- Some DL problems are intractable
- Complete Coffee Lake CPU list - Anandtech
- Jetson Nano
- Google Stadia - The Verge
- ESRGAN texture packs
- Playable raytraced Quake 2
- ARM Helium vector extensions
- GradientFlow - Training ImageNet in 1.5 Minutes on 512 GPUs
- Thinkpad X210
Episode 32 - It's a Proper Chunky Boy
CES is still a thing?
- Alienware Area 51 M
- LG Rollable TV - The Verge video
- Intel Keynote
- Ian Cutress tweet about 28 core high-clock Xeon pricing
- Gsync on freesync
- Microsoft 896 core servers
- Global Kernel locks in APFS
- Iain’s tweet about Google Bart
- AWS re:Invent
- Intel Cascade Lake Xeon - Pincount news!
- Sub $1000 IBM Power9 motherboard
- Adversarial Examples that Fool both Human and Computer Vision
- Fast and accurate object detection in high resolution 4K and 8K video using GPUs
Episode 31 - A Series of Small Fires
Doug and Iain reminisce about old computers
- Skype white noise
- Threadripper 2 Performance on Linux vs Windows
- Nvidia RTX
- Nvidia Video to Video Synthesis
- Deep Angel
- iOS 12 Core ML benchmarks
- Intel CPUs
- Reminiscing about old PCs
- Commodore 64
- ZX Spectrum
- Amstrad PC1512
- Amstrad PPC 640
- BT Gold
- School computers
- Apple iBook
- Iain’s cooling “solution”
- HP Jornada
- Dell Streak
- How the computing world was represented in high-school computing.
Episode 30 - Throw It All In The Sea
- More on local Python web servers.
- Intel 10nm
Episode 29 - Total Donkey
ARM servers, and our favourite command line apps
- More ARM on servers
- 10nm Intel CPUs in the wild
- Facebook job ad for a chip designer
- Xeon with integrated FPGA - Anandtech
- Command line utilities
- GNU parallel
- SimpleHTTPServer - For Python 2
- strace - Linux
- Bash tips:
ctrl-rfor reverse history search.
ctrl-ato go to the start of a line.
ctrl-eto go to the end of a line.
rm -rf /to test your backups.
- Fish Shell
Episode 28 - Taking a Baby
News, and Newsletters.
- Nvidia GTC 2018
- More Intel 8th Gen Core CPUs
- ARM PC followup
- PCIe SD cards
- Google Knusperli
- Generating images from human brain activity
- Newsletters we recommend
- More Computer Vision Trolling:
- Amazon Transcribe
Episode 27 - Either Which Way Round
What’s caught Doug’s eye over the break, and the future of ports.
- Bitmain buying 20k 16nm wafers a month from TSMC
- Nvidia Turing
- Nvidia Ampere
- Facebook Detectron
- Fast Photo Style from Nvidia
- Deep Learning Classification Explain Tool
- Google TPU Public Beta
- Chip reliability
- Branchless Doom
Episode 26 - The Three is Silent
Titan V Benchmarks, CPU bugs, and the new webscale stack.
- CGP Grey
- Titan V Deep Learning Benchmarks
- Turning Design Mockups into Code with Deep Learning
- S80186: 16-bit 80186 compatible IP core
- Google’s Learned Image Compression Challenger
- Meltdown and Spectre
- Why Raspberry Pi Isn’t Vulnerable
- What Spectre and Meltdown Mean for Webkit
- Meltdown Paper Summary
- Why performance will be hit by fixes - Ars Technica
- Performance impact on Windows systems
- Performance impact on Mac OS
- Meltdown Proof of Concepts
- Other CPU bugs
Episode 25 - Prime Nonsense
Iain and Doug discuss computational photography, and guess what might be next
- Toshiba sampling UFS 2.1
- Qualcomm Centriq 2400
- Nvidia Titan V - Anandtech
- Topic - Computational Photography
- Google Research blog posts
- Style Transfer for artistic impression
- Turning day into night
- Interactive editing of a video scene
- Occlusion removal
- Cats? No, you want cat memes!
- Doug watches interactive editing video
Episode 24 - I'm Not Special Any More
Some news, and a discussion on the future of networking
- Apple’s OLED display performance
- Topic - Networking
Episode 23 - I Ate a Lot of Pretzels
Iain goes to GTC Europe, Linux laptops, and moar RGB
- Correction on Coffee Lake
- Camera calibration for AR in Android
- Apple’s open source Core ML tools
- Intel shipping Nevana Neural Network processors
- AVX-512 in Cannon Lake consumer CPUs
- Intel delivers 17 qubit quantum processor
- GTC EU
- Off the shelf autonomous cars e.g. from AutonomouStuff
- Project Denver
- Caffe 2
- Side rant on Linux notebooks:
- Image - Facebook’s “small” deep learning cluster
- Image - RGB Man
- Video - RGB robot
- Image - MapD Bezos chart
- New show segment
Episode 22 - 1.2 Million Dollars of Hardware
ARM followup, more new Intel CPUs, and everybody has their own silicon.
- Higher accuracy GPS chips for smartphones
- iPhone 8 and iPhone X
- NVDLA - Nvidia Deep Learning Accelerator
- Microsoft Project Brainwave
- Google TPU Performance Analysis - Anandtech Live Blog
Episode 21 - ARM Macs?
We discuss the possibility of ARM Macs in the near future.
- eGPU support for Nvidia Quadro cards
- Video encoding on Amazon EC2 F1 instances
- More fooling deep learning algorithms
- AMD Threadripper review from Anandtech - CPU Modes
- Intel Cannonlake details
- Intel Coffee Lake leaks
- ARM Macs
Episode 20 - How does the Internet Work?
Doug goes webscale
- Intel Skylake, Kaby Lake Hyperthreading errata
- Amazon EC2 G3 Instances
- IBM Z14 - Buzzword driven computing
- Nvidia V100 in the wild
- Nvidia DGX-1 V100 pricing
- Neural Network hardware for Microsoft HoloLens
- Google Glass for Industry
- Adobe Flash EOL
- How does the Internet Work?
- What is “Webscale”?
- Ruby on Rails
- Vint Cerf
- MVC - Model-View-Controller
- Framework Song
- Global Interpreter Lock
- Phoenix web framework
- Getting to Facebook
- Content Delivery Networks
- Seconds count in web performance
Episode 19 - It's More of a Tea Tray
Doug tells Iain about WWDC, Iain pontificates on what might be coming in AR and VR.
- Intel Aye Nine
- Computex Taipei
- WWDC 2017
- CoreML (Not MLKit)
- Mac OS support for eGPU
- HEVC / H.265
- Apple Podcast Analytics
- Safari performance benchmarks
- More from Google IO
- New 10.5 Inch iPad Pro
- iMac Pro
- The Talk Show Live 2017
- StackOverflow helping developers escape vim
Episode 18 - Depressed Grad Student Slash Intern
Doug and Iain talk about SIGGRAPH paper preview videos
- Nvidia Volta - Anandtech
- Vega reveals itself
- Open source MapD
- New Cray supercomputers
- AI on ARM
- Google Fuscia
- Computer Vision and Graphics
- The Final Itanium
- Intel Aye Nine - Aye Right!
- AMD Socket 4049 for ThreadRipper
- Interview with Intel’s Mark Bohr
- CPU Utilization is wrong - Brendan Gregg
Episode 17 - Procount
What is a Pro Computer?
Episode 16 - A Flexible Omnishambles
Catching up on the news, and the mysteries of Thunderbolt 3
- AI literacy - Lisa Daly
- Chainer AI framework
- AMD Naples CPU
- AMD Ryzen 5 CPU Anandtech review
- Sony’s super fast image sensor is now available in a phone
- Facebook’s new Open Compute Project servers
- Microsoft’s “Project Olympus” Open Computer Project servers
- See episode 6 for LGA3647 chat.
- Actual Intel Optane products
- Google Guetzli perceptual JPEG encoder
- Google TPU
- Google’s custom NICs
- Thunderbolt 3
Episode 15 - What is AI?
A 15th episode special, Iain educates Doug with his views on what is and is not AI.
- Modal Logic
- Backus–Naur form
- Journal paper based on Iain’s PhD thesis
- Robot or Not?
- Netflix Prize
- The entity benefiting most from AI hype is… The government of Anguilla
- Iain’s computer vision tracking paper
- There are four giraffee species
- Deep Learning results on Google Trends
- Imagenet Challenge
- Generative Visual Manipulation on the Natural Image Manifold
- Deep Learning skin cancer identification
- Argumentation Theory
- Evolved Antenna
- HTML5 Genetic Cars
- Getting Started
Episode 14 - You've Munged This Whole Thing
Lightning news roundup, and what's still rubbish in 2017
- Short Topics
- Intel 10nm plans
- Nintendo Switch development platforms
- Modifying some simple code to use SIMD
- Taxi analysis - This time on 4x Phi!
- Stratus, 24 years uptime and counting
- Apple’s ARM Co-Processor
- High-Speed Sony Camera Sensors for Smartphones
- Faster Kaby Lake
- Faster high core count Xeons
- Webkit WebGPU API Proposal
- Fuzzing PCIe
- Things That Still Suck in 2017
- Collaborative document editing
- VOIP + conference call etiquette
- Batteries, kind of.
- PC laptops
- Mac laptops
- AMD, for now.
- What Programming Languages Are Used Most on Weekends?
Episode 13 - Random Tech News Podcast
Doug and Iain sweep up the newsroom floor, our favourite tech over the winter break.
Doug recorded using the wrong microphone, that’s why he sounds like he’s in another room this week. Sorry about that.
- Rick Branson’s tweet about Amazon EC2 R4 instance network performance
- ENA - Ethernet Network Adaptor
- PPS - Packets Per Second
- “R4 instances are optimized for memory-intensive applications and offer better price per GiB of RAM than R3.”
- Joe Domato’s Linux Kernel Receive Tuning Guide
- Amazon EC2 F1 instances
- Rick Branson’s tweet about Amazon EC2 R4 instance network performance
- Power8 Performance Numbers
- Calculating time remaining on Mac OS 10.12.2
- Leaked AMD Ryzen Benchmarks
- Anandtech Intel i7-7700k “Kaby Lake” Review
- Razer Project Valerie
- Dell 8k Monitor
- Benedict Evans tweet with picture of Ford’s self-driving-car computer
- Alexa, order me a dollhouse
- Paper - Hidden Voice Commands
- ML all the things - Google’s ML image compression
- GTA V + OpenAI Universe
Episode 12 - That'd Be a Pretty Small Truck
AMD Deep Learning, various Intel news and trivia, and a whole swathe of new AWS products.
- AMD Deep Learning Announcements
- Intel GPU Internals
- Agner’s CPU Blog - Knights Landing
- Intel x86 documentation has more pages than the 6502 has transistors
- Unlocked Intel i3 Processor: i3-7350k
- Intel self driving cars
- Emulated Intel on arm
- Bluetooth 5 now available
- “Next year it will work great” - John Gruber
- AWS re:Invent 2016
- James Hamilton Talk
- ASICS for their own switches. @0:23:00
- 7B transistors
- 128 ports of 25Gbe = 3.2Tbps (!!!!)
- Own silicon for NICs. @0:30:20
- Dropping networking to hardware rather than kernel
- Storage nodes @0:40:50
- 2015 design is 1,110 disks/rack
- 11PB with current-gen HDD
- Weighs 2,778 lbs (1,262 kg)
- Backblaze Pod
- Compute nodes @0:42:00
- Half-empty rack-mount case
- Optimised for thermal and power efficiency, not density of compute
- 3-5x less dense than off-the-shelf “cloud” servers
- ASICS for their own switches. @0:23:00
- AWS Snowball Edge
- AWS Snowmobile
- Elastic GPUs
- GPUs that can be attached to ANY EC2 instance.
- Examples show Windows desktop applications using them.
- OpenGL support only.
- 1, 2, 4, 8GB sizes.
- Dolphinics PCIe switches
- EC2 F1 Instances w. FPGA
- FPGA choo choo
- James Hamilton Talk
- Liberouter Combo cards
Episode 11 - 13,700 Files
- SuperComputing ‘16
- Intel Xeon E5-2600v4 Deep Dive
- Deeper on Apple MacBook Pro 16GB RAM limit
- X86 gaming onsoles vs gaming PCs
Episode 10 - Long on the Internet
Tiny enterprise SSDs, massive upcoming Intel CPUS, and have we reached 'peak laptop'?
- iPhone 7 Plus sensor sizes round 3. The two sensors are different sizes. This time. Really.
- Intel Announcements and Rumours
- New Macbook Pros
- USB-C cable nightmares
- Macbook Pro Thunderbolt 3 asymettric port speeds
- No more startup chime
- Restoring the startup chime
- 16GB RAM ceiling due to power concerns
- More on the RAM ceiling
- Rumours on 2017 Macbook Pros
- ATP on the new Macbook Pro: 1, 2, 3.
- Penny Arcade on the Microsoft Surface Studio
Episode 9 - Serious Beards and Facial Hair
A jumble of news, Iain goes to GTC, and watching Erlang the Movie.
- AMD AM4 Socket
- John Gruber on big.LITTLE
- 2Tb NVMe SSDs from Samsung
- FPGA in Microsoft servers
- Image compression with Neural Networks
- High Definition displays
- Sharp 8k 120hz Monitor
- IBM medical imaging displays (22” 4k, not 17” 2k as Doug guessed)
- Intel i7-7700k “Kaby Lake” Desktop benchmark
- Training TensorFlow models for use with iOS
- Apple hiring Nvidia driver engineers
- GTC Europe
- Nvidia announced Volta architecture with Xavier “AI Supercomputer”
- Ultra-Wide Curved Dell monitors
- New unannounced Quadros
- So much compute
- DGX-1? How mainstream.
- IBM Power 8
- Bull Sequoa
Episode 8 - Only 50 Euros Shipping
Followup on Xeon Phi, more deep learning GPUs from Nividia, and some maths on the iPhone 7 Plus camera.
- New Nvidia GPUS
- iPhone 7 Plus
- John Gruber’s iPhones 7 review
- LITTLE cores not used in Geekbench Benchmarks
- big.LITTLE bug in Mono
- Nine-levels of depth detection
- Remours of Intel modems in iPhone 7
- Some Sensor Sums
- On a 7+, the wide is 3.99mm /_f_1.8 on 1/3”, and the tele a 6.6mm /f2.8 on 1/3.2”
- Angle of view is α = 2arctan( d / 2f), where d is the sensor width.
- So decrease f, to get the same α you need to keep (d / 2f) constant, so d decreases.
- Calculate f number like: N = f/D, where f is focal length and D aperture diameter.
- So double f and 22 the area of glass needed.
- But f isn’t doubled on the tele, as the sensor is smaller so doesn’t need to double to half the FOV (2x zoom) - see the above point about α.
- Speculation and deep dive on future ceramic iPhones
- Nilay Patel’s iPhone 7 review on The Verge
- Intel buys Movidius
- The BBC’s Turing Codec
Episode 7 - Special Apple September Keynote Flash Express Edition™
CPUs, Cameras, and waterproofing from Apple's September Keynote
- Apple Watch Updates
- iPhone Water Resistance
- iPhone Camera
- iPhone CPU
Episode 6 - Waving Their Latest and Greatest Around
Flash Summit, IDF, Hot Chips and silly names
- Flash Week
- Doug says H.264 instead of H.265 @ 12:57.
- Nvidia doesn’t like Intel’s Benchmarks
- Pincount 3647!
- Phi in the Ark
- Why Didn’t Larrabe Fail
- Amazon charts
- AMD do CPUs too
- Hot Chips
- Aftershow - Fun Computer Names
Episode 5 - Glass and Gold Velcro
Display technology, HP's "The Machine", microwaves.
- The Machine
- Bad microwave interface design
- KnobFeel - the non-porn Tumblr
- Iain’s student microwave, Comrades.
Episode 4 - How the Sausage is Made
Facebook Surround 360, what powers the Pincount website, and who's drinking what.
- Facebook Surround 360
- How the Sausage is Made
Episode 3 - Silly Names
Titan X Pascal, software vs. video encoders, FPGA vs ASICS.
- Episode 1 audio
- Doug mis-spoke
- Mediatek 10 Core SoC x20 Dev Board
- Titan X, Again!
- New Intel Chips - Coffee Lake
- Are software encoders relevant?
- Xeons specifically for video
- Intel VCA board (Xeon PCIE boards)
- FPGA vs ASICS
Episode 2 - HR in VR
VR tracking, possible VR specific silicon, upcoming video codecs.
- Qualcomm Project Tango - Lightweight 3D tracking in silicon for mobile.
- Lenovo Phab 2 Pro
- Apple WWDC Metal Part 2 Replay
- Inside-Out Tracking
- How the Vive Lighthouse Works
- Cortex Episode 32 - Dropping Acid - Grey and Myke experience VR for the first time.
- CHOO CHOO
- AMD Sulon
- Scale in VR
- Lands End - The game Iain couldn’t remember.
- Video nonsense
- HEVC / H.265
- Intel E5-1500 V5 CPUs with hardware HVEC encoding.
- Qualcomm Snapdragon 800 Series supports H.265 decoding in hardware.
Episode 1 - Being Cheeky
Deep learning on iOS, speculation on upcoming AMD and Nvidia GPUs.
- Metal and deep learning
- Nvidia mac speculation
- RX-480 Release