April 6 th, 2018 - The WAN Show Document

[LinusTech]

 

 

 

Main News Topics

LINUS IS PISSED - Nvidia AIB BS

Source 1: AluminiumTech https://linustechtips.com/main/topic/913534-nvidia-their-aibs-are-deceiving-customers-yet-again-with-new-gt-1030-variant/

Source 2: https://www.anandtech.com/show/12614/nvidia-quietly-rolls-out-slower-lower-tdp-geforce-gt-1030-with-ddr4-vram

 

• Sometime last month, NVIDIA quietly released the GeForce GT 1030 DDR4, with several models filtering out through add-in board partners.

• In addition to slower-clocked DDR4 VRAM, the GT 1030 DDR4’s core clocks have also been reduced, though the exact boost clocks differ between custom boards.

• The end result of 2100Mbps DDR4 from 6008Mbps GDDR5 reduces the bandwidth from around 48 GB/s to 16.8 GB/s, though with no change to memory size and bus width.

• It appears that the DDR4 variants are also lower power, with a 20W TDP as opposed to 30W.

• Some of the listed documentation shows two different core names: “GP108-300” and “GP108-310.”

  • But while “GP108” is affiliated with the lower-clocked, lower TDP DDR4 variant, certain GDDR5 models (AERO ITX 2G OC, AERO ITX 2G OCV1, 2G LP OC, 2G LP OCV1, 2G LP OCV2) list both core names in their specification sheets but with everything else unchanged, leaving it unclear what the differences are between these two GP108 bins.

• The official NVIDIA GeForce GT 1030 product page mentions nothing of the change, though it does note that specifications of partner board may vary.

  • each of the model names except one differentiate with a “D4” moniker.

 

Facebook Messenger Scans & Censors Chats

Source 1: AluminiumTech https://linustechtips.com/main/topic/912936-facebook-private-chats-arent-actually-private-and-theyre-actively-censored-by-facebook/

Source 2: https://www.bloomberg.com/news/articles/2018-04-04/facebook-scans-what-you-send-to-other-people-on-messenger-app

 

• Zuckerberg has confirmed that messages in the Facebook Messenger app are scanned by automated systems before they are sent to ensure they don’t breach community standards

• Infringing content is never sent

• Instagram, Facebook Messenger, and Facebook operate the same way

• Data gleaned from scanned messages is NOT used for advertising

  • Why not? Gmail does did this for years.

• The forum thread erroneously implies that this censorship applies to Facebook Messenger’s E2EE function (Secret Messages)

 

Intel gives up on patching Spectre variant 2

Source 1: CommandMan7 https://linustechtips.com/main/topic/912801-intel-gives-up-on-patching-spectre-variant-2/

Source 2: https://www.zdnet.com/article/intel-we-now-wont-ever-patch-spectre-variant-2-flaw-in-these-chips/

 

• Intel is dropping plans to patch certain CPU families affected by the Spectre variant 2

• The chipmaker has spent the past few months releasing and re-releasing microcode updates to fix the Spectre variant 2 flaw, but while it's rolled out updates for all processors launched in the past five years, it has now revealed some older CPUs won't be patched at all.

• Intel's latest Microcode Revision Guidance applies a new 'stopped' status to several CPU product families for which it had been developing microcode updates, including

  • Bloomfield, Clarksfield, Gulftown, Harpertown Xeon C0, Harpertown Xeon E0, Jasper Forest, Penryn/QC, SoFIA 3GR, Wolfdale C0 and M0, Wolfdale E0 and R0, Wolfdale Xeon X0, Wolfdale Xeon E0, Yorkfield, and Yorkfield Xeon.

• Intel says it stopped developing the mitigations for at least one of three main reasons

  • Micro-architectural characteristics that preclude a practical implementation of features mitigating variant 2 .

  • Limited commercially available system software support.

  • Based on customer inputs, most of these products are implemented as 'closed systems' and therefore are expected to have a lower likelihood of exposure to these vulnerabilities."

• Is that fair?

 

What’s Coming Out on Floatplane? Go Over Current Week.

Rapid Fire

Rumour: PS5 Coming in 2018?

Source 1: Master Disaster https://linustechtips.com/main/topic/913109-playstation-5-leakrumours/

Source 2: https://www.youtube.com/watch?v=jTa10-XXGac

Source 3: https://www.express.co.uk/entertainment/gaming/941817/PS5-release-date-Sony-PlayStation-E3-reveal-specs-VR

 

• Previous speculation by some analysts put the PS5 release around 2019 or 2020, citing a sustained wave of PS4 Pro sales as Sony could push that system’s price toward $250

• But now rumours are circulating about a holiday 2018 launch of the PS5

  • Rumours first published by Semiaccurate.com, behind a $1,000 per year paywall ($100 for students)

• The console will reportedly:

  • AMD’s Navi as its base architecture,

  • An 8 core Zen CPU

  • be backwards compatible with PS4 Pro games.

  • Have a strong emphasis on VR support

• A lot of people are not convinced given history:

  • PS3 debuted in 2006 while PS4 wasn’t out until 2013. (7 years)

  • PS4 Pro arrived in 2016 (3 years later, 2 years ago)

  • PS5 in 2018 would mean a five year life cycle for PS4, and another upgrade for fans just two years after Pro.

  • Sony also started teasing PS4 in February 2013 before its November 2013 release, and here we are in April 2018 with nothing.

• An announcement at E3 2019 seems plausible (late May, early June)

 

Jaywalking in Shenzhen

Source 1: https://www.reddit.com/r/Cyberpunk/comments/89am3j/thank_you_for_cooperating/?st=JFJTONK2&sh=1da0ab9d

Source 2: https://www.cnet.com/news/jaywalking-in-china-surveillance-system-will-sms-you-a-fine/

 

• In Shenzen, a city with a population of over 12 million, a company called Intellifusion has worked with local police to set up AI-powered billboards that will display your face and family name if you’re caught jaywalking by facial recognition-armed outdoor cameras.

• Now, Intellifusion is looking to work with mobile carriers and messaging platforms like WeChat, China's version of WhatsApp, to notify jaywalkers of their transgression on their phones.

• Over a period of 10 months, ending in February this year, the boardsAI displayed 13,930 faces of jaywalkers

• only 10 percent of the city's population can be identified by the system, but this number will rise dramatically once different provincial governments merge there databases.

• Is jaywalking different in China?

  • Stop lights may you wait longer (90 seconds vs 60 in the West)

  • Many drivers don’t respect pedestrians even when they have the right of way

  • Therefore there’s more jaywalking..

  • ..which results in major traffic disruption

• Some cities have turnstiles that only open when the light changes

 

Facebook Smart Speaker

Source 1: https://www.bloomberg.com/news/articles/2018-03-28/facebook-is-said-to-delay-home-speaker-unveil-amid-data-crisis

Source 2: https://techcrunch.com/2017/07/25/facebook-speaker/

 

• Facebook Inc. has decided not to unveil new home products at its F8 developer conference in May, in part because the public is currently so outraged about the social network’s data-privacy practices

 

• The new hardware products, connected speakers with digital-assistant and video-chat capabilities, are undergoing a deeper review to ensure that they make the right trade-offs regarding user data

• The “Portal” devices, codenamed Aloha and Fiona:

  • Include a 15-inch touch-screen sourced from LG

  • A smart camera

  • Voice control

  • a magnesium-aluminum-alloy chassis.

  • Assembled by Chinese iPhone manufacturer Pegatron.

  • Designed by Facebook’s secretive new hardware lab Building 8

  • Facebook has signed music licensing contracts with Sony and Universal Music to enrich the device's applications

• The original rumoured launch was May 2018, then July. Now it looks like fall.

  • a small pilot run has already been conducted

 

Rumour: NVIDIA GPP, Meet ASUS AREZ

Source 1: https://videocardz.com/75783/nvidia-gpp-meet-asus-arez-radeon-series

Source 2:

 

• ASUS could be launching a new AREZ brand for Radeon series.

• These cards will not carry any ROG branding, and might even lose ASUS branding altogether.

• There won’t be any AREZ ROG STRIX series, but AREZ STRIX instead.

  • Same applies to other sub-series, such as Phoenix, Expedition and so on. These are called AREZ Phoenix or AREZ Expedition respectively

• Is it pronounced Air-ease or Are-ez?

 

In-Display Fingerprint Sensors

Source 1: https://www.anandtech.com/show/12593/goodix-ships-optical-indisplay-fingerprint-sensor-for-smartphones

Source 2:

 

• Goodix has launched its first optical in-display fingerprint reader for use in smartphones with AMOLED screens

• One of the largest suppliers of fingerprint readers, Goodix says that its optical in-display fingerprint sensor works faster than conventional capacitive fingerprint readers

• The reader is already used inside Vivo’s X21 and Huawei’s Porsche Design Huawei Mate RS smartphones.

• Where would you want the fingerprint reader to be? Reachable by thumb?

• Would we be losing our rear-mounted fingerprint readers?

• Will these perform better or worse than conventional when you’ve got wet hands?

  • Synaptics’ under-screen fingerprint sensor works even under water.

• What if you crack your screen?

 

Hairio

Source 1: https://www.youtube.com/watch?v=8JV2D7gJ5HI

Source 2:

 

• gross.

 

T-Mobile Austria Stores All Customer Passwords In Clear Text

Source 1: LAwLz https://linustechtips.com/main/topic/913735-t-mobile-austria-confirms-they-stores-all-customer-passwords-in-clear-text/

Source 2: https://twitter.com/tmobileat/status/981418339653300224

 

• Hilariously ignorant and rude agent knows nothing of cybersecurity

• T-Mobile Austria is not owned by T-Mobile, but is a subsidy of  Deutsche Telekom Group

 

MIT’s new headset reads the ‘words in your head’

Source 1: http://news.mit.edu/2018/computer-system-transcribes-words-users-speak-silently-0404

Source 2: https://techcrunch.com/2018/04/06/mits-new-headset-reads-the-words-in-your-head/

 

• AlterEgo is a jaw-mounted device that’s capable of reading neuromuscular signals through built-in electrodes.

• Electrodes in the device pick up signals in the jaw and face that are triggered by internal verbalizations — saying words “in your head” — but are undetectable to the human eye.

• The signals are fed to a machine-learning system that has been trained to correlate particular signals with particular words.

• In a study in which 10 subjects spent about 15 minutes customizing the application to their own neurophysiology, then another 90 minutes using it to execute computations, the system had an average transcription accuracy of about 92 percent.

• The device also includes a pair of bone-conduction headphones

[Tech_Dreamer]

Thanks for the after party m8.

[Canada EH]

Hes got no made up angry mad pissed face for the thumbnail. And another new face in the warehouse.

[Shreyas1]

ANOTHER smart speaker?

[Daniel644]

I TOTALLY created my twitter account just to enter giveaways and it worked, I won a 1080ti from that Esports thing that happened back in November.

[Canada EH]

23 minutes ago, Daniel644 said:

twitter account

What is twitter?

 

 

Sounds like something to stay far far far away from.

Maybe log in once every month or two, yes!

 

Well thats good you won a gpu, now go mine

...or the question shall be asked, how much you made on that gpu

[Daniel644]

30 minutes ago, Canada EH said:

What is twitter?

 

 

Sounds like something to stay far far far away from.

Maybe log in once every month or two, yes!

 

Well thats good you won a gpu, now go mine

...or the question shall be asked, how much you made on that gpu

added it to the other 1080ti I already had and sent them to the mines just days prior to the Nicehash hack, they are currently gathering over a thousand Pigeons a day in hopes that coin goes to even Raven levels when it reaches some exchanges, heck if it reaches even half of Raven's value thats December+ Level profits per day.

[ahuckphin]

First we relished that Nvdia had two versions for the MX150 

and now we relished there's two version for the GT1030. 

[adamantiumxt]

The hypothetical thing Linus was talking about in China virtually exists: https://www.wired.com/story/age-of-social-credit/. It's quite long but very worth a read.

[Nicnac]

I kinda like that we are seeing new "brands" after GPP. Maybe finally an opportunity for different colors and designs?

[Egg-Roll]

Twitter what's that?

https://twitter.com/is_a_fail

 

13 hours ago, LinusTech said:

Is jaywalking different in China?

• Stop lights may you wait longer (90 seconds vs 60 in the West)

• Many drivers don’t respect pedestrians even when they have the right of way

• Therefore there’s more jaywalking..

• ..which results in major traffic disruption

And this is different how?

If you don't have placebo buttons where you live it comes down to the systems reliability and local authorities willing to upkeep their equipment.

Where I live I've waited upwards to 4 minutes for a light, and there are light systems which would completely ignore my request for 3 cycles. Plus the numerous places the buttons don't work at, and the fact that 95% of the drivers here rather ignore pedestrians rights over letting them go for 5 seconds...

 

Linus should visit Kennedy and Steeles. Just make sure when you cross legally anywhere in that area you wear a full body armour suit  It's completely possible where LMG workers live there are good drivers, equally if they mostly drive many drivers fail to see the wrong or errors in others or themselves as drivers. Just saying.

 

It's like last weeks WAN show where people took videos of the area where the pedestrian died, that proves very little due to how weather works. Also I noticed in one video where she crossed is actually a dark patch where you still can miss a person crossing depending on speed. IMO those who shot those videos using their phone in their hand should get fined, regardless of their intent they are driving with a device in their hand.

[JCHelios]

"Here they are doing it again, and it's only been a few months [since the GTX 1060 launched]."  — @LinusTech 

 

Summer 2016, dude.  Almost two years.  July '16 for the 6GB; August for the 3GB.  

As James said, our perception of time keeps speeding up the more of our lives are behind us.  ?⚰️?

[Lathlaer]

With all due respect Linus, if you should get criticism it's not for pairing a strong CPU with weak GPU while doing GPU testing but for your insistence of doing majority CPU tests in games in 4k.

 

I mean:

 

 

something like that. 

 

"Not quite sure why that would be?" Hmm, maybe cause you tested Deus Ex and Tomb Raider in 4k as a CPU test and CS GO in 1080p? ;-)

[dalekphalm]

I really like James. I would like him more if he ate Jerky (JK bro, you're still cool).

[Xilefian]

I made an account just to comment on something that was touched on in this WAN show.

 

At 00:38:30, with the discussion of the "PS5 rumours", James asks 'prior to Ryzen, all the games/engines/APIs are optimised for single-threaded', Linus says "Strongly depends".

 

It felt like Linus knew that James was a bit flawed with where he was coming from with that single-threaded optimisation comment, but didn't know exactly what.

 

The history is a lot more interesting and complex. There's loads of stuff going on, but it would indeed be incorrect to say that prior to Ryzen games were optimised for single-threaded execution.

 

About Graphics APIs and Multi-threaded CPUs

 

The modern, explicit graphics APIs (Mantle, Metal, DX12, Vulkan) are designed to allow easy multi-threaded use, where you can dedicate any number of threads to building up command buffers before sending them off to the GPU driver - which handles the commands and does whatever it needs to before issuing actual GPU instructions.

 

For older graphics APIs, OpenGL and DX11, the API calls must be called from a single thread, you can't have two threads issuing API calls (in general - there are crazy ways you can do this in OpenGL, but OpenGL support in games hasn't been great compared to DX).

The solution that game engine programmers came up with is to create a system where multiple threads can issue their own dummy draw-calls, and then the single graphics thread receives these calls, organises/optimises them, then hands them off to the GPU driver (via traditional OpenGL/DX11). A lot of game engines have been built this way, and these game engines are the ones that had the easiest transition to DX12/Vulkan.

What this essentially created was a situation where game developers were creating systems where you would build explicit draw commands on any thread, send them to a single graphics thread which sorts them into neat OpenGL/DX11 API calls, then sends them off to the GPU driver to do whatever it has to do (including checking state, analysing what you're doing, inserting error correction).

So we have been optimising these old graphics APIs for multi-threaded environments by writing a fat abstraction layer, but with Vulkan/DX12 that extraction layer can be removed (at least, made lighter) as they natively allow the use of commands to be built on multiple threads. Multi-threaded support is just a tiny slice of what modern graphics APIs give us, there's more stuff that's a bigger deal (in my opinion) than multi-threaded API calls, but that's off topic for the multi-threaded optimisation question.

 

DOOM 2016 is an example game of this technique being used.

About Game Engines and Multi-threaded CPUs

When the Xbox 360 and the PS3 were launched the state of game engines was definitely "Everything on a single thread", however these consoles both introduced some form of parallel compute that you really needed to take advantage of if you were to write an optimised, graphically impressive title for these systems.

The 360 has a triple core processor, each core with two threads, so to really take advantage of that CPU power you had to rethink your technology and the industry had to come up with some new paradigms.

The PS3 was an even bigger deal when it came to parallelism. The PS3 has a single core processor, with two threads, but has those crazy 8 SPEs that can absolutely churn through compute operations all in parallel. The PS3 was weaker than the 360 when it came to single-core processing, and it was left in the dust when it came to multi-threaded CPU work loads, but it was massively ahead when it came to parallel stream compute with the SPEs. In some incredibly specific compute cases, the PS3 can outperform an original model PS4.

This is part of why the 360/PS3 launch games looked terrible, but the consoles had a very lengthy life-span and the end-of-life games looked rather excellent. The industry had to create new paradigms to best utilise the parallel compute hardware available; the PS3 in particular popularised the job-manager paradigm^* where threads can queue up jobs to be churned through by execution units (SPEs in the PS3's case, CPU threads in the 360's case).

The PS3's parallel compute lives on to this day with GPU compute - where the GPU is used for parallel mathematics. Pretty much everything that was done on a PS3's SPU is these days done on the GPU - this is stuff like particle simulation, physics, AI path-finding (a notorious computer science exercise to implement for parallel compute!) and even image processing.

That console generation made it clear that the technology needs to be multi-threaded if it wants to take full advantage of what's available. At the end of that console generation all the big game engines were taking advantage of multiple CPU threads in some form or another - using multiple threads for building GPU API calls is just one example of this.

 

Rage (2010) is an example game of these parallel compute paradigms being used, as is Doom BFG Edition (2012).

About multi-core in the PC space

 

When Ryzen was released pretty much no game developer expected performance to be better with more cores. I think that's a misconception that gamers would have created on their own (perhaps AMD marketing teams perpetuated this kind of belief). The CPU is not much of a bottleneck on games now, games are pretty lightweight on processing (with it all being spread across multiple threads). It's almost always the GPU that is acting as the bottleneck.

There are engines were the CPU can still be a bottleneck - Unity is notorious on the PC for its hefty interpreter engine being pretty killer on single-threads, but even Unity itself is getting more and more optimised for multi-threaded work loads (the graphics engine team at Unity is doing a really fantastic job right now).

The only thing that needed optimising for Ryzen on release were the compilers where some aspects of Ryzen could be better taken advantage of. I remember Ryzen exposing some MSVC compiler bugs where pipeline flushes and cache clears were being incorrectly triggered - the talk amongst peers is that Ashes of the Singularity's "Ryzen optimisation" was literally just recompiling the game after Microsoft fixed the compiler bug. Game developers pretty much have no work required for "Ryzen optimisation" (just make sure you compile your project with non-buggy compilers).

 

Any other low-level benefit Ryzen itself gives is best dealt with by the compiler developers, not game developers. The multi-core factor applies to all CPUs, Intel and AMD, so optimising for multiple cores is not a Ryzen optimisation.

On console hardware the compiler optimisation can be even more specific is it only needs to target a specific piece of known hardware. There are cases where an engine programmer will treat the CPU of a console differently to a PC, but in these cases the beefy CPUs that a PC has will stomp all over the hardware specific optimisations of a console. Every console-specific CPU optimisation that comes to my mind also applies to every desktop PC processor - Intel and AMD.

GPU is a different ballgame when it comes to consoles, thanks to unified memory and pretty much zero driver overhead (which is why D3D12 has pretty much zero benefit on XBone compared to PC).

---

There's loads, loads more that I can go into as the journey to optimising game engines for multiple threads has been a long one, and there's still room to go.

 

It is definitely incorrect to suggest 'prior to Ryzen, all the games/engines/APIs are optimised for single-threaded'.

---
 

^* The job-manager paradigm is great because it scales with any number of CPU threads available, and in cases like the PS3 and GPU compute you can let some jobs be handled by the stream compute hardware.

[DaiGurenMK42]

Luke & James really is the best combo for news and fun.

 

Luke & Linus is the best combo when there are no tech topics and when Lukes Mom calls. 

 

And occasional guest appearances from the rest of the crew is totally fine. This shows for entertainment not news. 

[Canada EH]

29 minutes ago, DaiGurenMK42 said:

This shows for entertainment not news.

I would agree on only one part of your statement

 

Spoiler

its not news