Mine still takes several seconds to boot android TV just so it can display the HDMI input, even if not connected to internet. It has to be always plugged on the power because if there is a power cut, it needs to boot android TV again.
My old dumb TV did that in a second without booting an entire OS. Next time I need a big screen, it will be a computer monitor.
Even switching to other stuff right after the boot (because the power-on can't be called a simple power-on anymore) the tv is slow.
I recently had the pleasure of interacting with a TV from ~2017 or 2018. God was it slow. Especially loading native apps (Samsung 50"-ish TV)
I like my chromecast. At least that was properly specced. Now if only HDMI and CEC would work like I'd like to :|
Signage TVs are good for this. They're designed to run 24/7 in store windows displaying advertisements or animated menus, so they're a bit pricey, and don't expect any fancy features like HDR, but they've got no smarts whatsoever. What they do have is a slot you can shove your own smart gadget into with a connector that breaks oug power, HDMI etc. which someone has made a Raspberry Pi Compute Module carrier board for, so if you're into, say, Jellyfin, you can make it smart completely under your own control with e.g. libreELEC. Here's a video from Jeff Geerling going into more detail: https://youtu.be/-epPf7D8oMk
Alternatively, if you want HDR and high refresh rates, you're okay with a smallish TV, and you're really willing to splash out, ASUS ROG makes 48" 4K 10-bit gaming monitors for around $1700 US. HDMI is HDMI, you can plug whatever you want into there.
I don't have a TV, but doesn't a smart TV require internet access? Why not just... not give it internet access? Or do they come with their own mobile data plans now meaning you can't even turn off the internet access?
They continually try to get ob the Internet, it's basically malware at this point. The on board SoC is also usually comically underpowered so the menus stutter.
OK, that's really fucked. What the hell? Wait a moment... that means they could turn the use of the TV into a subscription at any time! That's crazy...
Right now it's easier to find projectors without it and a smart os. Before long tho it's gonna be harder to find those without a smart os and AI upscaling
IoT devices are already getting owned at staggering rates. Adding a learning model that currently cannot be secured is absolutely going to happen, and going to cause a whole new large batch of breaches.
A processor that isn't Turing complete isn't a security problem like the TPM you referenced. A TPM includes a CPU. If a processor is Turing complete it's called a CPU.
Is it Turing complete? I don't know. I haven't seen block diagrams that show the computational units have their own cpu.
CPUs also have co processer to speed up floating point operations. That doesn't necessarily make it a security problem.
I would pay for AI-enhanced hardware...but I haven't yet seen anything that AI is enhancing, just an emerging product being tacked on to everything they can for an added premium.
In the 2010s, it was cramming a phone app and wifi into things to try to justify the higher price, while also spying on users in new ways. The device may even a screen for basically no reason.
In the 2020s, those same useless features now with a bit of software with a flashy name that removes even more control from the user, and allows the manufacturer to spy on even further the user.
My Samsung A71 has had devil AI since day one. You know that feature where you can mostly use fingerprint unlock but then once a day or so it ask for the actual passcode for added security. My A71 AI has 100% success rate of picking the most inconvenient time to ask for the passcode instead of letting me do my thing.
It's hardware specifically designed for running AI tasks. Like neural networks.
An NPU, or Neural Processing Unit, is a dedicated processor or processing unit on a larger SoC designed specifically for accelerating neural network operations and AI tasks. Unlike general-purpose CPUs and GPUs, NPUs are optimized for a data-driven parallel computing, making them highly efficient at processing massive multimedia data like videos and images and processing data for neural networks
An NPU, or Neural Processing Unit, is a dedicated processor or processing unit on a larger SoC designed specifically for accelerating neural network operations and AI tasks.
it does seem like a good translator for the less human readable stuff like regex and such. I've dabbled with it a bit but I'm a technical artist and haven't found much use for it in the things I do.
I'm generally opposed to anything that involves buying new hardware. This isn't the 1980s. Computers are powerful as fuck. Stop making software that barely runs on them. If they can't make ai more efficient then fuck it. If they can't make game graphics good without a minimum of a $1000 gpu that produces as much heat as a space heater, maybe we need to go back to 2000s era 3d. There is absolutely no point in making graphics more photorealistic than maybe Skyrim. The route they're going is not sustainable.
The point of software like DLSS is to run stuff better on computers with worse specs than what you'd normally need to run a game as that quality. There's plenty of AI tech that can actually improve experiences and saying that Skyrim graphics are the absolute max we as humanity "need" or "should want" is a weird take ¯\_(ツ)_/¯
The quality of games has dropped a lot, they make them fast and as long as it can just about reach 60fps at 720p they release it. Hardware is insane these days, the games mostly look the same as they did 10 years ago (Skyrim never looked amazing for 2011. BF3, Crysis 2, Forza, Arkham City etc. came out then too), but the performance of them has dropped significantly.
I don't want DLSS and I refuse to buy a game that relies on upscaling to have any meaningful performance. Everything should be over 120fps at this point, way over. But people accept the shit and buy the games up anyway, so nothing is going to change.
The point is, we would rather have games looking like Skyrim with great performance vs '4K RTX real time raytracing ultra AI realistic graphics wow!' at 60fps.
The quality of games has dropped a lot, they make them fast
Isn't the public opinion that games take way too long to make nowadays? They certainly don't make them fast anymore.
As for the rest, I also can't really agree. IMO, graphics have taken a huge jump in recent years, even outside of RT. Lighting, texture quality shaders, as well as object density and variety have been getting a noticeable bump. Other than the occasional dud and awful shader compilation stutter that has plagued many PC games over the last few years (but is getting more awareness now) I'd argue that game performance is pretty good for most games right now.
That's why I see techniques like DLSS/FSR/XeSS/TSR not as crutch, but as just as one of the dozen other rendering shortcuts game engines have accumulated over the years. That said, it's not often we see a new technique deliver such a big performance boost while having almost no visual impact.
Also, who decided that 'we' would rather have games looking like Skyrim? While I do like high FPS very much, I also do like shiny graphics with all the bells and whistles. A Game like 'The Talos Principle 2' for example does hammer the GPU quite a bit on its highest settings, but it certainly delivers in the graphics department. So much so that I've probably spent as much time admiring the highly detailed environments as I did actually solving the puzzles.
Isn't the public opinion that games take way too long to make nowadays? They certainly don't make them fast anymore.
I think the problem here is that they announce them way too early, so people are waiting like 2-3 years for it. It's better if they are developed behind the scenes and 'surprise' announced a few months prior to launch.
Graphics have advanced of course, but it's become diminishing returns and now a lot of games have resorted to spamming post-processing effects and implementing as much foliage and fog as possible to try and make the games look better. I always bring Destiny 2 up in this conversation, because the game looks great, runs great and the graphical fidelity is amazing - no blur but no rough edges. Versus like any UE game which have terrible TAA, if you disable it then everything is jagged and aliased.
DLSS etc are defo a crutch and they are designed as one (originally for real-time raytracing), hence the better versions requiring new hardware. Games shouldn't be relying on them and their trade-offs are not worth it if you have average modern hardware where the games should just run well natively.
It's not so much us wanting specifically Skyrim, maybe that one guy, but just an extreme example I guess to put the point across. It's obviously all subjective, making things shiny obviously attracts peoples eyes during marketing.
I see. That I can mostly agree with. I really don't like the temporal artifacts that come with TAA either, though it's not a deal-breaker for me if the game hides it well.
A few tidbits I'd like to note though:
they announce them way too early, so people are waiting like 2-3 years for it.
Agree. It's kind of insane how early some games are being announced in advance. That said, 2-3 years back then was the time it took for a game to get a sequel. Nowadays you often have to wait an entire console-cycle for a sequel to come out instead of getting a trilogy of games on during one.
Games shouldn’t be relying on them and their trade-offs are not worth it
Which trade-offs are you alluding to? Assuming a halfway decent implementation, DLSS 2+ in particular often yields a better image quality than even native resolution with no visible artifacts, so I turn it on even if my GPU can handle a game just fine, even if just to save a few watts.
Which trade-offs are you alluding to? Assuming a halfway decent implementation, DLSS 2+ in particular often yields a better image quality than even native resolution with no visible artifacts, so I turn it on even if my GPU can handle a game just fine, even if just to save a few watts.
Trade-offs being the artifacts, while not that noticable to most, I did try it and anything in fast motion does suffer. Another being the hardware requirement. I don't mind it existing, I just don't think mid-high end setups should ever have to enable it for a good experience (well, what I personally consider a good experience :D).
I agreed with you in the first half but the rest makes no sense and leads me to believe you aren't aware that most games come with graphics settings. You don't need a $1000 GPU to play 99% of new games...my PC hasn't had a current-gen video card in like 15 years and I've never run into a game I couldn't play with reasonable FPS.
There is absolutely no point in making graphics more photorealistic than maybe Skyrim.
Just because you don't appreciate an aspect of art doesn't mean there is "no point" to it. Graphics come second to gameplay, of course, but do you really not see the appeal in visual realism in games? I was wowed by the ray tracing in Control and it really added to the immersion for me
I figure they're those "early adopters" who buy the New Thing! as soon as it comes out, whether they need it or not, whether it's garbage or not, because they want to be seen as on the cutting edge of technology.
raytracing is something I'd pay for even if unasked, assuming they meaningfully impact the quality and dont demand outlandish prices.
And they'd need to put it in unasked and cooperate with devs else it won't catch on quickly enough.
Remember Nvidia Ansel?
yeah but it didn’t try to lock him into a subscription plan or software ecosystem
Not AI fault, the first one (killed) was a remotely controlled by the product of a big corp (Skynet), the other one was a local, offline one.
Moral of the story: there's difference between the AI that runs locally on your GPU and the one that runs on Elon's remote servers... and that difference may be life or death.
Look, me too, but, the average punter on the street just looks at AI new features and goes OK sure give it to me. Tell them about the dodgy shit that goes with AI and you'll probably get a shrug at most
When they start calling everything AI, soon enough it loses all meaning. They're gonna have to start marketing things as AI-z, AI 2, iAI, AIA, AI 360, AyyyAye, etc. Got their work cut out for em, that's for sure.
Just saying, I’d welcome some competition from other players in the industry. AI-boosted upscaling is a great use of the hardware, as long as it happens on your own hardware only.
I mean, if framegen and supersampling solutions become so good on those chips that regular versions can't compare I guess I would get the AI version. I wouldn't pay extra compared to current pricing though
I'm interested in hardware that can better run local models. Right now the best bet is a GPU, but I'd be interested in a laptop with dedicated chips for AI that would work with pytorch. I'm a novice but I know it takes forever on my current laptop.
Maybe people doing AI development who want the option of running local models.
But baking AI into all consumer hardware is dumb. Very few want it. saas AI is a thing. To the degree saas AI doesn't offer the privacy of local AI, networked local AI on devices you don't fully control offers even less. So it makes no sense for people who value convenience. It offers no value for people who want privacy. It only offers value to people doing software development who need more playground options, and I can go buy a graphics card myself thank you very much.
I would if the hardware was powerful enough to do interesting or useful things, and there was software that did interesting or useful things. Like, I'd rather run an AI model to remove backgrounds from images or upscale locally, than to send images to Adobe servers (this is just an example, I don't use Adobe products and don't know if this is what Adobe does). I'd also rather do OCR locally and quickly than send it to a server. Same with translations. There are a lot of use-cases for "AI" models.
Okay, but here me out. What if the OS got way worse, and then I told you that paying me for the AI feature would restore it to a near-baseline level of original performance? What then, eh?
I would pay extra to be able to run open LLM's locally on Linux. I wouldn't pay for Microsoft's Copilot stuff that's shoehorned into every interface imaginable while also causing privacy and security issues. The context matters.
That's why NPU's are actually a good thing. The ability to run LLM local instead of sending everything to Microsoft/Open AI for data mining will be great.
I hate to be that guy, but do you REALLY think that on-device AI is going to prevent all your shit being sent to anyone who wants it, in the form of "diagnostic data" or "usage telemetry" or whatever weasel-worded bullshit in the terms of service?'
They'll just send the results for "quality assurance" instead of doing the math themselves and save a bundle on server hosting.
All your unattended date will be taken (and some of the attended one). This doesn't mean you should stop to attend your data.
Even of you're somehow forced to use Windows instead open alternative, it doesn't mean you can't dual boot or use other privacy conscious devices when dealing with your sensitive data.
Closed/proprietary OS and hardware driver can't be considered safe by design)
I completely agree. There are some killer AI apps, but why should AI run on my OS? Recall is a complete disaster of a product and I hope it doesn't see the light of day, but I've no doubt that there's a place for AI on the PC.
Whatever application there is in AI at the OS level, it needs to be a trustless system that the user has complete control of. I'd be all for an Open source AI running at that level, but Microsoft is not going to do that because they want to ensure that they control your OS data.
I think it could definitely automate some roles where you aren't necessarily thinking and all decisions are made based on information internally available to the PC.
For sure these exist but some decisions need human input, I'm not sure how they automate out those roles just because they see stuff happening on the PC every day.
If anything I think this feature is used to spy on users at work and see when keystrokes fall below a certain level each day, but I'm sure that's already possible for companies to do (but they just don't).
I can't tell how good any of this stuff is because none of the language they're using to describe performance makes sense in comparison with running AI models on a GPU. How big a model can this stuff run, how does it compare to the graphics cards people use for AI now?
A big letdown for me is, except with some rare cases, those extra AI features useless outside of AI. Some NPUs are straight out DSPs, they could easily run OpenCL code, others are either designed to not be able to handle any normal floating point numbers but only ones designed for machine learning, or CPU extensions that are just even bigger vector multipliers for select datatypes (AMX).
Even DLSS only works great for some types of games.
Although there have been some clever uses of it, lots of games could gain a lot from proper efficiency of the game engine.
War Thunder runs like total crap on even the highest end hardware, yet World of Warships has much more detailed ships and textures running fine off an HDD and older than GTX 7XX graphics.
Meanwhile on Linux, Compiz still runs crazy window effects and 3D cube desktop much better and faster than KDE. It's so good I even recommend it for old devices with any kid of gpu because the hardware acceleration will make your desktop fast and responsive compared to even the lightest windows managers like openbox.
TF2 went from 32 bit to 64 bit and had immediate gains in performance upwards of 50% and almost entirely removing stuttering issues from the game.
Batman Arkham Knight ran on a heavily modified version of Unreal 3 which was insane for the time.
Most modern games and applications really don't need the latest and greatest hardware, they just need to be efficiently programmed which is sometimes almost an art itself. Slapping on "AI" to reduce the work is sort of a lazy solution that will have side effects because you're effectively predicting the output.
When a decent gpu is ~$1k alone, then someone wants you to pay more $ for a feature that offers no tangible benefit, why the hell would they want it? I haven’t bought a PC for over 25 years, I build my own and for family and friends. I’m building another next week for family, and AI isn’t even on the radar. If anything, this one is going to be anti-AI and get a Linux dual-boot as well as sticking with Win10, no way am I subjecting family to that Win11 adware.
I'm fine with NPUs / TPUs (AI-enhancing hardware) being included with systems because it's useful for more than just OS shenanigans and commercial generative AI. Do I want Microsoft CoPilot Recall running on that hardware? No.
However I've bought TPUs for things like Frigate servers and various ML projects. For gamers there's some really cool use cases out there for using local LLMs to generate NPC responses in RPGs. For "Smart Home" enthusiasts things like Home Assistant will be rolling out support for local LLMs later this year to make voice commands more context aware.
So do I want that hardware in there so I can use it MYSELF for other things? Yes, yes I do. You probably will eventually too.
I wish someone would make software that utilizes things like a M.2 coral TPU, to enhance gameplay like with frame gen, or up scaling for games and videos. Some GPUs are starting to even put M.2 slots on the GPU, if the latency from Mobo M.2 to PCIe GPU would be too slow.
As with any proprietary hardware on a GPU it all comes down to third party software support and classically if the market isn't there then it's not supported.
Assuming theres no catch-on after 3-4 cycles I'd say the tech is either not mature enough, too expensive with too little results or (as you said) theres generally no interest in that.
Maybe it needs a bit of marturing and a re-introduction at a later point.
I could see use for local text gen, but that apparently eats quite a bit more than what desktop PCs could offer if you want to have some actually good results & speed. Generally though, I'd rather want separate extension cards for this. Making it part of other processors is just going to increase their price, even for those who have no use for it.
Yes, I know - that's my point. But you need the necessary hardware to run those models in a performative way. Waiting a minute to produce some vaguely relevant gibberish is not going to be of much use. You could also use generative text for other applications, such as video game NPCs, especially all those otherwise useless drones you see in a lot of open world titles could gain a lot of depth.
Well yeah, because dedicated DACs have a tangible benefit of better audio. If you want better audio you need to buy a quality DAC and quality cans.
I also used to think it's dumb because who cares as long as you can hear. But then I built a new PC and I don't know if it was a faulty mobo or just unlucky setup but the internal DAC started picking up static. So I got an external DAC and what I noticed was that the audio sounded clearer and I could hear things in the sound that I couldn't hear before. It was magical, it's like someone added new layers into my favorite songs. I had taken the audio crack.
I pretty quickly gave away my DAC along with my audio technicas because I could feel the urge. I needed another hit. I needed more. I got this knawing itch and I knew I had to get out before the addiction completely took over. Now I live in static because I do not dare to touch the sun again.
Soundblasters may be shit but the hardware they're supposed to sell is legit, it has a tangible benefit to whomever can tell the difference. But with AI, what is the tangible benefit that you couldn't get by getting a better GPU?
I would already like to buy a 4k TV that isn't smart and have yet to find it. Please don't add AI into the mix as well :(
Look into commercial displays
The simple trick to turn a "smart" TV into a regular one is too cut off its internet access.
Except it will still run like shit and may send telemetry via other means to your neighbors same brand TV
I've never heard of that. Do you have a source on that? And how would it run like shit if you're using something like a Chromecast?
I don't know about the telemetry, but my smart tv runs like shit after being on for a few hours. Only a full power cycle makes it work properly again.
https://www.amazon.com/Amazon-Sidewalk/b?ie=UTF8&node=21328123011
https://aws.amazon.com/iot-core/features/
Mine still takes several seconds to boot android TV just so it can display the HDMI input, even if not connected to internet. It has to be always plugged on the power because if there is a power cut, it needs to boot android TV again.
My old dumb TV did that in a second without booting an entire OS. Next time I need a big screen, it will be a computer monitor.
Still uses the shitty 'smart' operating system to handle inputs and settings.
I just bought a commercial display directly from the Bengal stadium. Still has Wi-Fi.
I was just thinking the other day how I'd love to "root" my TV like I used to root my phones. Maybe install some free OS instead
You can if you have a pre-2022 LG TV. It's more akin to jailbreaking since you can't install a custom OS, but it does give you more control.
https://rootmy.tv
We got a Sceptre brand TV from Walmart a few years ago that does the trick. 4k, 50 inch, no smart features.
All TVs are dumb TVs if they have no internet access
I just disconnected my smart TV from the internet. Nice and dumb.
Still slow UI.
If only signage displays would have the fidelity of a regular OLED consumer without the business-usage tax on top.
What do you use the UI for? I just turn my TV on and off. No user interface needed. Only a power button on the remote.
Even switching to other stuff right after the boot (because the power-on can't be called a simple power-on anymore) the tv is slow.
I recently had the pleasure of interacting with a TV from ~2017 or 2018. God was it slow. Especially loading native apps (Samsung 50"-ish TV)
I like my chromecast. At least that was properly specced. Now if only HDMI and CEC would work like I'd like to :|
Signage TVs are good for this. They're designed to run 24/7 in store windows displaying advertisements or animated menus, so they're a bit pricey, and don't expect any fancy features like HDR, but they've got no smarts whatsoever. What they do have is a slot you can shove your own smart gadget into with a connector that breaks oug power, HDMI etc. which someone has made a Raspberry Pi Compute Module carrier board for, so if you're into, say, Jellyfin, you can make it smart completely under your own control with e.g. libreELEC. Here's a video from Jeff Geerling going into more detail: https://youtu.be/-epPf7D8oMk
Alternatively, if you want HDR and high refresh rates, you're okay with a smallish TV, and you're really willing to splash out, ASUS ROG makes 48" 4K 10-bit gaming monitors for around $1700 US. HDMI is HDMI, you can plug whatever you want into there.
I don't have a TV, but doesn't a smart TV require internet access? Why not just... not give it internet access? Or do they come with their own mobile data plans now meaning you can't even turn off the internet access?
Anti Commercial-AI license
They continually try to get ob the Internet, it's basically malware at this point. The on board SoC is also usually comically underpowered so the menus stutter.
I never needed a TV, but now I for sure am not getting one.
Anti Commercial-AI license
IDK why people are downvoting you, I am sure you're not alone with that sentiment.
A lot of TVs are requiring an account login before being able to use it.
OK, that's really fucked. What the hell? Wait a moment... that means they could turn the use of the TV into a subscription at any time! That's crazy...
Anti Commercial-AI license
I'm sure that's coming up.
Right now it's easier to find projectors without it and a smart os. Before long tho it's gonna be harder to find those without a smart os and AI upscaling
The dedicated TPM chip is already being used for side-channel attacks. A new processor running arbitrary code would be a black hat's wet dream.
It will be.
IoT devices are already getting owned at staggering rates. Adding a learning model that currently cannot be secured is absolutely going to happen, and going to cause a whole new large batch of breaches.
The “s” in IoT stands for “security”
Do you have an article on that handy? I like reading about side channel and timing attacks.
TPM-FAIL from 2019. It affects Intel fTPM and some dedicated TPM chips: link
The latest (at the moment) UEFI vulnerability, UEFIcanhazbufferoverflow is also related to, but not directly caused by, TPM on Intel systems: link
That's insane. How can they be doing security hardware and leave a timing attack in there?
Thank you for those links, really interesting stuff.
It's not a full CPU. It's more limited than GPU.
That's why I wrote "processor" and not CPU.
A processor that isn't Turing complete isn't a security problem like the TPM you referenced. A TPM includes a CPU. If a processor is Turing complete it's called a CPU.
Is it Turing complete? I don't know. I haven't seen block diagrams that show the computational units have their own cpu.
CPUs also have co processer to speed up floating point operations. That doesn't necessarily make it a security problem.
I would pay for AI-enhanced hardware...but I haven't yet seen anything that AI is enhancing, just an emerging product being tacked on to everything they can for an added premium.
In the 2010s, it was cramming a phone app and wifi into things to try to justify the higher price, while also spying on users in new ways. The device may even a screen for basically no reason.
In the 2020s, those same useless features now with a bit of software with a flashy name that removes even more control from the user, and allows the manufacturer to spy on even further the user.
It's like rgb all over again.
At least rgb didn't make a giant stock market bubble...
Anything AI actually enhanced would be advertising the enhancement not the AI part.
My Samsung A71 has had devil AI since day one. You know that feature where you can mostly use fingerprint unlock but then once a day or so it ask for the actual passcode for added security. My A71 AI has 100% success rate of picking the most inconvenient time to ask for the passcode instead of letting me do my thing.
DLSS and XeSS (XMX) are AI and they're noticably better than non-hardware accelerated alternatives.
Already had that Google thingy for years now. The USB/nvme device for image recognition. Can't remember what it's called now. Cost like $30.
Edit: Google coral TPU
I use it heavily at work nowadays. It would be nice to run it locally.
You don't need AI enhanced hardware for that, just normal ass hardware and you run AI software on it.
But you can run more complex networks faster. Which is what I want.
Maybe I'm just not understanding what AI-enabled hardware is even supposed to mean
It's hardware specifically designed for running AI tasks. Like neural networks.
https://github.com/huggingface/candle
You can look into this, however it’s not what this discussion is about
Exactly what we are talking about.
Stick to the discussion of paying a premium for hardware not the software
Not sure what you mean? The hardware runs the software tasks more efficiently.
The discussion is whether people should/would pay extra for hardware designed around ai vs just getting better hardware
I'm curious what you use it for at work.
Not the guy you were asking but it's great for writing powershell scripts
I'm a programmer so when learning a new framework or library I use it as an interactive docs that allows follow up questions.
I also use it to generate things like regex and SQL queries.
It's also really good at refactoring code and other repetitive tasks like that
it does seem like a good translator for the less human readable stuff like regex and such. I've dabbled with it a bit but I'm a technical artist and haven't found much use for it in the things I do.
I'm generally opposed to anything that involves buying new hardware. This isn't the 1980s. Computers are powerful as fuck. Stop making software that barely runs on them. If they can't make ai more efficient then fuck it. If they can't make game graphics good without a minimum of a $1000 gpu that produces as much heat as a space heater, maybe we need to go back to 2000s era 3d. There is absolutely no point in making graphics more photorealistic than maybe Skyrim. The route they're going is not sustainable.
The point of software like DLSS is to run stuff better on computers with worse specs than what you'd normally need to run a game as that quality. There's plenty of AI tech that can actually improve experiences and saying that Skyrim graphics are the absolute max we as humanity "need" or "should want" is a weird take ¯\_(ツ)_/¯
The quality of games has dropped a lot, they make them fast and as long as it can just about reach 60fps at 720p they release it. Hardware is insane these days, the games mostly look the same as they did 10 years ago (Skyrim never looked amazing for 2011. BF3, Crysis 2, Forza, Arkham City etc. came out then too), but the performance of them has dropped significantly.
I don't want DLSS and I refuse to buy a game that relies on upscaling to have any meaningful performance. Everything should be over 120fps at this point, way over. But people accept the shit and buy the games up anyway, so nothing is going to change.
The point is, we would rather have games looking like Skyrim with great performance vs '4K RTX real time raytracing ultra AI realistic graphics wow!' at 60fps.
Isn't the public opinion that games take way too long to make nowadays? They certainly don't make them fast anymore.
As for the rest, I also can't really agree. IMO, graphics have taken a huge jump in recent years, even outside of RT. Lighting, texture quality shaders, as well as object density and variety have been getting a noticeable bump. Other than the occasional dud and awful shader compilation stutter that has plagued many PC games over the last few years (but is getting more awareness now) I'd argue that game performance is pretty good for most games right now.
That's why I see techniques like DLSS/FSR/XeSS/TSR not as crutch, but as just as one of the dozen other rendering shortcuts game engines have accumulated over the years. That said, it's not often we see a new technique deliver such a big performance boost while having almost no visual impact.
Also, who decided that 'we' would rather have games looking like Skyrim? While I do like high FPS very much, I also do like shiny graphics with all the bells and whistles. A Game like 'The Talos Principle 2' for example does hammer the GPU quite a bit on its highest settings, but it certainly delivers in the graphics department. So much so that I've probably spent as much time admiring the highly detailed environments as I did actually solving the puzzles.
I think the problem here is that they announce them way too early, so people are waiting like 2-3 years for it. It's better if they are developed behind the scenes and 'surprise' announced a few months prior to launch.
Graphics have advanced of course, but it's become diminishing returns and now a lot of games have resorted to spamming post-processing effects and implementing as much foliage and fog as possible to try and make the games look better. I always bring Destiny 2 up in this conversation, because the game looks great, runs great and the graphical fidelity is amazing - no blur but no rough edges. Versus like any UE game which have terrible TAA, if you disable it then everything is jagged and aliased.
DLSS etc are defo a crutch and they are designed as one (originally for real-time raytracing), hence the better versions requiring new hardware. Games shouldn't be relying on them and their trade-offs are not worth it if you have average modern hardware where the games should just run well natively.
It's not so much us wanting specifically Skyrim, maybe that one guy, but just an extreme example I guess to put the point across. It's obviously all subjective, making things shiny obviously attracts peoples eyes during marketing.
I see. That I can mostly agree with. I really don't like the temporal artifacts that come with TAA either, though it's not a deal-breaker for me if the game hides it well.
A few tidbits I'd like to note though:
Agree. It's kind of insane how early some games are being announced in advance. That said, 2-3 years back then was the time it took for a game to get a sequel. Nowadays you often have to wait an entire console-cycle for a sequel to come out instead of getting a trilogy of games on during one.
Which trade-offs are you alluding to? Assuming a halfway decent implementation, DLSS 2+ in particular often yields a better image quality than even native resolution with no visible artifacts, so I turn it on even if my GPU can handle a game just fine, even if just to save a few watts.
Trade-offs being the artifacts, while not that noticable to most, I did try it and anything in fast motion does suffer. Another being the hardware requirement. I don't mind it existing, I just don't think mid-high end setups should ever have to enable it for a good experience (well, what I personally consider a good experience :D).
We should have stopped with Mario 64. Everything else has been an abomination.
I agreed with you in the first half but the rest makes no sense and leads me to believe you aren't aware that most games come with graphics settings. You don't need a $1000 GPU to play 99% of new games...my PC hasn't had a current-gen video card in like 15 years and I've never run into a game I couldn't play with reasonable FPS.
Just because you don't appreciate an aspect of art doesn't mean there is "no point" to it. Graphics come second to gameplay, of course, but do you really not see the appeal in visual realism in games? I was wowed by the ray tracing in Control and it really added to the immersion for me
Running my 1080 + i7-6700 into the ground. Playing 10+ yr old games i never got around to like a G
Cum play Old school RuneScape
Or they know a guy named Al and got confused. ;)
Maybe I'm in the minority here, but I'd gladly pay more for Weird Al enhanced hardware.
Hardware breaks into a parody of whatever you are doing
Me - laughing and vibing
A man walks down the street He says why am I short of attention Got a short little span of attention And woe my nights are so long
I figure they're those "early adopters" who buy the New Thing! as soon as it comes out, whether they need it or not, whether it's garbage or not, because they want to be seen as on the cutting edge of technology.
I am generally unwilling to pay extra for features I don't need and didn't ask for.
raytracing is something I'd pay for even if unasked, assuming they meaningfully impact the quality and dont demand outlandish prices.
And they'd need to put it in unasked and cooperate with devs else it won't catch on quickly enough.
Remember Nvidia Ansel?
We're not gonna make it, are we? People, I mean.
Didn't John Connor befriend the second IA he find?
yeah but it didn't try to lock him into a subscription plan or software ecosystem
It locked him into the world of the terminators? Imo a mighty subscription
/j
Not AI fault, the first one (killed) was a remotely controlled by the product of a big corp (Skynet), the other one was a local, offline one.
Moral of the story: there's difference between the AI that runs locally on your GPU and the one that runs on Elon's remote servers... and that difference may be life or death.
I was recently looking for a new laptop and I actively avoided laptops with AI features.
Look, me too, but, the average punter on the street just looks at AI new features and goes OK sure give it to me. Tell them about the dodgy shit that goes with AI and you'll probably get a shrug at most
What does AI enhanced hardware mean? Because I bought an Nvidia RTX card pretty much just for the AI enhanced DLSS, and I’d do it again.
When they start calling everything AI, soon enough it loses all meaning. They're gonna have to start marketing things as AI-z, AI 2, iAI, AIA, AI 360, AyyyAye, etc. Got their work cut out for em, that's for sure.
Instead of Nvidia knowing some of your habits, they will know most of your habits. $$$.
Just saying, I’d welcome some competition from other players in the industry. AI-boosted upscaling is a great use of the hardware, as long as it happens on your own hardware only.
Let me put it in lamens terms..... FUCK AI.... Thanks, have a great day
FYI the term is "layman's", as of you were using the language of a layman, or someone who is not specifically experienced in the topic.
Sounds like something a lameman would say
Well, when life hands you lémons...
The biggest surprise here is that as many as 16% are willing to pay more...
I mean, if framegen and supersampling solutions become so good on those chips that regular versions can't compare I guess I would get the AI version. I wouldn't pay extra compared to current pricing though
Who in the heck are the 16%
The ones who have investments in AI
The ones who listen to the marketing
The ones who are big Weird Al fans
The ones who didn't understand the question
I would pay for Weird-Al enhanced PC hardware.
Those Weird Al fans will be very disappointed
I'm interested in hardware that can better run local models. Right now the best bet is a GPU, but I'd be interested in a laptop with dedicated chips for AI that would work with pytorch. I'm a novice but I know it takes forever on my current laptop.
Not interested in running copilot better though.
Maybe people doing AI development who want the option of running local models.
But baking AI into all consumer hardware is dumb. Very few want it. saas AI is a thing. To the degree saas AI doesn't offer the privacy of local AI, networked local AI on devices you don't fully control offers even less. So it makes no sense for people who value convenience. It offers no value for people who want privacy. It only offers value to people doing software development who need more playground options, and I can go buy a graphics card myself thank you very much.
I would if the hardware was powerful enough to do interesting or useful things, and there was software that did interesting or useful things. Like, I'd rather run an AI model to remove backgrounds from images or upscale locally, than to send images to Adobe servers (this is just an example, I don't use Adobe products and don't know if this is what Adobe does). I'd also rather do OCR locally and quickly than send it to a server. Same with translations. There are a lot of use-cases for "AI" models.
Okay, but here me out. What if the OS got way worse, and then I told you that paying me for the AI feature would restore it to a near-baseline level of original performance? What then, eh?
I already moved to Linux. Windows is basically doing this already.
One word. Linux.
I don't think the poll question was well made... "would you like part away from your money for..." vaguely shakes hand in air "...ai?"
People is already paying for "ai" even before chatGPT came out to popularize things: DLSS
Pay more for a shitty chargpt clone in your operating system that can get exploited to hack your device. I see no flaw in this at all.
I would pay extra to be able to run open LLM's locally on Linux. I wouldn't pay for Microsoft's Copilot stuff that's shoehorned into every interface imaginable while also causing privacy and security issues. The context matters.
That's why NPU's are actually a good thing. The ability to run LLM local instead of sending everything to Microsoft/Open AI for data mining will be great.
I hate to be that guy, but do you REALLY think that on-device AI is going to prevent all your shit being sent to anyone who wants it, in the form of "diagnostic data" or "usage telemetry" or whatever weasel-worded bullshit in the terms of service?'
They'll just send the results for "quality assurance" instead of doing the math themselves and save a bundle on server hosting.
Yes, obviously, especially if you are running all open source software.
All your unattended date will be taken (and some of the attended one). This doesn't mean you should stop to attend your data. Even of you're somehow forced to use Windows instead open alternative, it doesn't mean you can't dual boot or use other privacy conscious devices when dealing with your sensitive data.
Closed/proprietary OS and hardware driver can't be considered safe by design)
I replied to the person above "locally on Linux".
Even in Windows, local queries give the possibility of control. Set your firewall and it cannot leak.
I'm willing to pay extra for software that isn't
My old ass GTX 1060 runs some of the open source language models. I imagine the more recent cards would handle them easily.
What’s the “AI” hardware supposed to do that any gamer with recent hardware can’t?
Run it faster.
A CPU can also compute graphics but you wait significant more time than using hardware accelerated graphics hardware.
Fuck, they won't upgrade to TPM for windows 11
Still havent turned mine on, don't want no surprises after a long day at work
I completely agree. There are some killer AI apps, but why should AI run on my OS? Recall is a complete disaster of a product and I hope it doesn't see the light of day, but I've no doubt that there's a place for AI on the PC.
Whatever application there is in AI at the OS level, it needs to be a trustless system that the user has complete control of. I'd be all for an Open source AI running at that level, but Microsoft is not going to do that because they want to ensure that they control your OS data.
I think it could definitely automate some roles where you aren't necessarily thinking and all decisions are made based on information internally available to the PC. For sure these exist but some decisions need human input, I'm not sure how they automate out those roles just because they see stuff happening on the PC every day.
If anything I think this feature is used to spy on users at work and see when keystrokes fall below a certain level each day, but I'm sure that's already possible for companies to do (but they just don't).
You borked your link
Here is the article.
That still needs an FPGA. While they certainly seems to be able to use smaller ones, adding an FPGA chip will still add cost
I can't tell how good any of this stuff is because none of the language they're using to describe performance makes sense in comparison with running AI models on a GPU. How big a model can this stuff run, how does it compare to the graphics cards people use for AI now?
A big letdown for me is, except with some rare cases, those extra AI features useless outside of AI. Some NPUs are straight out DSPs, they could easily run OpenCL code, others are either designed to not be able to handle any normal floating point numbers but only ones designed for machine learning, or CPU extensions that are just even bigger vector multipliers for select datatypes (AMX).
Even DLSS only works great for some types of games.
Although there have been some clever uses of it, lots of games could gain a lot from proper efficiency of the game engine.
War Thunder runs like total crap on even the highest end hardware, yet World of Warships has much more detailed ships and textures running fine off an HDD and older than GTX 7XX graphics.
Meanwhile on Linux, Compiz still runs crazy window effects and 3D cube desktop much better and faster than KDE. It's so good I even recommend it for old devices with any kid of gpu because the hardware acceleration will make your desktop fast and responsive compared to even the lightest windows managers like openbox.
TF2 went from 32 bit to 64 bit and had immediate gains in performance upwards of 50% and almost entirely removing stuttering issues from the game.
Batman Arkham Knight ran on a heavily modified version of Unreal 3 which was insane for the time.
Most modern games and applications really don't need the latest and greatest hardware, they just need to be efficiently programmed which is sometimes almost an art itself. Slapping on "AI" to reduce the work is sort of a lazy solution that will have side effects because you're effectively predicting the output.
When a decent gpu is ~$1k alone, then someone wants you to pay more $ for a feature that offers no tangible benefit, why the hell would they want it? I haven’t bought a PC for over 25 years, I build my own and for family and friends. I’m building another next week for family, and AI isn’t even on the radar. If anything, this one is going to be anti-AI and get a Linux dual-boot as well as sticking with Win10, no way am I subjecting family to that Win11 adware.
The other 16% do not know what AI is or try to sell it. A combination of both is possible. And likely.
Bro, just add it to the pile of rubbish over there next to the 3D movies and curved TVs
I'm fine with NPUs / TPUs (AI-enhancing hardware) being included with systems because it's useful for more than just OS shenanigans and commercial generative AI. Do I want Microsoft CoPilot Recall running on that hardware? No.
However I've bought TPUs for things like Frigate servers and various ML projects. For gamers there's some really cool use cases out there for using local LLMs to generate NPC responses in RPGs. For "Smart Home" enthusiasts things like Home Assistant will be rolling out support for local LLMs later this year to make voice commands more context aware.
So do I want that hardware in there so I can use it MYSELF for other things? Yes, yes I do. You probably will eventually too.
I wish someone would make software that utilizes things like a M.2 coral TPU, to enhance gameplay like with frame gen, or up scaling for games and videos. Some GPUs are starting to even put M.2 slots on the GPU, if the latency from Mobo M.2 to PCIe GPU would be too slow.
As with any proprietary hardware on a GPU it all comes down to third party software support and classically if the market isn't there then it's not supported.
Assuming theres no catch-on after 3-4 cycles I'd say the tech is either not mature enough, too expensive with too little results or (as you said) theres generally no interest in that.
Maybe it needs a bit of marturing and a re-introduction at a later point.
Unless you're doing music or graphics design there's no usecase. And if you do, you probably have high end GPU anyway
I could see use for local text gen, but that apparently eats quite a bit more than what desktop PCs could offer if you want to have some actually good results & speed. Generally though, I'd rather want separate extension cards for this. Making it part of other processors is just going to increase their price, even for those who have no use for it.
There are local models for text gen - not as good as chatGPT but at the same time they're uncensored - so it may or may not be useful
Yes, I know - that's my point. But you need the necessary hardware to run those models in a performative way. Waiting a minute to produce some vaguely relevant gibberish is not going to be of much use. You could also use generative text for other applications, such as video game NPCs, especially all those otherwise useless drones you see in a lot of open world titles could gain a lot of depth.
Not even on my phone
AI-en{hanced,shittified}
Just need the right name for it. Soundblasters are still being produced aren't they? There's always a market.
Well yeah, because dedicated DACs have a tangible benefit of better audio. If you want better audio you need to buy a quality DAC and quality cans.
I also used to think it's dumb because who cares as long as you can hear. But then I built a new PC and I don't know if it was a faulty mobo or just unlucky setup but the internal DAC started picking up static. So I got an external DAC and what I noticed was that the audio sounded clearer and I could hear things in the sound that I couldn't hear before. It was magical, it's like someone added new layers into my favorite songs. I had taken the audio crack.
I pretty quickly gave away my DAC along with my audio technicas because I could feel the urge. I needed another hit. I needed more. I got this knawing itch and I knew I had to get out before the addiction completely took over. Now I live in static because I do not dare to touch the sun again.
Soundblasters may be shit but the hardware they're supposed to sell is legit, it has a tangible benefit to whomever can tell the difference. But with AI, what is the tangible benefit that you couldn't get by getting a better GPU?
And the other 16% would also pay you $230 to hit them in the face with a shovel
Tbh this is probably for things like DLSS, captions, etc. Not necessarily for chatbots or generative art.
The only reason I have any enthusiasm about CoPilot+ PCs (AI PCs or whatever new name they get in 6 months) is because of ARM and battery life.
Heck, I'll trade them all the AI features for no ads.
Predictable outcome, common tech company L.
I feel like the sarcasm was pretty obvious in that comment, but maybe I'm missing something.