Going for a new computer - please, some help will be cool

[ronnie]

Cores, good. Threads, bad. DSP, best.

[tlaskows]

That's the thing. How is it that DSPs are so much faster than CPUs? I mean the current CPUs have crazy amount of mflops/s but these old Sharks are still rocking. I can run 7 instanced of the Minimax of 6 voices each on 45 DSPs. Can't do that with Diva. They sound just as good. Sonic Core developers are the best. I know that DSPs have dedicated instructions for audio/video, but I didn't think it would make this much of a difference. Maybe I just need a new chip

-Tom

[garyb]

CPUs have a lot of jobs to do....

[Nestor]

They connect to your host in a completely different way. CPUs have been made for many different general purpose jobs, Pulsars have been made to work only for audio, this is the difference, all its raw power is directed toward one single object, rather and divided into many.

[jksuperstar]

The theoretical CPU mflops/sec is only theoretical...it only accounts for "do math". To support real calculations, the CPU is a load-store architecture. Load data, do math, store data. You can see how fast those ghz can be burned during load & stores.

The DSP has everything designed around Do Math. So there isn't load-math-store, only do math, do math, do math.

It's like comparing a single lane bridge at 100mph vs a 10 lane bridge at 20mph. Which gets more cars across faster?

[tlaskows]

I shall choose the DSP to do the hard work

-Tom

[Sounddesigner]

I can run 7 instanced of the Minimax of 6 voices each on 45 DSPs. Can't do that with Diva. They sound just as good. Sonic Core developers are the best.

-Tom

Yep, comparing the number of Minimax's loaded on one's dsp's (or another great sounding SCOPE synth) verses Diva (or another one of the better Native synths) on Computer processor is a telltale to me and lets me know that dsp's are VERY powerfull in a relative sense, especially since SCOPE allows you to run many instances of great sounding synths at a latency unachievable by Native. The more you lower the Native buffer the less the load it can handle to the point where nothing or next to nothing can be done. Plus the SCOPE synths sound better IMO and when Native achieves better sound the plugins tend to get even hungrier in a insane way. This type benchmark comparison might not be that scientific of a test technically speaking far as measuring the power of computers vs dsp's but it is VERY scientific when it comes to personal needs, wich is what matters most.

SCOPE allows for a nice amount of high-quality synths to be ran at VERY low-latency. It's still very powerfull.

[stonberg]

How is it that DSPs are so much faster than CPUs?

DSPs are designed to do very specific tasks very quickly; their hardware performs these tasks very efficiently and perhaps more importantly, perform a series of parallelised tasks in just one execution cycle.

The tasks they are designed to do are those mathematical operations which are used in digital signal processing; typically multiplications and additions.

Here's an algorithm for a simple bessel filter, in fact one we used in the laser diode spectroscopy project I worked on for the last couple of years (we ran a few in series to get the required number of poles):

O(n) = a * O(n-2) + b * O(n-1) + c * I(n)

This reads as follows:

The nth output value is equal to the n-2th output value times a constant, plus the n-1th output value times a constant, plus the nth input value times a constant. i.e. the previous output values are fed-back into the calculation.

As you can see, there's nothing magic going on other than addition and multiplication. DSPs are designed to perform these calculations with lightning speed and with minimal execution cycles.

As to whether a DSP chip is faster than a CPU, that's a nonsense really, it's the same as asking this question:

What is faster, a boat or a car?

The answer ofc depends on whether they are running on land or in water; the car is going to beat the boat on land, but the boat will be faster in water.

[tlaskows]

Yes, I know some filter theory. I would have thought that Intel (or AMD) have added DSP like MAC instructions. Maybe it doesn't make sense to have them on a desktop CPU, I don't know...

-Tom

[stonberg]

Intel started adding SIMD type instructions way back with the MMX instruction set (1999?), so I guess the lines between what is a general purpose processor and what is a digital signal processor are somewhat blurred these days.

Some chips we've worked with in the past have both a general purpose CPU and a DSP on the same silicon, whilst others (e.g. some of the Cortex M-series) have DSP instruction extensions.

[garyb]

a CPU CAN do what a DSP does, more or less, but not while running the OS and Sequencer and various other processes...

[tlaskows]

Yes, I do believe you, Mr. Gary. I just checked Activity Monitor on my MacBook... oh boy. Over a 1000 threads and 250 processes. The CPU is very busy doing other stuff. I don't even have that much open!

-Tom

[garyb]

that's part of the fallacy that says "CPUs are getting so powerful, DSPs are passe". the truth is that sloppy coding and extra tasks are the result of the extra power. little of the extra power is left for the audio apps...

[Nestor]

Well..., the prove of this it is in the fact that I am getting an old Pulsar 1 as an amazing addition to my existing system with already a Pulsar 1 on it, and despite the many years that has passed away from the creation of these DSP based cards, they still are and will be shining in a new system with all sort of new technology. It is amazing! That proves DSP is the best way to go for a DAW, and most important, trouble free in most cases.

[Nestor]

Reading about new technology and all, I'm a bit confused about which one of those two MOBO to take, which one would be better for my multimedia job? I think I don't need many of the features of the Z97, futuristic stuff I will definetely not be using, and the Z87 seems to be more suited for Pulsar, and more stable perhaps, don't really know. Anyway, here are the contenders:

Asus Z97_K

or

Asus Z87-C

Based in your experience guys, which one would you take and why? Cheers!

[garyb]

Z87 looks better somehow. it's probably not a big deal either way, i just like the extra PCI slot. with 13 usb ports, it won't be a big deal when you have to disable a few usb controllers...

people have had good luck with MSI lately, too. actually, the two or three 1150 boards that did not work were ASUS. i don't think there's meaning there. traditionally ASUS has been one of the best choices for Scope cards, so i think there are just more ASUS boards being used. that means that the actual ratio of good boards to bad(defective) boards is the same as any other maker. to the best of my knowledge, the defective boards were warranted and exchanged and the replacements all worked as they should.

i'm not really sure why i feel compelled to write all this extra. i guess it's a left-handed assurance that if the boards don't work, they're just defective, it's not a compatibility issue. this is unlike the earlier socket 1155 boards. if one of those DID work, it was defective...

[tlaskows]

Using a z87-k here with no issues. If you don't need the extra features and want to save a bit of money, it's the way to go. Just make sure you do a MV test. You should get around 10.

-Tom

[Nestor]

I have the same feeling here… Z87 looks more suitable to me to… and particularly the Z87-K, it seems to be a better build than the Z97.

I will definitely check the MSI too.

It is true that the quality of MSI has raised, this is also my impression. Their latest boards seem very, very good.

I’ll check them out immediately, as I have to decide, and will report you here. Thank you for your friendship!

[Nestor]

Well guys..., I investigated many MOBOs, I read many, many reviews and came finally to the conclusion that I prefer to get the Azus Z87-C. I really hope to have been wise when taking this decision.

[Nestor]

Here it is the manual, for all the details, it is a great board it seems:

http://dlcdnet.asus.com/pub/ASUS/mb/LGA ... _Z87-C.pdf