Slashdot Mirror

Since HECV became popular the price of a build went from $150 to $450

Get yourself a low-end card that does HEVC in hardware (such as a non-Ti GTX 750/950/960) and call it a day.

Nvidia won't release a VR headset because they don't care about it. They care about AI.

https://research.nvidia.com/re...

Yeah, but aside from maybe watching on a table while in a car/plane, who would want to actually watch a crappy copy like that?

So, Nvidia includes vastly improved hardware screencap encoders in their new RTX cards... and now 4k iTunes rips are appearing. Coincidence???

I'm in the Nvidia GeForce NOW beta, which is a similar service. The games run on their servers and the display is streamed back to you. I must say that it works much, much better than I ever expected. I'm in a fairly rural area but I have a decent connection at about 110 Mbps down, 8 Mbps up. It's usually quite playable, even over wifi. It does chew through a lot of bandwidth and latency can be an issue. The biggest caveat is that I like single player turn-based games. Even when I play real-time first-person games they tend to be Tomb Raider-ish, with more of a stealth and exploration focus rather than reaction time and precision. But it lets me play games with higher minimum requirements than my old-ish laptop meets, and lets me play a world of PC games that have never been ported to my Mac.

This kind of technology can work reasonably well for a lot of people. It's definitely not ready yet for the super competitive players, and I doubt it ever will be (barring some great revolution in subspace communications...) But I think there's a market for it and I'm glad to see competitors in the space.

These statistics are really lousy and does not say anything about if an expensive card will really increase the player performance.

1. The "low" frame rate is 120 fps which would not by any means be considered "low" by most gamers.

2. They only asses GPUs and do not include displays and other hardware or network lag. They have no idea of which frame rates are actually experienced by the players. There is even a graph to show that the more expensive cards are helpful even on 60 fps screens, despite that all studied GPUs produce much more than 60 fps.

3. As many have pointed out, this is very likely a skewed data set to start with. The serious gamers buy more expensive cards. Nivida actually have the data to confirm or refute this but don't show it. They could show the distribution of hours played per week, grouped per GPU type. They don't show it which is cause for concern.

There is little reason to believe gamers would experience any significant improvement going from a 120 fps to 200 fps system. That would be equivalent of reducing absolute worst case lag from 8.3 ms to 5.0 ms (on average it would actually be half those numbers). Then consider that the human reaction time to visual stimuli is around 120-150 ms in ideal setups.

Nivida claims that "pros in our labs have been able to consistently discern and see benefits from even a handful of milliseconds of reduced latency". That may be true, but it just proves that the handful of milliseconds that you could possibly gain can only be discerened by pro players on extremely high-performing systems in controlled environments.

Extraordinary claims require extraordinary evidence. In this case Nividia does not even produce ordinary evidence.
This is purely a marketing ploy.

Exactly. And we are here to help! There's only one way you can solve this: https://www.nvidia.com/en-gb/s...

Regards
Your friendly neighbourhood NVIDIA representative.

the GeForce Experience is now the ONLY way to access Nvidia drivers.

Bullshit. You can no longer download a driver package which doesn't include the install files for geforce experience, but you don't have to install it. Nice FUD, how long have you been working for AMD?

Please link to the fraudulent statement so I can point out where your reading comprehension failed.

Jeesh some people need to be spoon fed. Why not follow the links in TFA through to the actual legal filing?

"Specifically, defendants: (1) assured investors that NVIDIA followed the market closely and could adjust to rapid changes in the cryptocurrency markets; (2) touted that NVIDIA and its executives are “masters at managing our channel, and we understand the channel very well.”; and (3) assured investors that surging demand for graphics processing units (“GPUs”) among cryptocurrency miners would not have a negative impact on NVIDIA because of strong demand for GPUs by NVIDIA’s core customer base of computer gamers. When the true details entered the market, the lawsuit claims that investors suffered damages."

Want to do your own legal analysis http://investor.nvidia.com/ Go your hardest.

Indeed.

Considering they use the phrase ... critical "inferencing" computation, called multiply and accumulate ... instead of the terms fused MAC or FMAC I'm wondering why are they just using nVidia's Tensor Cores ?

Not a question of marketing, I promise. The Iray render engine (built into a lot of CG rendering suites and apps nowadays) requires CUDA-enabled drivers, or else the render kicks to CPU for calculations, causing render times to go up by factors. This means a 30-minute render suddenly takes, say, an hour and a half... if you're lucky.

Microsoft's WHQL GPU/video driver has CUDA disabled, so you're stuck with CPU (not GPU) rendering - and you usually don't find out until after it begins. Also, when you have a 6-12GB GPU card, all that RAM goes to waste under Microsoft's driver. :/

Re games support and RTX. https://nvidianews.nvidia.com/...

Right, Nvidia bought 3dfx after they had bled

By bled you mean they filed a patent lawsuit...which of course happened after 3dfx had done exactly the same thing to nvidia.

them dry

That link points to a lawsuit filed by the trustees of 3dfx's bankruptcy proceedings against nvidia.

with lawsuits.

That link is a lawsuit between nvidia and samsung, it's nothing to do with 3dfx. did you even read it?

It was Nvidia that stole IP knowledge from SGI (Silicon Graphics Inc).

Hang on, first you were saying they stole from 3dfx, which is wrong, then you changed it to be that they just sued them into bankruptcy (which actually was started by 3dfx themselves) and now you're arguing that nvidia stole from SGI which was actually patent infringement (which is not theft) and which is what 3dfx first sued nvidia for.

Right, Nvidia bought 3dfx after they had bled them dry with lawsuits. 3dfx did throw the first patent punch. It is too bad for 3dfx that their Voodoo 3 (their first attempt at a flagship 2D/3D combo card) was absolutely utter rubbish, and they decided to produce it all by themselves, in Juarez, Mexico (at the STB facility they'd recently purchased). Additionally, they couldn't make a decent AGP 4x card. So, basically, the dropped their current business plan for a completely untested business plan. Or they took a double-barreled shotgun and carefully aimed at first one foot and then the other.

As for the patent lawsuits, I was confused. It was Nvidia that stole IP knowledge from SGI (Silicon Graphics Inc). As part of the settlement with SGI, SGI turned over all their IP and graphic engineers to Nvidia.

Fun factoid: the 'Unix' computer that was used in Jurassic Park was actually an Silicon Graphics Crimson running Irix 6, and that graphical file system navigation tool (FSN) was actually created to help market and sell the Crimson. Sadly, I cannot link to FSN as it appears that the siliconbunny site has gone down.

https://developer.nvidia.com/e...
TLDR: 10x better power efficiency than the TX2 and 20x the performance.

How about NVIDIA Ray Tracing?

With motion capture technology getting better and cheaper - and capable of real-time capture - incorporating a body in VR should be a solved problem soon. Startups are flocking to this space:

Perception Neuron Hardware
https://neuronmocap.com/produc...

Kigurumi Live Animator (real time animation running on Perception Neuron)
https://kila.amebaownd.com/

See Kigurumi Live Animator Animation demo here:
https://www.youtube.com/watch?...

PrioVR:
https://yostlabs.com/priovr/

Ikenema Orion:
https://ikinema.com/orion

Motion Shadow:
https://www.motionshadow.com/v...

Coming "real soon now" Mocap with your phone camera: (open beta available)
https://blogs.nvidia.com/blog/...

Every year a good percentage of computer science and computer engineering undergrads write a small kernel for their coursework. It's not hard to write a kernel that solves a narrow set of requirements. When you have an every expanding scope, like in the Linux world, it gets hard. Linux has to run kiosks, mobile phones, desktops and supercomputers. And it's probably not the best possible kernel for any one of those problems, and certainly more complex than a kernel designed for a single specific purpose.

Why do we need Google Fuchsia? We don't really. But a whole lot of people leverage Little Kernel for their projects (my own company is using LK in 3 totally different ways). I blame LK for projects like Fuschia as it has turned into sort of a DIY Operating System kit.

If Linux and the BSDs are too complicated or you're just looking for some kernel you can hack up to meet your own special needs:
* NewOS
* Xv6 - a teaching OS. but people have patches for virtual memory and other goodies
* LK (little kernel) and for an example of a fork TLK (Trusted Little Kernel). LK is quite a good starting place for an aspiring osdev'r
* basekernel - rough starting place for making your own kernel
* FUZIX - a UNIX-like kernel geared toward 8-bit CPUs. but can be ported to bigger CPUs (there is a 68K port for example)
* TinyOS
* Femto OS - a kernel suitable for multitasking on small microcontrollers
* PonyOS - a graphical OS for people who love ponies (OMG Ponies). If you're looking for a more serious version see Toaruos.

There are hundreds of these hobby and learning OSes and several more complete and better established ones (like FreeRTOS).

I'd really appreciate the ability to switch between OSes like I can virtual desktops. Modern hardware certainly supports this potential.

There'll never be a product that truly allows you to switch the OS on the fly at will. There cannot be at least not without redefining our typical hardware concept of a PC. An OS by definition is the "operating system" and there can be one. You could certainly have a hypervisor built into the OS but a hypervisor cannot be the OS itself.

What they're asking for is to be able to run multiple operating systems and switch between them. This functionality already exists, and the technology which makes it work as the user expects is GPU Virtualization. AFAIK you need to use real VMware and not just player or workstation to use it, but I'm not actually sure about that. This lets multiple VMs share the GPU without hacky translation layers. Obviously, you need another OS underneath your operating systems; as you say, we call this a hypervisor. You also need a "professional" video card; can't just use a GEforce, because nVidia locks that functionality down to Quadro or maybe even only GRID cards.

I'm very much looking forward to installing this. I recently put together a nice Machine Learning / linux workstation / build machine.... https://pcpartpicker.com/list/... And Linux pre 4.14 just flubbed pretty bad with the processor... https://www.phoronix.com/scan.... I got things working somewhat smoothly with Manjaro linux, but getting the CUDA support working was a total hack (currently GCC 6.3 is all they support, not 6.4, much less 7.3 and the arch linux "fix" is very much an admitted dirty hack), and getting Caffe 2 to compile right was turning into more work than it was worth ... http://docs.nvidia.com/cuda/cu... ... So yes, these problems are largely the fault of Intel and NVIDIA clutching their proprietary pearls, but I am looking forward to running a well supported and stable version of linux that can support and be optimized to the latest hardware that came out 6 months ago.

Many of the Linux ports are shoddy and it showed in the number of issues they suffered. It's not the fault of Linux or of some inherent superiority of Windows but of time, resources, effort on behalf of the developer.

It's economies of scale. Obviously developers aren't going to put the same effort into servicing a very tiny portion of the market as they do into servicing the majority of the market. Most gamers also don't care about the OS at all, they'll play on whatever it runs best on because you can always just dual boot.

And not just of the game dev, but also the graphics driver. NVidia / AMD push out frequent Windows driver updates that optimize performance or fix issues for specific games. Linux gets none of that.

That's just completely incorrect. The changelog on one of the very recent linux drivers for nVidia's GPUs shows multiple fixes for specific games:

http://www.nvidia.com/object/linux-display-ia32-304.37-driver

Moreover shader compiler optimizations are shared across platforms as they aren't related to the OS.

Self driving cars are neither run by AIs nor by Deep Learning Neural Networks.

Nvidia made a clear point that their platform was indeed pure DNN (aka "end-to-end deep learning self driving cars").
(Because it was more a demo of the application of their tensor acceleration rather than a practical implementation of driving).

But also has given rise to criticism about a technology that we can't completely understand (simply by not being based on a set of rules).

Recognizing a street sign might be so ...

Yes, in most cars (except Nvidia), the final decision is rules based (as in "if there an object that's too near, hit the brakes")

The problem here, is what goes upstream.
Neural nets use go way beyond recognizing signage.
We have moved past using Hough-space to recognize lane in early LDAS a long time ago. Neural nets are used to make better sense out of the sensors input, like tagging all the various object in the scene seen by the cameras.
(Comma.ai's blog gives a nice overview of what goes in a modern car system)
It's these modern classifier that enable auto maker to start detecting new object like cyclists, etc.
(e.g.: Volvo's various announcement of the improvment of their City-Safety)
But it's this type of image recognition that is the weirdest to understand and might get fooled in weird ways (Volvo having problem with kangoroos crossing the road in Australia, because by jumping vertically they might seem to move forward and backward to the monoscopic camera's perspective)
This makes the signals coming out the object dections/tagging of the sensors less reliable.

Thus a simple rule like the above "if there an object that's too near, hit the brakes" can become much less reliable in practice because the autonomous system will have trouble determining what's an object and were it is.

System are already reaching border condition where the standard input filtering gets in the way.
  - Uber mentioned that their accident might be caused by the radar normally filtering out signals that don't move at a different speed (it tries to find incoming cars, but will miss a bike going perpendicular).
  - Tesla mentioned that their car crash might be partially caused by the radar trying to filter out abnormal reflection that could be caused by metallic junk / metallic signage.

Now just try to imagine what happens when it's the neural net used to tag objects in the camera that goes miss prediciton.
It will be even harder to understand *why* the rules kicked into action and the autonomous car took a specific decision.

I think the current demos were also done on some insanely high-end Quad GPU configured workstation, that costs around $60k ( https://arstechnica.com/gaming... and https://www.nvidia.com/en-us/d... ).

And even with all of the above hardware, I think they are just running at 24fps?

It looks like this is a long ways off, for a reasonably priced high-end home gaming machine (ie. $4k budget).

More specifically, Nvidia made custom hardware to do the calculations.

I value convenience (as I'm getting old) and I like the large apt package set, lots of stuff pre-packaged and ready to run by a a single command line.

Debian (apt), Fedora (yum), Arch/Manjaro (pacman) or Gentoo (emerge) can do this, likely others as well. I think Debian's apt repo is quite a bit larger than Ubuntu's.

The other options like MINT, Peppermint, Bodhi, CrunchBang, etc are going to have smaller package repos than Debian, if that matters to you. I could probably list all the Linux apps I need on a Post-It note and find them in the vast majority of distros. (vim, gcc/binutils/make, SDL2, Firefox, pidgin, VLC, audacity, MilkyTracker, GIMP, LibreOffice, Scribus)

I like KDE's features and configurability, but don't like the bloat. I've tried XFCE (&Co) on my lo-end machines, like the speed but they lack some features.

KDE and XFCE are available on Debian, Fedora, Arch, Gentoo, Devuan, Slackware, etc. I use WindowMaker + Thunar myself.

I don't really care if I run a BSD or Linux kernel and user space.

FreeBSD, DragonFlyBSD, etc. the BSDs can be a good choice if you can put a little work in and as long as you're not too dependent on running proprietary drivers. (I believe there are NVIDIA drivers for FreeBSD)

Is it time for me to turn to Debian? Or Manjaro? Or... go hard core Arch? Am I too lazy for those?

If you do run Debian, I recommend running from "unstable". You likely won't be happy with how old the software is in "stable".

Arch is intimidating to install for most people. But the instructions are thorough and walk you through every step. Arch is pretty mindless to install if you have a copy of the installation guide on a phone, tablet, another computer or print out. Arch's support for installing packages from source is the best I've seen, typing 'makepkg -i' and you're done. Makes getting the most up-to-date packages very easy, much of the work is already done for you by other users and posted on AUR.

Don't most of the major players in Web Services have Nvidia GPU farms for processing? I'd think they would be putting most of their focus there.

GPU supercomputing cards are ideal for this:
https://www.nvidia.com/en-us/d...

This is not entirely accurate. You have unified memory access in CUDA, and it's been that way for years. The CUDA driver has system-level privileges.

A quick google turned up this NVIDIA blog post. You can dig into the details on CUDA Zone if you're really curious.

Unified memory is also supported in DirectX 12 if the underlying hardware supports it.

In both cases, the driver shuffles data transparently. There are already a lot of attacks that rely on manipulative accesses of memory/cache to ensure that data is being read from desirable locations, so it is conceivable that GPU code could expose kernel memory. After all, the driver that provides memory management would have access to it.

nVidia's brand of Android tablet with game pad (been out for a few years now): https://www.nvidia.com/en-us/s...

I've been using the nVidia Shield Android TV to do in-house game and media streaming from my home office PC to my bedroom TV for a couple years now. I hear the new version even streams games in 4k now.

Since Nvidia is a high profile player, who also happens to design the SOC and VPU inside, you get the best driver support. They constantly update the system, also the Android version.

For a comparison to other contenders (incl. Roku) chech out the list at: https://www.nvidia.com/en-us/s...

Get an Nvidia Shield TV. It is absolutely great for Kodi and Plex and has native apps for all or most streaming services, like Netflix, Prime and Youtube that you mentioned. It also doubles as a Chromecast.

If someone tells you they know how a neural network makes its decisions, they are lying to you.

Nope, you have been misinformed, even I can tell you that!

Neural networks make their decisions by using gradient descent to segment an N-dimensional hyperspace with N-1 dimensional hyperplanes.

Researchers who know how ANNs work have known this for a long time, and can extract more "human readable" explanations from that understanding - it's just that there's never been an impetus to do so before. There is now, so we see researchers actually providing said info now. For example:

NVidia's ANN for self-driving cars

Well there is the TITAN Xp Star Wars Edition. But it's a much lower spec card than TITAN V and more like $1200 retail instead of $5000.

obviously you should run both the dark side and the light side in the same machine to have a proper Star Wars experience.

I see your point, but these specialized GPUs tend to cost much more than a motherboard. To get rid of all the serious bottlenecks, you'll probably be buying a lot of hardware anyway, it's not easy. Check out this device. When people are spending $100k USD on machine learning setups, they're willing to get custom motherboards.

What you've described is essentially nVidia's Quadro line, vs their normal retail/gamer lines: http://www.nvidia.com/object/q...

"Topping out at less than a third the power consumption of a 1080" yeah, except not. Normally I wouldn't even both correcting just another ransom person spouting completely wrong information. But you have been going on and on about it, talking about how you *refuse* to use something so power hungry and inefficient, and how you're so clever cause what you use is *of course* way better perf/watt.

That would be nice and all if it were true, but it's complete bullshit. TDP of a 960 is 120W, per Nvidias site. See for yourself: https://www.geforce.com/hardwa...

The 1080 TDP is 180W, Nvidia significantly improved the power usage of the 10 series compared to the previous 900 series cards for equal or greater performance levels, like the 1080 vs 980 ti. https://www.nvidia.com/en-us/g...

If you were actually right, your continued mentioning of it would be kind of annoying, but that's ok. But since you're actually not even close to right, it's just obnoxious to keep reading the same bullshit spewed out over and over again. So you aren't oddly principled about using only the best perf/watt components, you're just ignorant. Sorry.

NVIDIA's new building is only about 6 miles away from Apple's and it is essentially one giant room with 11.5 acres of cubical. I exaggerate, obviously the full 500,000 sqft of floor space is not cubicles, my rough calculations show it more like 2.75 acres of cubes (6' x 8' * 2500 people).

The switch should have no problem with 1080p, it is using the Tegra X1 chip, which is capable of 4k at 60fps.

This is so painfully wrong that I have to correct it. The Tegra has a hardware decoder that can handle 4K video. It specifically supports the H.265 and VP9 codecs, which are widely used for streaming.

I would be absolutely shocked if the Tegra can render 3D scenes at 4K and 60 Hz, given that high end desktop GPUs pulling 200W have trouble with it. In fact, the Tegra X1 has only 40% of the power of an outdated mid-level GeForce card. The Tegra X1 has 256 CUDA cores compared to the 640 cores in the GeForce 750 Ti.

And Nintendo's patch notes are so vague as to be meaningless. You can speculate on what they did all you want. But based on your gross misunderstanding of the SOC's capabilities, I do not assign much credibility to your speculation.

> I thought that industry primarily used the Quadro line

It really depends on your needs.

For game developers, the artists are probably using Quadro's and programmers the non-Quadro's (GTX) to better match the _actual_ specs of the gamers. i.e. You can probably count on one hand the number of gamers with a Quadro card.

Prices due to VRAM options are all over the place for the Quadro line. Notice how nVidia doesn't list prices on the Quadro line.

If you're doing Modelling / CAD such as Maya / Max / SolidWorks / etc. then yeah, you're probably using a Quadro since the Quadro's prices are a drop in the (price) bucket -- relatively speaking. i.e. The M6000 with 24 GB of VRAM is selling on Amazon for $4,529.00 -- which is *already* discounted !

If not then Quadro's are hideously expensesive (significantly north of $1K) for the "rest of us". i.e. The Quadro P6000 has (had?) a MSRP of $7,000. The average gamer doesn't really "need" more then 6 GB VRAM.

It is kind of like F1 cars. They cost a fortune due to limited supply and demand but technology "trickles" down so us "mere mortals" can afford it. A rule-of-thumb for /oblg. PC Master Race is

* get the fastest single card you can afford
* re-sell it every 2~3 years to re-coop costs
* Don't spend more then $1,000 unless you actually _need_ it.

Performance has always been on a exponential curve. Every time you double (*) the cost your performance gains go down by 1/2 (*).

(*) SWAG. These are NOT actually numbers -- I pulled them out of my ass -- I just used them for illustrative purposes.

--
Microshat, noun, fucking you over with their crap since Winblows 95 by trying to lock you into proprietary technology that is dead within the decade. i.e. How is that Silverlight and XNA working out guys?

From wiki for TOP500:

Since 1993, performance of the #1 ranked position has grown steadily in accord with Moore's law, doubling roughly every 14 months. As of November 2014, Tianhe-2 was fastest with an Rpeak[6] of 54.9024 PFLOPS, is over 419,102 times faster than the fastest system in November 1993, the Connection Machine CM-5/1024 (1024 cores) with Rpeak of 131.0 GFLOPS.[7]

And then there's this article from a company that knows a little bit about parallel processing.

As the trend continues for increases of processing power while cost decreases, the only real question to me is at what point is there critical mass enough where simulation can evolve strong AI. Unless there is a firmly held belief that our consciousness is the result of a spirit entity which is destined for a higher plane of existence after temporarily inhabiting a meat sack, why does it seem so difficult to assume that natural process that created human intelligence could not be reproduced and condensed into a fractional timespan, using software tools similar to Avida.

NVIDIA did some research on that topic a few years back: https://research.nvidia.com/pu...

No idea if the work is continuing, though.

IMHO

You're welcome to your own opinion, but my point is that the opinion of the far-better-informed investors is rather more credible. The number of years you've been hearing about it is not particularly relevant to the feasibility of the product.

the headset they actually have is the size of a helmet

Sure, that's a working prototype - the one that performs the lightfield projections that (AFAIK) no other company has demonstrated anything close to (NVIDIA have showed a low-res display-only system, and Microsoft's Hololens, while impressive in many ways, uses ordinary LCD displays rather than lightfields or holograms). The sunglasses are the production target, and they're still finding the best engineering process to get a reasonable display squeezed into that form. It's not some surprising revelation that they've hit the occasional roadbump along the way.

the demo they posted on youtube over a year ago ... was actually made with film studio special effects

The one with the big Weta logo in the corner? What a shock :-) If you want a reliable picture of what it can currently do, look instead at the videos that actually claim to be shot through the Magic Leap equipment.

there wouldn't be any need for this level of wierd cloak and dagger secrecy

The secrecy was precisely to manage the expectations of the inevitable "hype machine". You may have noticed; the internet has this tendency to blow things out of proportion, both positive and negative.

Incidentally, while I believe they genuinely do have some interesting technology, and am looking forward to what they (eventually) come out with, I'm not expecting magic, nor anything shipped in the next couple of years. For a well-informed and balanced (yet somewhat critical) look at what we know of Magic Leap's technology, the best I've is on Karl Guttag's blog.

For Nvidia (Support for 10-bit and up to 8k video):
https://developer.nvidia.com/n...

For Intel:
https://software.intel.com/en-...

"4th Generation Intel Core processors (Haswell CPU 2- 3.5GHz, 4 Cores): Includes an HEVC Software Decoder capable of real time decode of HEVC 4K streams.
5th Generation Intel Core processors (Broadwell): Supports HEVC 8-bit software/hybrid encode.
6th Generation Intel Core processors (Skylake) Supports hardware accelerated HEVC 8-bit decode and encode."

Many phones, tablets, and PCs have hardware acceleration for VP9. AMD, Nvidia, and Intel support it. By default, Microsoft Edge turns on VP9 support when hardware acceleration is available (you can override it to turn VP9 on all the time).

The "no hardware acceleration" argument is tired. Things have moved on.

Not to mention that updated versions for those cards were still available for recent kernels up to 2012

Deep learning and scientific calculation are also dependent on nVidia GPUs, and I seldom hear anyone suggesting Apple for that use case. Maybe a MacBook for SSH.

Well, maybe if you realized that nVidia supports Apple's F/OSS OpenCL Project on all CUDA-compatible GPUs, you would realize that there may just possibly be people doing things with things that you know nothing about.

I grew up here. I've driven on them a lot. I've also regained control during a "uncontrollable" condition.

And it only got into that uncontrollable position because I screwed up at a previous step. Over corrected a steer, blipped the accelerator too hard. A finely tuned feed back controller would have done a much better job than I did.

Cars have been racing each other on and off track in conditions worse than most roads at speeds your average driver would never be able to manage on the same course.

I think the most interesting thing is it will bring VR to notebooks - most current notebooks doesn't work with VR, even if the GPU is strong enough to support it. Problem with VR on current notebooks:
“The problem is that even if the dedicated card generates an image, the integrated card is what outputs that image to a monitor,” Lyons told me. “With VR, that monitor is your headset. Unfortunately integrated cards just aren’t powerful enough to output images to a VR headset without latency. There are workarounds to make VR work on a laptop with Optimus, but since the HDMI port is connected to the integrated card there is no way to bypass it.”
https://www.rockpapershotgun.c...

Problem solved with geforce 10 series notebooks \o/
https://blogs.nvidia.com/blog/...

https://devblogs.nvidia.com/parallelforall/how-build-gpu-accelerated-research-cluster/

No, because I don't see a point in it. I just run Windows natively to game.

I use VMware quite a lot, and had ESX and Workstation installed until I had to replace my server board and bought one that was approved to work with VMware*

*-except the hard disk controller...the only thing that matters in VMware

So, I am now using Hyper V on my server and can't load the Hyper V management tools along with VMware Workstation, they don't play well together.

http://www.nvidia.com/object/d...

It isn't like it is hard to do GPU Passthrough, there are articles about how to set it up.

And even more important once we see single pass widely implemented it will truly be 90 fps with no confusion.