My best performing settings - Bios, system, etc. for Cubase

Steinberg Lounge
1

Hi there,

I’ve spent a few weeks with a new system and have found some surprising performance changes just by combining settings.

For reference, my tests were using the DAWBENCH RXC-EXT test in Cubase 6 and then Cubase 8. The test runs highly multibanded processing on a free Reaper plugin, stacking up the plugs until the computer can’t stream in realtime anymore.

I’ve not found a significant performance difference between Cubase 6 and 8 on the same system.

At 512 buffer on my RME FF800:
(more Plugins are better)

INITIAL SYSTEM
130 Plugins - (single core Xeon 2620v3 - before upgrade)
NEW CPUS
230 Plugins - (Dual core Xeon 2630v3 - after upgrade)
SOFTWARE & BIOS TWEAKS
246 Plugins - (Hyperthreading off, bios tweaks (disable runtime power mgmt, pcie performance mode, no c-states, disable usb3, Win7 High Performance Power mode although this would be disabled by the bios)
SOFTWARE & BIOS TWEAKS 2
271 Plugins - (Hyperthreading off, runtime power management ON, no power saving in bios, disable pcie performance mode, runtime power management ON, Steinberg Power scheme ON, Xeon turbo mode enabled, disabled USB3)

Other bios tweaks:
NUMA disabled
Intel Management interface (AMT) disabled - this polls the system every 5 seconds and can create glitches on certain driver versions (19.3 and above).

Conclusion

So I obtained a 17% performance increase just by setting up the computer more effectively. Many of the power tweaks I was making in the OS were not fully realised as the runtime power management in the BIOS was disabled.

The Turbo mode for the xeons allows an 8% overclock - clock speed of 2.4 to 2.6ghz. With this, the cores are all around 85-99% utilised in the task manager, although Cubase doesn’t fully optimise the usage of the cores, as it seems one processor is always 5-10% under-used, even when audio dropouts are occurring. This means 8 cores on my system are still not being maxed out.

Interestingly, it was this last set of tweaks that got Cubase 64-bit to perform the same as Cubase 32-bit (usually it was 20-30% slower).

Hope this helps someone

Back to writing music now!

2

Hmm, I got very different results from you with some of those settings on my i7-5930K system (running Windows 7 64-bit with 64-bit Cubase 7.5.2).

Disabling hyperthreading significantly hurt my DAWBench RXC scores, but disabling C-states and other power management CPU features had no positive effect; they were working as intended- dialing back on the power management when the system was running under heavy strain.

I’ve never worked with Xeons, but I’m curious why the results would be so different?

EDIT: Above scores at 44.1kHz with a 288-sample buffer on a Roland Duo-Capture EX.

3

That’s interesting!

Everything is at 48khz. At 44.1khz I get 303 plugins.

Are you running the RXC project, or the RXC-EXT project?

I ran the normal RXC and am getting about 1536 plugins on that version.

*The tweaks in terms of disabling power management are more related to stability rather than maximum processing power.

I’ve switched on Hyperthreading and it makes a negative performance difference.

4

cool blog btw

5

Interested how you are getting these results - SCAN could only get 262 from an i7-5960X…

6

Just the RXC project, not EXT, and I am running at stock speeds, no overclocking- so I’m not surprised by the general difference in numbers.

But I was wondering why disabling hyperthreading seemed to have opposite effects for us, and why disabling the CPU power management settings seemed to give you a boost but had zero effect for me.

7

Ah that explains alot! Thought you had some magic going on in there.

Well, I was reading about hyperthreading, it is beneficial in certain situations but creates a slight overhead. So if you’re nailing the CPU cores, adding that overhead isn’t going to be beneficial.

I found the ASIOGuard was only beneficial with HT ON, it made no difference with it off.