You are here

New PC time. Why shouldn't I go rackmount?

For anything relating to music-making on Windows computers, with lots of FAQs. Moderated by Martin Walker.

Re: New PC time. Why shouldn't I go rackmount?

Postby AlasdairEaston » Mon Nov 18, 2019 10:49 pm

That's great info Pete. Thanks for taking the time to cover all those details. In answer to your question, I'm usually running the buffer at about 192.

Cheers,
Ali.
AlasdairEaston
Poster
Posts: 36
Joined: Fri Feb 10, 2017 2:17 pm

Re: New PC time. Why shouldn't I go rackmount?

Postby Pete Kaine » Tue Nov 19, 2019 10:52 am

Yeah, there shouldn't be any real reason that you can't hit the 64 buffer on that interface other than running out of overhead, the figure of 10ms also sounds in the right ballpark for that buffer setting.

The laptop chip your running, I only mentioned the multicore score and I guess the single core should be noted here too. The QM chip you have pulls about 1800, whereas the newer chips tend to start around 4000, so we're talking about a load more overhead per thread.

The biggest trade off with laptops is that mobile chips tend to have low baseclocks, in this instance the 3720QM runs with 2.6 base clocks and 3.6 single core turbo.

Due to fact that it's rather complicated to split processing chains and tends to introduce lag which wastes resources, sequencers (as far as I'm aware, all of them) will aim to process whole channel chains on the same thread. Then you have to factor in routing will cause dependencies, so for instance side-chains will cause both channels to be linked and processed on the same thread, at this point you should be able to see how it's possible to end up with a number of complicated chains all being processed by the same core whilst the rest might be sat around doing largely nothing.

I tend to find that the slowest core will have an impact in the overall real world performance, so I tend to favour the solution that has the highest. With desktops starting in the mid 3GHz where you current chip finishes and the generational IPC gains (increase in instructions per cycle), I agree that you should be able to get the increase in responsiveness through buffer reduction that you're looking for.
User avatar
Pete Kaine
Frequent Poster (Level2)
Posts: 3178
Joined: Thu Jul 10, 2003 12:00 am
Location: Manchester
Kit to fuel your G.A.S - https://www.scan.co.uk/shop/pro-audio

Re: New PC time. Why shouldn't I go rackmount?

Postby Mixedup » Tue Nov 19, 2019 1:27 pm

I've had a couple of rackmount machines. They've been fine. But they don't cool as well as towers (heat rises), which might be an issue for overclocks etc?

I've just last week got a new machine — I opted for a tower (from Scan, surprise, surprise!). There's little difference really. Arguably the tower is more portable because it's smaller... just as easy to have a flightcase built to fit. And it has panels that can be opened on all four sides which makes servicing/installing things easier.

The only benefit of a rack case to me would be that I have rackspace available to accommodate it :headbang:
User avatar
Mixedup
Jedi Poster
Posts: 4411
Joined: Wed Sep 03, 2003 12:00 am
Location: Cambridgeshire, UK

Re: New PC time. Why shouldn't I go rackmount?

Postby Pete Kaine » Wed Nov 20, 2019 12:29 pm

Mixedup wrote:I've had a couple of rackmount machines. They've been fine. But they don't cool as well as towers (heat rises), which might be an issue for overclocks etc?

The older models with larger cases, you could get away with it, but I don't allow overclocks on our current racked machines for this reason.

Admittedly I had less concern when the CPU's were topping out at 95W TDP, rather than the 160w+ bursts we see on the current generation models.
User avatar
Pete Kaine
Frequent Poster (Level2)
Posts: 3178
Joined: Thu Jul 10, 2003 12:00 am
Location: Manchester
Kit to fuel your G.A.S - https://www.scan.co.uk/shop/pro-audio

Re: New PC time. Why shouldn't I go rackmount?

Postby AlasdairEaston » Wed Nov 20, 2019 7:16 pm

Pete Kaine wrote:Yeah, there shouldn't be any real reason that you can't hit the 64 buffer on that interface other than running out of overhead, the figure of 10ms also sounds in the right ballpark for that buffer setting.

The laptop chip your running, I only mentioned the multicore score and I guess the single core should be noted here too. The QM chip you have pulls about 1800, whereas the newer chips tend to start around 4000, so we're talking about a load more overhead per thread.

The biggest trade off with laptops is that mobile chips tend to have low baseclocks, in this instance the 3720QM runs with 2.6 base clocks and 3.6 single core turbo.

Due to fact that it's rather complicated to split processing chains and tends to introduce lag which wastes resources, sequencers (as far as I'm aware, all of them) will aim to process whole channel chains on the same thread. Then you have to factor in routing will cause dependencies, so for instance side-chains will cause both channels to be linked and processed on the same thread, at this point you should be able to see how it's possible to end up with a number of complicated chains all being processed by the same core whilst the rest might be sat around doing largely nothing.

I tend to find that the slowest core will have an impact in the overall real world performance, so I tend to favour the solution that has the highest. With desktops starting in the mid 3GHz where you current chip finishes and the generational IPC gains (increase in instructions per cycle), I agree that you should be able to get the increase in responsiveness through buffer reduction that you're looking for.


Really interesting, thanks Pete, especially the bit about workload sharing between cores and why good single core performance is key. And if I can get down to 64 on the buffer as you suggest then that would be fantastic. I'm going through your range now experimenting with the configuration options. Great website BTW.

Cheers,
Ali.
AlasdairEaston
Poster
Posts: 36
Joined: Fri Feb 10, 2017 2:17 pm

Previous