I’ve been keeping a secret… I have been running this whole operation from a 2012 MacBookPro since founding A/Vian Eco in 2017. I held on to this machine for dear life for it’s robust nature and upgradable design. The whole case is a glorious aluminum heat sink making it near bear-proof. It well outlasted it’s plastic-bodied counterparts of the same vintage.
Unlike most MacBook Pro Unibodies, Bear-proof’s case came with a removable back panel making it a snap to upgrade to high-capacity solid state internal drive. Our first jump to a solid state was a huge leap forward in processing speed. If you’re looking for an affordable and accessible upgrade for your bioacoustic analysis machine get yourself a high-capacity internal SSD, stat! Also important for large data download this vintage of MacBook Pro was of the era of “the more ports the better”, so no throatled transfer speeds or dongle hell for me!
Bearproof, the 2012 MacBook Pro, still purrs at it’s peak capacity however, in the unspeakable year of 2020 Apple broke my heart. They stopped supporting iOS upgrades for “vintage” MacBook Pros older than mid-2013. I knew it was time to build a desktop workhorse that could crush huge batch processes for high data capacities (terabytes), quickly index high numbers of audio files, and handle machine learning when it inevitably becomes a standard practice in bioacoustics. The new PC-build is a processing beast and I like to call her Fluffy.
Transparency:
For the sake of transparency, while building Fluffy the PC we did have a budget. Do not hold up these components as the “perfect analysis machine” as all analysts and workflows will have their own specific needs. We chose to find a balance between price and performance for this build. The rationales are suggested considerations for purchasing new or optimizing an existing set up for acoustic analysis. There are no hard rules.
For the interest of my students, the IT professionals on my clients’ teams, and any other interested techno-naughts, below I describe:
- the components of the new A/Vian Eco Bioacoustic Analysis PC.
- the rationale behind each selected component from a bioacoustics perspective.
- a challenging benchmark using KailedescopePro’s cluster analysis.
Monitors:
Analysis Primary Screen: Dell 27” 4K S2721QS IPS monitor–
Purchased: December 2020 for ~$430 from Dell
Rationale:
- As per the Rtings.com review, this dell monitor was one of the most affordable 4K-resolution, true-colour screens on the market. It is hard to beat the price for image clarity like this. The great image quality can improve visualizations of spectrograms helping with those hard-to-define, quiet, and far away vocalizations.
- The adjustability of the included stand is great for ergonomics during long analysis sessions.
- The screen real estate is an upgrade from my former 24” monitor and 13” MacPro screen configuration. Larger screens allow me to view greater stretches of time on each spectrogram, reducing the overall amount of time spent loading new spectrograms views.
- The IPS panels in both of my monitors allow for clear images at indirect viewing angles. This allows me to sit in-between the monitors and compare focal recording samples on my second screen to the current spectrogram view on the main screen with minimal neck craning.
Secondary productivity screen is the AOC I2367FH 24” HD (1080) IPS monitor (discontinued).
Purchased: 2017 for ~$220 from Memory Express.
Rationale:
- Despite the lack of 4k resolution and a cooler overall colour temperature, this has been a fantastic monitor. I originally used this as my analysis monitor, as the 24” screen allowed more screen real estate than the 13” MacBook Pro alone.
- The only complaint I have is the limited range of adjustment from the included stand. For improved ergonomics I’ve purchase a riser for this monitor.
Motherboard with built in sound card:
Asus TUF Gaming X570-Plus
Purchased: December 2020 for $225 from Memory Express
Rationale:
- This motherboard has built in isolation/shielding, stereo sound, and a reported Signal-to-noise (SNR) output of 108dB. It is possible to get a slightly higher SNR output with some intel motherboards or stand-alone audio cards but this was very good clarity at a reasonable price.
- Signal-to-noise is a measure of how loud the desired signal (sound) is by comparison to any artificial sound created by electrical circuits. High numbers here are good, but once north of 100dB it’s unlikely I will have the sound up high enough to hear the noise.
- If you’ve ever plugged a pair of headphones into a computer and immediately heard a hiss or static, that computer or specific 3.5mm jack likely has terrible SNR. From this motherboard there is none of that non-sense.
- This model is also Ryzen-ready allowing me to purchase the desired processor (see below).
- Of note from use of this motherboard:
- The amplitude range is huge. Low “volume” settings are quite sufficient for most studio recorded music. This means there is lots of room to increase amplitude for listening on any “field-oops” recordings with too-low gain settings or for weather or cow-spit dampened microphone diaphragms.
- There is also more than sufficient pre-amplification to the 3.5mm jack to share interesting vocalizations with other colleagues through desktop speakers.
*note I pulled a noob here and ordered the version without in-built wifi. I recommend purchasing this one instead. Wifi cards are relatively cheap (here’s the one I bought), but all-in-one removes some building steps.
Processor
AMD Ryzen 7 3700X 8-core
Purchased: December 2020 for $450 from NewEgg.
Rationale:
- At the time of purchase this processor was one of the best price-to-power models on the market. 8 cores, and 16-threads speeds up batch processing of acoustic data significantly (see the benchmark below).
- This is also one of the least expensive models on the market that will allow processing of machine learning models. There are rumblings and some organizations actively working on these types of tools for sound data processing, A/Vian Eco will be ready when they are.
Graphics card
MSI Geforce GTX 1650- dual fan
Purchased: December 2020 for $215 from NewEgg
Rationale:
- To be frank, it was extremely tough even to get my hands on a graphics card with pandemic and crypto-mining related silicon and card shortages. I was lucky to get this reasonably priced model.
- The programs currently available for acoustic analysis do not put a ton of strain on a graphics card, making it less likely to be a pinch point for data processing. Spectrograms generation is a pittance by comparison to VR gaming or heavy 3D mapping models.
- This card handles 4K video on the new monitor very well.
- The one feature I did hold out for was a dual-fan layout, as it is important for ambient sound. If the card is ever taxed the dual-fan layout spins up more slowly than single-fan while still adequately cooling the card. This coupled with the case’s sound-proofing making the overall sound output very minimal. A quiet computer means reduced ambient sound while listening.
RAM
Corsair Vengence LPX- 16GB (2x 8GB)
Purchased: December 2020 for $104 from NewEgg
Rationale:
- Not much to say here, 16GB is more than sufficient for our current processing needs. Though this is an upgrade from Bearproof, most of the power and speed improvement comes from our SSD and processor choices.
- We skipped the LEDs to reduce power consumption. Finding PC build components without rainbow LEDs for responsible energy use is unnecessarily challenging.
Case
Fractal Define 7 Compact ATX
Purchased: December 2020 for $140 from Memory Express
Rationale:
- This case has the appropriate space for the components purchase, while remaining compact.
- There is sound-proofing on all sides except the glass panel, really reducing the ambient room sound for listening. Though the fans are rarely taxed to full-spin, the sound of the forced-air furnace in our office is louder than the fans in this case.
- Of note from use of this case:
- It is not possible to attach the USB-C port on the top-front of the case to the motherboard, however there is a USB-C port on the back.
- The 3.5mm jack on the top-front of case doesn’t add any noticeable electronic noise or show any amplitude losses from the lovely clear sounds of the motherboard. This makes it a nice quick-connect option for analysis or wired headphones with short leads.
Harddrive
Samsung 860 EVO 2TB SATA 2.5″ Internal Solid State Drive (SSD)
Purchased: December 2019 for $389 from Memory Express
- Moving to solid state on the Mac and subsequently the PC has really improved the speed to load each time segment of spectrogram and the time to complete batch processes. This is a good investment to improve the speed of your overall analysis process.
- I highly recommend going with a higher-capacity drive so larger projects can be reviewed in a single cluster or batch process without breaking them into small chunks. Because of the way computers access a drive, higher capacity also tend to have longer life-spans.
- The case and motherboard combination allow for 2 SSD and 2 NVMe drives. High capacity NVMe might be our next step as flash memory prices keep falling.
Other Peripherals and Power:
We purchased a Corsair RM650 Power Supply for this computer. For a very brief reference we use the following peripherals currently:
Headphones: Audiotechnica ATH-MSR7’s (2016),
Mouse: Logitech MX vertical mouse (2021),
Keyboard: Logitech k400r (2014),
Speakers: Logitech Z207 Computer Speakers(2019),
Webcam: Logitech C920 HD Pro (2021),
External Long-term Storage Drives: Multiple 8TB Seagate Expansion Drives (2018)
If there is interest in the future, we could do a breakdown of specific peripherals. Perhaps headphones?
Head-to-head Benchmark:
| Bear-Proof* | Fluffy | |
| Device Type | MacBook Pro 2012 | Custom PC 2021 |
| Motherboard | Inbuilt Apple Board | Asus TUF Gaming X570-Plus |
| Processor | Intel i5 “Ivy Bridge”, 2-core, 2.5GHz | AMD Ryzen 7, 3700X, 8-core |
| Graphics Card | Intel HD Graphics 4000 graphics processor | MSI GeForce GTX 1650 (dual fan) |
| Case | Unibody Aluminum case | Fractal Define 7 Compact ATX |
| Harddrive | 500GB Samsung 860 evo SSD (upgrade) | 2TB Samsung 860 evo SSD |
| RAM | 8GB (2x4GB) 1600 MHz DDR3 | CORSAIR Vengeance- 16GB (2 x 8GB) 3200 MHz DDR4 |
Our test was a mix of short (60s) and long (1hr) acoustic recording files compiled into a folder with subfolders for a total of 101GB of acoustic data. These recordings came from a mix of device types and sampling frequencies (SM4 at 44.1kHz and AudioMoth at 32kHz), but were all the same file type (>118000 16bit-WAV files).
The test data folder and it’s subfolders were run through KailedescopePro’s Cluster analysis. The KPro Cluster settings were realistic to pull detections (144873 total) on a variety of species within the audible range (0-12000kHz). KPro was directed to use maximum compute resources for both computers.
*We gave Bearproof the Mac the best advantage we could by placing it on in fan-cooling stand. This is typically how we’ve run clusters in the past.
BearProof completed the task in 57min…
Fluffy completed the task in 15min. Wow. Just wow. We’re flying now friends.
Thank you!
Setting aside tongue-in-cheek, we also wish to extend an extremely big thank you to our clients, colleagues, and professional network. If it weren’t for their support, 2020 would have been a different story for A/Vian Eco. We are extremely grateful and hope the purchase of equipment will help us serve our clients and community better.
Another enormous thank-you goes out to my partner for freely lending his programmer-eye and valuable tech-pinions to the process.