|
Llamadeus posted:IMO it's a short-term problem for AMD anyway, in the long-term they'll have to develop a direct DLSS competitor using ML on specialized hardware. Or at least they should. I hope so. That's the crazy part for me; we should be rooting for AMD because it's a total net win for everyone if they gain more market share, it just kind of sucks that we can't do that in this case because it's so underhanded  wargames posted:So me on the charts where nvidia performs worse than amd. He didn't say that? He said they paid to have it run worse than it otherwise could, which is objectively true...
|
#
?
Aug 18, 2023 17:43
|
|
|
|
| # ? Jun 4, 2024 07:46 |
|
|
Llamadeus posted:IMO it's a short-term problem for AMD anyway, in the long-term they'll have to develop a direct DLSS competitor using ML on specialized hardware. Or at least they should. I think their current approach is the correct one, FSR2 works on drat near every AMD gpu/igpu from the steamdeck to old vega cards, limiting the upscaler to only the newest dedicated hardware wouldn't help AMD that much, the igpu market takes a very long time to get the latest hardware stack from the dedicated gpu side of things.
|
|
#
?
Aug 18, 2023 17:46
|
|
|
power crystals posted:Isn't XeSS also vendor agnostic? The ideal is still that it's not controlled by one of the GPU hardware vendors regardless of being able to run on anybody's cards. XeSS has an Intel path and a generalized path. The Intel path performs WAY better (in terms of IQ). It could probably be adapted to run on Nvidia GPUs but why would Intel do extra work to make a worse version of DLSS? AMD is basically trying to disguise how bad an idea it is for them to be super focused on raster at the expense of RT and ML performance by kneecapping the use of those things in games. At some point they'll probably get competent ML hardware and shortly after a more unified solution will materialize out of nowhere, having absolutely nothing to do with AMD no longer being at a significant hardware advantage. K8.0 fucked around with this message at 17:54 on Aug 18, 2023 |
|
#
?
Aug 18, 2023 17:46
|
|
|
K8.0 posted:Nvidia has done some bullshit but your gaming-related examples are awful. And GPU compute... I mean everyone else has themselves to blame for letting Nvidia build a position of total dominance because they were too lazy and stupid to see that GPU compute would inevitably keep growing bigger and bigger. I have been saying this for years but it's baffling to me that people didn't understand what Jensen was saying about NVIDIA being a software company. People were laughing at it on social media as late as like 2016-2018 when it was obvious that it had shaped every part of their strategy including both GPGPU and gaming and everything else. Haha no jensen you make GPUs, lay off the coke and get back to loving work and make Turing not suck!!! this segment of a talk Jensen gave is so key to understanding how NVIDIA works. The customer doesn't know what they want, you have to have a vision and then show them what it could be like, and pick technical directions that get them to the things they want. If you asked IBM in 2005 if they wanted to do compute on GPUs they would have said no, and they did! Gsync worked the same way. Nobody knew how important framepacing was, NVIDIA showed them, and then dumped a tech which completely changed the game on framepacing. Same for DLSS, really. Even today people don't want sub-native rendering but it does produce big gains in asymptotic performance-per-transistor and in a post-moores law world that's how you keep going. And customers do want faster products even if they don't like the mechanism. https://www.youtube.com/watch?v=Xn1EsFe7snQ&t=1034s he's right that graphics is an insatiable market, you could 10x performance and the market would eat it all up, there is never a "good enough" in computer graphics. Even today, if you 10x'd performance we'd throw away raster entirely and start playing with pathtraced everything, and so on. And moore's law was a lever that let them relentlessly push the bounds of what they could do, and then provide the solutions that implement that new feature. dude is like a steve jobs level businessman and I mean that in both the positive and negative contexts. Not too many OG tech CEOs still leading their company 30+ years after it was founded. And the reason is his ability to see what the tech could be and what it could mean, and then build the tech to get it there. jensen also directly lays out the nature of his business: he's a middleman between TSMC and the customer, he is just slinging transistors that TSMC's node shrinks provide. And when that treadmill stopped (after the video) it's pretty obvious in hindsight that DLSS is the pivot to keep going afterwards. More performance from less transistors is a tough battle. The easy gains were squeezed a long time ago with maxwell, and actually some of those gains have had to be rolled back because things like software scheduling weren't sustainable. Paul MaudDib fucked around with this message at 18:05 on Aug 18, 2023 |
|
#
?
Aug 18, 2023 17:50
|
|
|
K8.0 posted:XeSS has an Intel path and a generalized path. The Intel path performs WAY better. It could probably be adapted to run on Nvidia GPUs but why would Intel do extra work to make a worse version of DLSS? ML performance on RDNA3 is very competitive with the Nvidia cards in the same price tier, what are you talking about? AMD making an AMD specific upscaler would be silly given their market position.
|
|
#
?
Aug 18, 2023 17:57
|
|
|
It is and it isn't. FSR and the generic XeSS path have to balance the use of their GPU to do the compute for upsampling, because it's taking away from render performance. Nvidia does work on the tensor cores for it and because of that DLSS can do a ton more work without a significant hit to rendering performance, and AFAIK Intel has similar but not quite as separate ML optimization in their compute so they can do much more efficient ML work without giving up a huge chunk of rendering performance. Two pieces of hardware could bench identically in two separate loads, but wildly differently if you run both those loads at the same time, and that's a significant part of why AMD can't compete in upsampling right now.
|
|
#
?
Aug 18, 2023 17:59
|
|
|
Yudo posted:ML performance on RDNA3 is very competitive with the Nvidia cards in the same price tier, what are you talking about? AMD making an AMD specific upscaler would be silly given their market position. it's not an AMD-specific upscaler though, there's no reason that (like Intel) they can't offer versions of it that are ported/quantized to run on the execution units of other ML hardware but also have versions of it that take full advantage of their own hardware. like if it's impossible because marketshare, how come Intel not only did it with much less marketshare, but theirs is actually much better than FSR2 even in its DP4a version that doesn't use any proprietary accelerators? also again, AMD doesn't have a marketshare problem, they control a majority of the overall market lmao, that's the whole thing they're leaning on with FSR2 as well!
|
|
#
?
Aug 18, 2023 18:00
|
|
|
spunkshui posted:This. you're literally mad because they don't let you use the correct lovely upscaler on your 3080Ti card that doesn't need upscaling in the first place man. just take the W OTOH i'm starting to think amd was correct in doing this poo poo if it makes people freak out over incorrect upscaling. next they should make FSR like DLSS so it doesn't work on nvidia at all, imagine the tears holy poo poo
|
|
#
?
Aug 18, 2023 18:04
|
|
|
Paul MaudDib posted:it's not an AMD-specific upscaler though, there's no reason that (like Intel) they can't offer versions of it that are ported/quantized to run on the execution units of other ML hardware but also have versions of it that take full advantage of their own hardware. It isn't impossible; rather, it is pointless.
|
|
#
?
Aug 18, 2023 18:06
|
|
|
Unless Bethesda’s changed how their engine works someone’s probably gonna add dlss to Starfield themselves. It’ll be fine in the long run.
|
|
#
?
Aug 18, 2023 18:09
|
|
|
Truga posted:you're literally mad because they don't let you use the correct lovely upscaler on your 3080Ti card that doesn't need upscaling in the first place man. just take the W amd making technologies that make their customers' experiences better is good! AMD doesn't have to give everything they build away, and nor does NVIDIA. paying to make other customers' experiences worse is a different kettle of fish and is no different from Intel writing a check to keep AMD out of their partners' products. that's anticompetitive and bad, in a way that developing proprietary technologies is not. we want people to develop new technologies! and again, like openCXL, or adaptive sync, once the market is proven, if there's some value there then it'll eventually be cloned. And that's fine. XeSS kind of is exactly such a (conceptual) clone of DLSS2 and that's fine/good! Everyone seems to like it pretty well and it's doing great. AMD just doesn't want to pay to make one for their users, and that's fine too! The product has to include those support costs in the price, and AMD offers less support and that's reflected in their pricing. When you have to wait 5 years for ROCm to work, that's the fifty bucks you saved in action. When FSR3 is 2 years late to market behind DLSS3, that's the fifty bucks you saved in action. NVIDIA's operating costs don't grow on trees, someone pays for all that devrel and software development. but paying to blunt the efforts of other companies who are pushing the frontier, because you don't want to spend enough on R&D to remain competitive, is a market failure. It is, again, exactly what AMD complained about Intel doing all those years ago. we don't want to incentivize companies not doing the R&D and then writing a check to remove the consequences, or we've created a disincentive for R&D and innovation. you don't have to pay to improve your own pool, but you can't come piss in everybody else's. Paul MaudDib fucked around with this message at 18:28 on Aug 18, 2023 |
|
#
?
Aug 18, 2023 18:11
|
|
|
Truga posted:you're literally mad because they don't let you use the correct lovely upscaler on your 3080Ti card that doesn't need upscaling in the first place man. just take the W You seriously don't think a 3080 Ti needs upscaling at something like 4K?
|
|
#
?
Aug 18, 2023 18:12
|
|
|
K8.0 posted:XeSS has an Intel path and a generalized path. The Intel path performs WAY better (in terms of IQ). It could probably be adapted to run on Nvidia GPUs but why would Intel do extra work to make a worse version of DLSS? intel did say they were going to open source xess which would in principle make it possible to generalize the XMX path to other hardware, but to date they haven't actually dropped the source
|
|
#
?
Aug 18, 2023 18:12
|
|
|
maybe nvidia should invest into making FSR, which is opensource, better instead of pushing for a proprietary cuda based solution, thenProfanity posted:You seriously don't think a 3080 Ti needs upscaling at something like 4K?
|
|
#
?
Aug 18, 2023 18:13
|
|
|
FuturePastNow posted:People getting way too worked up over fancy upscalers. Yeah, it's somewhat over the top
|
|
#
?
Aug 18, 2023 18:15
|
|
|
Truga posted:it does not You are dumb.
|
|
#
?
Aug 18, 2023 18:17
|
|
|
but i'm still right and you're wrong, funny how that works
|
|
#
?
Aug 18, 2023 18:19
|
|
|
Truga posted:you're literally mad because they don't let you use the correct lovely upscaler on your 3080Ti card that doesn't need upscaling in the first place man. just take the W I’m not the one that suffers. If I wanted I could buy a loving AMD card just to play a single video game. I don’t want to see companies paying money to block features from being implemented. More specifically I don’t want to give money to companies that do that so unless you’ve got a gun ain’t opening the wallet.
|
|
#
?
Aug 18, 2023 18:29
|
|
|
I have a 3080 (not TI) and a 4k 240hz screen and use DLSS/FSR2 whenever it's available. It's basically impossible to hit 240 in any game at 4k on this card and DLSS helps me get closer. Even if I don't need it I'd still use it on the highest quality setting because it's so much better at anti-aliasing than TAA.
|
|
#
?
Aug 18, 2023 18:29
|
|
|
spunkshui posted:I’m not the one that suffers. but nvidia, intel, samsung, etc, also do this all the time are you gonna just stop buying computers entirely? also you don't need to buy a radeon for fsr to work  EngineerJoe posted:I have a 3080 (not TI) and a 4k 240hz screen and use DLSS/FSR2 whenever it's available. It's basically impossible to hit 240 in any game at 4k on this card and DLSS helps me get closer. Even if I don't need it I'd still use it on the highest quality setting because it's so much better at anti-aliasing than TAA. okay but get this, there's people still on 1060s and rx 580s and they game on them just fine
|
|
#
?
Aug 18, 2023 18:33
|
|
|
Truga posted:okay but get this, there's people still on 1060s and rx 580s and they game on them just fine You certainly understand the wants and needs of gamers over the past 30 years 
|
|
#
?
Aug 18, 2023 18:35
|
|
|
So I already contacted Sapphire/Althon Micro to do an RMA, but their tech support just responded to my first support ticket, that was mostly a copy & paste of my post here, with the detailed response of "sounds like a hardware issue, please forward a copy of the purchase receipt." Which is a pretty good response since I don't have to spend a bunch more time troubleshooting, but I did find it kind of humorous as a response in contrast to all the information I sent.
Branch Nvidian fucked around with this message at 18:40 on Aug 18, 2023 |
|
#
?
Aug 18, 2023 18:36
|
|
|
Truga posted:maybe nvidia should invest into making FSR, which is opensource, better instead of pushing for a proprietary cuda based solution, then you may as well just ask nvidia to open source DLSS, merging their TAAU R&D on top of FSR would basically mean deleting most of the FSR code and replacing it with a big ML blob ship of theseus style it would be nice if they open sourced DLSS of course but they're not going to
|
|
#
?
Aug 18, 2023 18:49
|
|
|
Branch Nvidian posted:So I already contacted Sapphire/Althon Micro to do an RMA, but their tech support just responded to my first support ticket, that was mostly a copy & paste of my post here, with the detailed response of "sounds like a hardware issue, please forward a copy of the purchase receipt." Which is a pretty good response since I don't have to spend a bunch more time troubleshooting, but I did find it kind of humorous as a response in contrast to all the information I sent. It could also just be that it doesn't tick off anything on the big list of common issues that they have, and at that point it's more time effective for them (and thusly salary-effective for the company) to swap cards with you and see if that fixes it. There's a very real point with that kind of stuff where it makes a lot more sense to just RMA the thing and deal with any associated expense than spend time and resources trying to troubleshoot it remotely.
|
|
#
?
Aug 18, 2023 18:52
|
|
|
Branch Nvidian posted:So I already contacted Sapphire/Althon Micro to do an RMA, but their tech support just responded to my first support ticket, that was mostly a copy & paste of my post here, with the detailed response of "sounds like a hardware issue, please forward a copy of the purchase receipt." Which is a pretty good response since I don't have to spend a bunch more time troubleshooting, but I did find it kind of humorous as a response in contrast to all the information I sent. They probably looked at their troubleshooting checklist and saw that you already completed it for them
|
|
#
?
Aug 18, 2023 18:54
|
|
|
spunkshui posted:I’m not the one that suffers. You know FSR does work on nvidia cards right, and doesn't make you lose performance right?
|
|
#
?
Aug 18, 2023 18:55
|
|
|
Truga posted:but nvidia, intel, samsung, etc, also do this all the time Give an example then. Show me an another example software that has a paid sponsorship to make some hardware features not work.
|
|
#
?
Aug 18, 2023 19:05
|
|
|
wargames posted:You know FSR does work on nvidia cards right, and doesn't make you lose performance right? FSR2 is about 25% slower than DLSS2 at iso-visual-quality. You “don’t lose performance” in the same sense you “don’t lose performance” by buying a 6600XT instead of a 6700XT because you it can run the same frame rates if you turn down the settings or upscale. But of course it’s still lower visual quality at any given framerate target. Paul MaudDib fucked around with this message at 19:11 on Aug 18, 2023 |
|
#
?
Aug 18, 2023 19:07
|
|
|
I have a feeling that this conversation is not going to go anywhere. There are plenty of games that my 3080ti needs to use dlss to get good framerates on at 60fps for 1440p Them saying its not required, means they probably are just saying hey don't use those settings which then devolves the argument to saying you only need a 970 because you can play 2D games still for all the point the conversation will muster.
|
|
#
?
Aug 18, 2023 19:12
|
|
|
$99 Arc A380 https://www.newegg.com/asrock-arc-a380-a380-cli-6g/p/N82E16814930076 Part of me wants to get one purely for tinkering, but might also be fun for people who are Av1 curious.
|
|
#
?
Aug 18, 2023 19:16
|
|
|
slidebite posted:Has there been a game on the PC in the modern era that has been an actual, full blown "Nvidia" or "AMD" exclusive and simply not playable otherwise? I don't think there's been a PC game that works only with one GPU vendor in at least 20 years. Maybe the olds can tell us about some crazy poo poo on the Voodoo
|
|
#
?
Aug 18, 2023 19:20
|
|
|
spunkshui posted:Give an example then. Microsoft have been doing this for almost their entire history.
|
|
#
?
Aug 18, 2023 19:22
|
|
|
VostokProgram posted:I don't think there's been a PC game that works only with one GPU vendor in at least 20 years. Maybe the olds can tell us about some crazy poo poo on the Voodoo Even in the few years Glide was king, there was almost always a fall back to software rendering/OpenGL/DirectX (once that existed). I remember some pack-in games that required a specific card to run at all mostly being lazy ports from arcade machines with the same hardware, like the Quantum3D Raven / Voodoo Banshee card or the NV1, but those were pretty rare.
|
|
#
?
Aug 18, 2023 19:25
|
|
|
Mega Comrade posted:Microsoft have been doing this for almost their entire history. But can I get an example? Everyone keeps saying that a whole bunch of companies do it all the time then shouldn’t it be really easy to give an example? I’m looking for a program that has a 3rd party hardware feature that would be basically normal for it to be present, but it’s not present due to a sponsorship. Not like tesla locking you out of their own features that they sell, but like a different company locking you out of a feature that would be inside of your own car normally. spunkshui fucked around with this message at 19:34 on Aug 18, 2023 |
|
#
?
Aug 18, 2023 19:29
|
|
|
spunkshui posted:But can I get an example? Ok. For years Microsoft required OEMs to ship with a signed key that would block Linux installation. There history is riddled with anti competitive stuff like this. Its why the EU has fined them so many times. Even now edge will actively try and hinder you installing chrome.
|
|
#
?
Aug 18, 2023 19:57
|
|
|
nvidia hosed with disabling physx on mixed GPU setups for years, so if you kept an old nvidia gpu or one of the original ageia cards around with an ATI/AMD GPU for raster, it would disable the physx nvidia hasn't had to do something like that in a while because they're in the lead with features, so they lock new features behind new GPU purchases instead all of the companies are bad and are not your friend, they just want to extract money from you
|
|
#
?
Aug 18, 2023 19:59
|
|
|
spunkshui posted:But can I get an example? I guess to meet your very specific criteria of a company behaving in a very specific way that you dissaprove, take the Wintel monopoly hobbling non-Intel cpus for...what, two decades? And then your alternative wasn't fsr (a competitive technology), it was get a new cpu which back then was extremely expensive. Or how about the many, many products and standards that Microsoft killed by either locking them out of windows or via dirty tricks. Netware, for example, sucked, but it died because MS intentionally implemented support in the slowest, most rear end backwards way imaginable. What I am saying is that if you want to die on this hill, you need to uninstall windows. Maybe use slackware on a cyrix cpu with an s3 video card. Then you will be pure.
|
|
#
?
Aug 18, 2023 19:59
|
|
|
Cygni posted:$99 Arc A380 And $219 for the A750. It's been as low as $199 before, though.
|
|
#
?
Aug 18, 2023 20:00
|
|
|
Funniest one to me is game exclusives on console. When it's a studio Sony or Microsoft own then whatever, but them buying up studios is fairly recent, it used to just be a wod of cash to just not release it on the competitors hardware. And what's worse, somehow they got consumers to celebrate it
|
|
#
?
Aug 18, 2023 20:03
|
|
|
|
| # ? Jun 4, 2024 07:46 |
|
|
Mega Comrade posted:Even now edge will actively try and hinder you installing chrome. I wouldn't call it "actively hindering" as much as sad and pathetic every time you use Edge to download Chrome on a new Windows install you get faced with a couple "Please give us a chance it's a good browser we swear  messages.
|
|
#
?
Aug 18, 2023 20:06
|
|
