most aggressive quality settings and 16 note polyphony. Previously it worked pretty well but if pushed too hard it would choke.

 

Reaper: Boot-up time is now about 40 sec. down from 65. I don't multi-track more than 2- 4 tracks at a time so no comparison here.

 

This build is near dead silent with nothing else in the room masking the noise. And I have yet to put the cover back on.

 

Over-Clocking makes no difference on anything. I had it up to close to 5GHz but since it adds nothing I've since disabled it. I'll save it for if/when needed. Also going to keep the buffer at 64. I'm learning to live with the extra 1.5ms latency .

 

[ElmerJFudd]

Hey, are you using motherboard audio with ASIO4ALL?

Or something better. There are some PCI cards or even USB3 or Thunderbolt interfaces that can potentially get you down to 32ms maybe.

 

Also, test in a host other than reaper or in standalone versions of the plugins and see if any better.

 

 

[Markyboard]

Hey, are you using motherboard audio with ASIO4ALL?

Or something better. There are some PCI cards or even USB3 or Thunderbolt interfaces that can potentially get you down to 32ms maybe.

 

Also, test in a host other than reaper or in standalone versions of the plugins and see if any better.

 

Thanks EJF. I use a RME PCIe audio card with ASIO. And yes - stand-alone apps work better - but not perfect at buffer size = 32. I'm in (or was) in test mode so trying to stress the various VSTs to see where the breaking point is. I'd like to hear from someone that has the soft synths I listed above that can overcome any audio pops at 32 samples (btw it's not ms ).

 

 

[Markay]

I'm finally done - YAY! Put my previously problematic soft-synths/VSTs to the test - here are some performance details for anyone interested:

 

Accoustic Samples Wurlie plug (UVI/Reaper) : previously giving occasional audio pops at buffer size of 32 and 64. Now absolutely nothing at 64 but still occasional pop at 32 if I pedal a whole lot of notes.

 

This build is near dead silent with nothing else in the room masking the noise. And I have yet to put the cover back on.

 

Over-Clocking makes no difference on anything. I had it up to close to 5GHz but since it adds nothing I've since disabled it. I'll save it for if/when needed. Also going to keep the buffer at 64. I'm learning to live with the extra 1.5ms latency .

 

Well acccepting that the following is not of much use to the OP I matched the settings to how my set up compared.

 

I loaded up an existing large concert in MainStage which is set to 44.1 kHz sample rate and reduced the I/O buffer to 32 samples. 3.5ms output latency MS tells me.

 

Loaded AS Wurlie and played block chords both hands pedalling sustain on and off while playing. No pops.

 

At 16 samples I get some muted pops. MS says 2.8ms output latency

 

No pops at 16 samples with Pianoteq .

 

Safety buffer off.

 

However the CPU meter said the 4 core i7 was into the yellow zone at both 32 and 16 samples.

 

Focusrite Scarlett Solo Gen 2 audio interface, core audio driver.

 

My conclusions are the hardware is up to running glitch free at 32 samples, the issues come down to how efficiently the VI, host or DAW, OS and audio drivers work together.

 

[Markyboard]

I didn't know about 16 samples as none of my hardware allows that setting. But I'm wondering if your 32 samples are equivalent to mine as Reaper shows an input/output latency of 0.7/1.5ms when set to 44.1K and 32 samples. At 64 samples it shows 1.5/2.2 ms.

 

Either way I do agree with your conclusions. And in case anyone thinks I'm concerned about a couple of ms no need. This is strictly an academic exercise for my understanding.

 

 

Edit: Correction to my Acoustic Samples Wurlie results: The audio pops are not related to the pedal use - they appear to be random but infrequent

[Markay]

I think the buffer samples are correct and the latency values are guesstimates.

 

The RTL and lower output values shown by MainStage, and Logic I suspect, are much lower than the RTL values calculated by Tafkat over at GS for the best RME interface.

 

To use an Aussie colloquialism I would tell Reaper it is dreaming with those latency estimates.

 

Either way I cannot feel or notice latency at 128 samples so it is all academic.

[Markyboard]

No doubt Mark. Btw I believe I run much faster then my timer indicates. I just have a very large sample buffer. :cofee:

[Theo Verelst]

I rarely use windo's for anything but system maintenance, an occasional example program or parallel work while it's on for recording sat or cable TV, so it's been a long time I extensively used sequencers and other music software on it. There's an issue with the whole music software that has a slight but non-zero congruence with the general properties of the windows or mac OS: how does the software keep time (stamps) and organize the audio workload with respect to buffers.

 

I'm very drawn to the idea I used almost daily on Linux where audio buffers are an exact given, and preferably correctly time-stamped input is computed (in a parallel sense, probably) and inserted into output buffers correctly, i.e. with fixed delay.

 

Even though explicit size audio buffers with strict latency control might impose some considerably more delay's in the overall computation organization in the audio programs, the constancy of the buffer length and latency between key press and audio output is worth it, usually. Nothing as annoying, especially when so much attention goes to picking the right computer architecture, as "feeling" the software work fro complicated chords, or to have their virtual memory management mess up with certain types of music performance, instead of just delivering enough sample bandwidth (should be super easy) and making all the filters and virtual circuitry just work on time and with rock solid fixed delay.

 

Lately I've even become obsessed with making fast computers (as it so happens a I39-30k with a P9X79 mobo) produce audio paths that aren't even horrible to play at very large latency (though I don't particularly enjoy that even up to .1 seconds(!)) because the accuracy of the signal processing adds something, including sub-sample accuracy, that's at times more worth than zero latency.

 

Zero latency at the output of the DAC is hard. Most computer DACs, will, rightfully, have a form of filtering built in that exhibits a little delay. That's necessary because of the mathematics that underlie sampling theory, in some sense, therefor not really circumventable. So even though a single sample delay should be easy with CPU clocks in the Gigaherz range (1/44100 ~~ 0.022ms, 1/3GHz ~~ .00000033ms), there might be a part of a millisecond delay built in the DAC itself, apart from the signal path and busses in between the CPU and the DAC buffers.

 

I know from building me 1-sample (!) latency hardware DSP synth that it's good fun to have a DAC and digital signal path with fixed super-low delay, so I'd prefer it at a computer as well, but unfortunately the computations even on a modern CPU are usually done in a pipe-lined parallel fashion inside the CPU, which requires some buffer length to become efficient. And then there's the kernel that deals with the various programs and interrupt computation threads which isn't all too efficient in many cases.

 

T.

[Markyboard]

It would be interesting to compare performance under Linux. I've never gone there before. While Reaper has a Linux version I don't believe most of these plugins do.

[burningbusch]

My build is done and has been up and running for several weeks. I added a second M.2 NVMe (1TB) for samples and use another SATA SSD for additional sample libraries. Man, this thing is awesome. My Geekbench multicore scores are about 24,000 no overclocking.

 

Of the three DAWs I have installed, Pro Tools 2018, Studio One and Ableton Live 10, Pro Tools is noticeably more efficient. The same plugins run with less CPU impact and they never seem to glitch. It doesn't seem to have those occasional 100% peaks, CPU usage is smooth. I know they were one of the last to move to 64-bits and had to rewrite everything due to their propriety hardware. Whatever they did, it's working.

 

These 6-core i7s are definitely in the sweet spot for price/performance. They run about $55/core vs. $100/core for i9s and, I believe, $200 for equivalent Xeons.

 

Busch.

[Markyboard]

Cool Bill. You didn't state what sample buffer size you're running at but I assume your more than happy with the performance. If you do by chance have the Scarbee Classic 88 plug I'd appreciate you seeing if you have any pedal pops at the lowest size sample buffer.