Converting typical benchmarks to Cantabile use

You are describing my Sweetwater Sound “Creation Station CS400v7” exactly. Intel Core i7-10700K CPU @ 3.80GHz and shitty Intel video! :slight_smile:

It was built to be optimized for audio and be whisper quiet with a 4u rack mount case. I rarely hear any fans. It seems most video cards have fans on them these days if they are any good, and I don’t want to add that noise to my system - plus Nvidia used to be treacherous for audio work, introducing terrible latency glitches. I used to have to run LatencyMon all the time trying to optimize against them. Never bothered with that now.

Unfortunately, it seems Sweetwater is out of those units at the moment, which likely means they are about to release a newer version.

Terry

1 Like

BTW, LatencyMon is a bust on the latest i9 anyway. The bar graph page is showing all kinds of wonky results with massive latencies while the written report page is showing several orders of magnitude lower latencies. All while zero glitching running complex Reaper projects at 2ms latency. Something to be aware of.

2 Likes

Interesting indeed! Thanks for that news!

OK, I finally got everything working, and have some comparative data between Kaby Lake (Intel G7) and Coffee Lake (G8) systems with near identical surrounding pieces.

Sidebar: The first thing that I learned was that if a Dell PC (G7 or G8) can’t tell what power supply is attached, it defaults to 0.79GHz base frequency and no Turbo! The latest power supply I received reports as ‘unknown’ 95-98% of the time, leaving me scratching my head when results were not repeatable. While the i5-8600 outshines the original i5-8600T in my G8, I probably would have lived with it’s roughly i5-7600 performance, if things made sense sooner! :face_with_symbols_over_mouth:


Now that I’ve got the hardware sorted out, here’s what I found by running Cantabile and a range of VSTs on an i5-7600 vs. an i5-8600 in two very similar PCs:

  • Both PCs are Dell OptiPlex micro PCs, with DDR4 memory and the same complement of USB ports, memory slots, etc. Both have two 16GB SODIMMs, so they run dual-channel.
  • Both CPUs are 14mm Intel Core processors, although the newer one has minor improvements. The L1, L@, and L3 cache per core are the same.
  • Both hard drives were replaced with 1TB WD 570 NVMe SSDs.
  • Both use SoftPerfect RAMdrive to create a very fast 12GB ‘work’ drive for the VSTs, media files, and support data.
  • Both use a 3rd-gen Focusrite 4x4 audio interfaces over USB using ASIO.
  • Both are running ‘headless’ through Windows Remote Desktop to the laptop that runs my DAW; and rtpMIDI to pass keyboard data to the VSTs.

Now, for the differences:

  • The '7600 has 4 cores @ 3.5GHz (base clock)
  • The '8600 has 6 cores @ 3.1GHz
  • The '7600 has 2400MHz RAM, while the '8600 has 2666MHz.
  • The BIOS has more tweaks in the G8 PC. I tried to disable inconsistencies, but who knows?
  • The '8600 idles 2-3 degrees cooler, with the caveat that when I changed the CPU I used fresh, quality thermal paste; while the '7600 probably has the original 5-year-old thermal solution.

I used a variety of tests, including a couple of songs that I use live that I had to trim back on the '7600 to get them to run at < 10ms latency. Generally speaking, the G8 outperformed the G7, but by about the expected MIPS/FLOPS performance boost, which given the same basic CPU architecture should be –

(6 cores * 3.1GHz) / (4 cores * 3.5GHz) = 18.6 / 14 = 1.33 (133%)

And yeah, I saw a 25-30% improvement in either CPU Load/Time Load, or ASIO latency before glitching. In less technical terms, I found I could run 1-2 more VSTs on the 6-core machine; or I could drop the buffering to achieve 2-3mS better latency.

Now some of you are going to say, “Well of course it can run more VSTs on more cores,” but I also learned that the results aren’t that simple. The 4-core PC ran nearly identically with the Audio Engine set to Auto (4), 3 Cores, or 4 Cores. In fact, the 2-core performance was only off by 5-10% in most tests – although the Time Load often jumped above 100% a bit. This tells me that Cantabile does a really great job of juggling tasks to ‘play the hand it was dealt,’ so to speak.

Bottom line, a 65W CPU of either generation is a feasible platform for replacing something like a Muse Receptor or high-end workstation for live performance. Either PC works for my current needs; and the 6-core provides a little spare power for the future. In the end, though, I’ve decided to keep both PCs in my rack. The 4-core will handle (virtual) guitars, bass synth, and media files; as well as interface to my two keyboards over USB, pass the data on to the other PC, and communicate back to the DAW as necessary. The 6-core PC will handle as many synth and organ VSTs and it can, and seems OK running 2-3 of my ‘heavy hitters’ simultaneously.

This is arguably no better than trading in for a new, high power rig, except:

  • The whole setup cost me less than $900USD (reconditioned), even counting the second USB audio interface. I doubt I could touch anything with 10 ‘power’ cores for less than $1,500USD, if I stick with a top-tier PC brand (Dell, Acer, HP, etc.)
  • Each PC only dissipates 25-50W during performance, meaning an air-cooled thermal solution is practical.
  • Running multiple ASIO audio interfaces is straightforward (vs. almost impossible in a 1-PC environment), meaning I can stick with lower-cost audio solutions. I actually have 10 physical outputs going to the mixer!
  • Everything, including USB audio and power supplies, fits in 2U of rack space.

Coordinating two instances of Cantabile is straightforward in my setup, because I’ve always used my DAW as the master, so one slave running over Ethernet isn’t much different from two. (They are near-clones, just running different versions of each song.)

Regards,
-BW

3 Likes