Software Signal Flow

[dawman]

I know nothing about how instructions are sent and permissions given in a PC.

I need to find out a few things.

1) The OS loads into RAM. After this happens does the storage device play a role in performance after boot up, or do all of the cues needed then come from the CPU...?

2) Samples are loaded into RAM. Target buffers are created so once an instruction is given the low latency of RAM means the next order or performance comes from streaming from the storage device.
Again does this process need access to the OS data on a storage device, or is the CPU nowa the main factor in performance.

NVMe SSDs have become really fast due to the new protocol, but older AHCI models seem to have better sustained reads. Their peak bandwidth benchmarks help sell a device, but the only real world advantage I see is in multiple low latency I/O testing.

Thanks.

[dawman]

So it's safe to assume that folks here who have spent thousands on software and hardware don't know wtf they're doing...?

[borg]

great, isn't it? and to top things off... I sport Mac!

[w_ellis]

I think the problem is the answer is "it depends", which probably doesn't help you very much

You'll probably need to re-frame your questions to be more specific to get useful answers.

Being more generic, yes, applications load into RAM and at that point typically they themselves won't need to go back to disk. However, that's not always the case. For example games do stream load themselves as you progress.

Most major sampler programs these days stream samples from disk into RAM with clever algorithms that try to predict what you're likely to want to hear next. Gigasampler started this back in the olden days, but has been emulated by the other mainstream applications. This process is independent of the application logic itself, which may well have been fully loaded into RAM from what you are probably considering the OS drive (if that's where the application was installed).

Problem is that you can set up a PC in all sorts of different ways, by installing things in different places and applications can be implemented in all sorts of different ways. My advice would be only to worry about it if you have a problem and then ask about that specific problem.

[dawman]

Thank you very much chaps.

No specific problem, just a future proof concept as I start using PLAY for live real/time work.

Speeds of SSDs are insanely quick but I have already found that there's a threshold around 750MBps that seems to start having diminishing returns.
More than enough to do what I like doing, and basically is what PCI-e 2X M.2 devices are capable of.

So I am trying to get the lowest latency devices I can that can bench well in high I/O testing.
Seems to me that NVMe which is going to be brand new so rather costly would be a great OS+Apps device, then stay with PCI-e 2X M.2s like the Plextor M6e's which are cheap now, and sure to be even cheaper after August 5th when NVMe boards and devices start hitting the shelves.

Usually prefer to wait and see, but I already know I will be using many MIDI tracks triggering large instruments and sampled content in 2016.
Must prepare for that day way in advance and trying this new design out for 9 months will commence in 3 weeks.

You saved me lots of frustrating time trying to get developers of Kontakt or PLAY or even Omnisphere 2 to explain things.
They treat their apps as "Top Secret" and seem to clam up when such questions are asked.

Knew I could get the skinny from guys at the Z.
Never fails.

Cheers

[Neil B]

Only a quickie Jimmy that may help.
I was watching a video by John "Skippy" Lehmkuhl on you tube the other day. Search for Plugin Guru.
It was one of his new videos on Omni 2. Basically he had released a new bank Airwave 1 for Omni 2 and had to re-release with some lighter multis. Therre had been feedback that some of the multis were CPU hogs. Interestingly the problem was that ANY multitimbral synth will only use 1 core, whatever processor you've got. Now, I'm not into this technical stuff but you may find the video useful.
The video is Airwave V1 Bonus Vid Multis and Kiks
Neil

[dawman]

Thanks Neil B. good to see you again.
I bought those yesterday for 35 bucks and there are some great BPMs, Leads and Multi's.
Today I am checking out another Zebra2 guy who folks rave about.
But I am into the EDM stuff more now since I make leads and pads a certain way, mostly to emulate my beloved analog hardware.
If you like EDM though a Japanese guy calls his stuff Pokémon and he has mad skills.

Thanks for the heads up, great purchase.

FWIW I run my CPUs @ 4.2GHz since it's a 1U design.
These parts below have to keep things @ 53 C.
Other folks like scientific water cooled nitrogen injected stuff, I am a meat and potatoes guy,
keep it simple and cheap ankyuverymush.

[Neil B]

Hey Jimmy
I'm about here most days, but being a "Scope-less" honorary member at the moment, I don't have a lot to contribute !!
If you're into Omni 2, check out the Toxic bank from Skippy too.
Not sure if you watched the vid but it states that, apparently, Kontakt only uses 1 core too.
So much for the versatility of multi-timbral synths eh?

I just can't get my head around you getting into EDM, but it's nice to share my second childhood with someone else
Neil

[stonberg]

1) The OS loads into RAM. After this happens does the storage device play a role in performance after boot up, or do all of the cues needed then come from the CPU...?

Start-up time of applications and reading files (e.g. opening a DAW project file, a word document, etc.) will be faster due to the improved i/o speed of the SSD compared with a spinning platter hard drive. Any streaming from the drive (samples for example) will also be faster.

Note that a lot of moving data around in a computer is done under the auspices of the DMA controller, so the CPU doesn't necessarily get involved much in that type of operation.

2) Samples are loaded into RAM. Target buffers are created so once an instruction is given the low latency of RAM means the next order or performance comes from streaming from the storage device.
Again does this process need access to the OS data on a storage device, or is the CPU now the main factor in performance.

FSB (Front Side Bus) AKA RAM speed is a factor here. So are the sizes of the various CPU caches, and the efficiency of the cache coherency algorithms used, *especially* when dealing with multiple CPU cores on the same piece of silicon; coherency needs to be maintained across all cores! This usually involves some very clever hardware and OS design.

The OS (drivers, kernel, session manager, shell etc.) will be in RAM, there's not much to do with the OS which needs to be accessed from disk once the OS is up and running as far as I know.

Obviously the CPU is a factor, but it's not all about the number of cores, hyperthreading, and GiggleHz clock speeds. Single-threaded performance is very important, perhaps more so than multi-threaded performance as that's where the real time savings are made when running multiple threads; once any parallel processing is finished with, all threads have to wait on a synchronisation point which will depend on the single-threaded (serial) performance.

NVMe SSDs have become really fast due to the new protocol, but older AHCI models seem to have better sustained reads. Their peak bandwidth benchmarks help sell a device, but the only real world advantage I see is in multiple low latency I/O testing.
Thanks.

Hmm, dunno about that. SSDs were getting too fast for the SATA or IDE buses, so shoving them on the PCIe bus was a step in the right direction IMO. I'm actually working on 6- and 12- gig HDD enclosure management software presently; it's all going over to ethernet, man! Srsly, each individual HDD will be running its own linux operating system presenting an OO API and they all sit on blindingly fast fiber optic cables networked by ethernet. Completely bonkers:

http://www.seagate.com/gb/en/products/e ... netic-hdd/

Anyway, just how much bandwidth do you need? SATA can shovel 6 gb/s about quite merrily; very roughly that's 3,040 stereo channels of 44.1K/24-bit audio

[dawman]

Well I guess I will know in a few more weeks if the hardware I bought makes a difference.

Went with all NVMe, CAS 8 DDR4 2400 and even plopped for 2 x SM951s + a new Intel 750 NVMe connected on the ASRock M.2 slot with the U.2 connector.
If I don't see a big difference I can add an M.2 to each of the 3 x 1U PCs I have and share the wasted money I suppose.

The Z170 and i7 6700K is the only thing I am waiting on right now.
They tell me delivered by August 9th.

Fingers crossed.