A possible new real-time DAW benchmark test

[noiseboyuk]

Hello all. I'm on the cusp of buying a new rig, and so I've been up to my elbows in DAW bench and Geekbench scores. What I came to think is that our World Of Audio lacks a meaningful test for real time / single core performance. I currently run a 4930k at stock speed of 3.4ghz. The disk peroformance - despite SSDs and NMVes - is woeful, but that's one for another thread. I'm also underwhelmed by it's processor performance, and it's nearly always the single core / real time I have issues with. One demanding patch of a synth and it's into the reds.

DAWbench as a suite of tests took a little while to get my head around, but when I did I could see both its advantages as drawbacks. Both the VI streaming and DSP tests don't take real time performance into account at all, the very thing I'm struggling with. The best gauge we have for this is the Geekbench 4 single core test. My 4930k is a lowly 3500, while a shiny new 8700k overclocked to 4.7ghz is around 6,000 - promising a 1.7x improvement on this metric. That would be lovely!

So I made a very simple Reaper test project, based on the DAWbench SGA 1156 test. Instead of 40 channels of plugins, there is just one, with 8 instances of the freeware SGA 1156 (and the handful of standard Dawbench audio tracks which are used to determine if the CPU limit is breached). The only difference is that on the effect itself, CPU is switched to high (for the regular DAWbench 1156 test, it is set to low). Cubase cannot play a single track of this when switched to x4 oversampling! So the test has oversampling restricted to x1. My rig scores a 5 on this Reaper test, so I was imagining an 8700k could hit all 8, and maybe even try a test on x2 oversampling. However, a friend over at The Sound Board forum only managed 7 instances on his 8700 clocked at 4.7ghz, which seemed disappointing to me - where did that 1.7x improvement go?

If anyone out there fancies running the test, I'd be very interested to hear what you get. A zip of the Reaper project is here - https://www.dropbox.com/s/oj2myvmne1xh3 ... e.zip?dl=0 . Note - Reaper tends to me much more efficient than Cubase, so being as you can run it in evaluation mode I suggest Reaper would give the most meaningful results. So far, all tests have been run at a buffer of 256.

[Guenon]

Very cool!

What sample rate is to be used here? 44.1 kHz, at a buffer of 256 samples?

If so, the result of my workstation is: 7

(and 8 in a pinch; starts to skip very noticeably with 8, but doesn't crackle, cut out or "break down")

i7-3770k @ 3.8 GHz, Asus P8Z77-V, integrated Intel HD 4000 graphics
RME HDSPe AIO
Windows 7 64 bit

High performance power mode with all sleep modes disabled, core parking disabled, unnecessary devices like wifi and additional usb ports disabled.

Btw it's SGA 1566, not 1156 ... Everyone keeps saying 1156 for some reason. Just so that nobody searches for the 1156 before trying the project file (like I did, hehe)

[Guenon]

5 at 2x oversampling. Can do 2 at 4x, plus an additional one with 1x. Upping that third one to 2x starts to skip.

[pljones]

Echo AudioFire4 (dedicated FireWire card, v5.8 ASIO drivers), 44.1 kHz, at a buffer of 256 samples
AMD 8350 at 4GHz stock, 32Gb RAM, AMD Radeon HD 7450 (1GB)
Win10Pro.1709 fully updated

1x oversampling

...

Two (yes, 2) instances!



(The 32Gb might give away that I generally trigger samples with as little post-processing as possible.)

[noiseboyuk]

Thanks you too, very interesting. Guenon, I actually had mine set at 48k, but I've just retested and same result with 44.1 (44.1 may have been fractionally better, the break ups still happened on 6 but they seemed to manifest slightly less severely). Good result you got there! I don't know of anyone getting more than 7, but early days of course.