>can't make high performance raytracing >can't make upscaling that doesn't look ugly as sin>can't make igpu transcoding>can't make CPUs idle at low power consumption >can't make substantial CPU performance improvement even on a more advanced node>can't make security measures that don't shave off 30% fps on games and ruin performance stability on windows >can't make CPUs and GPUs that are as good for productivity tasks as they are for gamingWhat can they make?
moar coars
>What can they make?Browns shills seethe
>>102398103Ray tracing runs fine if you do it in a compute shader on AMD. Plenty of games that do it properly in a cross platform way and don't run like shit. >>102398103
>>102398103They have hardware RT acceleration specific units, and RDNA4 has them doubled compared to RDNA3, and it puts them ahead of the RTX 4000 series.FSR looks as about as good in motion as DLSS, and DLSS requires thousands of hours of model training, per game, just to still make text a garbled mess in tons of cases Phoenix/Hawk Point and Strix Point have great support in their video decoders.The monolithic APUs have very low idle power and are still some of the most power efficient chips ever made when you look at watts consumed vs frames generated with a discrete GPU.N4 is just some tweaks over N5. Intel needs an actual full node (N3) advantage to BARELY outcompete AMD.Why are you poopskins so dumb?
>>102398103>>>can't make igpu transcodingAMD having garbage media encoding/decoding makes zero sense
>R7 7800X3D in your pathFuck off intel shilling faggot. AMD is far superior to Gayntel in the efficiency department
>>102398103You forgot one>cant make GPUsAMD dont actually make GPUs and they wont until UDNA. Pic related.
>>102398479Oh cool, wasted die space for something I won't use
>>102398479>AMD dont actually make GPUs>graphics processing unit is not a graphics processing unit
>>102398666Cope and seethe Satan, AMD is a false economy. Go back to the shadow from whence you came GayMD devil.
>>102398103They can make intel go bankrupt
>>102398103more money than Intel lol
>>102398103>pic unrel
>>102398103>can't make security measures that don't shave off 30% fps on games and ruin performance stability on windowsafaik intel 13 & 14 gen cpus were the one crashing, does amd ones do the same?
>>102398730This, AMDs market share oblierates Intel
>>102398702>no argument
>AMD shill accidentally reveals himself to be SatanEvery time
>>102398588It's a nothingburger. All that's happening is that they're killing off CDNA, which is just a die shrunk Vega modified for gluing multiple dies together, rather than developing a new architecture for it in parallel with RDNA. They'll just use multiple RDNA5 dies glued together for the UDNA datacenter parts.
>>102398873>>102398702This actually makes a lot of sense. I actually used to have an AMD system and Id have recurring dreams with a paralysis demon, when it was time to upgrade about 18 months ago and I went with a 13700k and sold the old system the paralysis demon has never come back,
>>102398702>>102398873Jesse what the fuck are you talking about
>>102398103>>can't make high performance raytracing Outside of shills, who honesty cares? Serious RT is still done with specialized ASICs or CPUs. RTX mode is just a crude hack trying to stave off the increasing irrelevancy of discrete GPUs to the masses.>>can't make upscaling that doesn't look ugly as sinBoth forms of upscaling are shit, it is little more than a crutch that neither can handle RT without making massive compromises (assets, resolution)>>can't make igpu transcodingWrong>>can't make CPUs idle at low power consumption Wrong>>can't make substantial CPU performance improvement even on a more advanced nodeWrong>>can't make security measures that don't shave off 30% fps on games and ruin performance stability on windowsWrong>>can't make CPUs and GPUs that are as good for productivity tasks as they are for gamingWrong>What can they make?A vast number of solutions that go beyond silly gayming. You are clearly somebody who is still pretending it is the early 2010s. Times have changed. Nvidia even doesn't much care for silly gayming. It is just a side hustle at best for them
>>102398479>unifying datacenter and gaming architectures againWorked great for GCN, right?
>>102398103>can't make igpu transcoding...why did you specify iGPU? their dedicated gpu encoders are just as shit
>>102398479They have been making GPUs you dense idiot. You write it off just because it doesn't operate in your proprietary ecosystem of CUDA.
>>102398908What about matrix cores? Highly efficient silicon for exactly the thing many of your customers want to do would be silly to get rid of.
The OP is just a poor-fag coping that AMD has no interest or incentive to do another price war. They are seething that good number of silly gayming SKUs are commanding north of $499 and are about to enter into the ~$1999 range in the next generation at high-end.
>>102399268Radeon doesn't know what high end means
>>102399144>>can't make CPUs idle at low power consumption What is with these brain dead comments today? When is 44+50W low idle power consumption?
>>102398778>IntelExtremely subtle voltage spikes, unobservable without special tools, causing instability and accelerated degradation in chips. Due to the subtleness and slow burn nature the issue went unnoticed for years, but has been mitigated with microcode.>AMD*explodes*
>>102399561the x3d wasn't supposed to run slightly slower and cooler (lower voltage) cause the extra stacked memory can't be efficiently cooled?
>>102399561>The virgin 1.85v vs the Chad 3.0v
>>102399144People who like their games more aesthetically pleasing when its done well
>>102398103https://videocardz.com/newz/amd-confirms-fsr4-will-be-fully-ai-based-focusing-on-improving-power-efficiency-and-battery-life-for-handhelds
>>102399257>what about the thing they now need to add to their gaming GPUs to keep up with Nvidia on the spec sheet?You're not making a great case for keeping RDNA and CDNA separate.
>>102398103>>can't make high performance raytracingraytracing is a huge gimmick and lighting can easily be done better through traditional ways, it's just usually developer laziness>>can't make upscaling that doesn't look ugly as sinnot that difficult of a problem to solverest is nonsense
>>102399347>Radeon doesn't know what high end meansit doesn't exist, it's a graphics card they are all really the same except the ones that want to scam you to deliver performance below anything playable
>>102399744The good case would be RDNA having only enough of them for upscaling and frame gen and use the rest of die space for rendering. DLSS uses only 30% of Tensor Cores.
>>102399535Desktop systems idle at 30-50W (via extra controllers, perperhials on kitchen skin motherboards). You can tweak for some saving if you really care about it. The vast majority of this group do not care. Get a ultraportable or laptop if you are super concerned with idle power consumption.>>102399642Which isn't a thing under current implementations. The bigger problem there is no killer app for it. The vast majority of /v/tards care about super-fast FPS (most 4090 users are this) >>102399561>Extremely subtle voltage spikes, unobservable without special tools, causing instability and accelerated degradation in chips. Due to the subtleness and slow burn nature the issue went unnoticed for years, but has been mitigated with microcode. Intel misbinned their Raptor Lake silicon. They push harder and through that their 7nm node could tank it. The microcode is just a band-aid trying to regress it back to have its binned but the damage for existing units is already irrevocable>>AMDMotherboard vendors being retard with suicide run-tier overvolting. Early XMP implementations by motherboard vendors did this non-sense back in Sandy Bridge-E/Ivy Bridge era, although it was fried chips.
>>102399966>The vast majority of this group do not careYea they don't care but they feel obligated to shitpost and lie about power consumption. Imagine having to pay $0.15/hr just to run your PC.
What's the probability that FSR 4 will be compatible with RDNA 3 thanks to the architecture's WMMA? Or are we SOL and will need to wait for Navi 48 to release instead?
>>102398123That would been the easiest solution to the terrible generational improvements, bump ryzen 5 to 8/16, ryzen 7 to 10/20 and so on
>>102400039Not a problem, even for poor-fags. This is only problem for businesses and datacenters. This is little more then feel good non-sense by retarded "environment-fags" who don't understand that far more energy is spend on transportation, construction and materials of their eToys.
>>102399847I'm guessing that's how the consumer UDNA cards will be and the enterprise cards will have some SKUs with more. AMD can't continue to fail so hard on spec sheets, they're currently already losing sales before people even look at benchmarks. But they're also going to be focusing on the midrange market so they're not going to be pumping out cards with 20,000 tensor cores.I think a big thing that people are missing with the whole UDNA thing is that it doesn't mean AMD won't have different core designs for consumer and enterprise, it just means that those different cores will use the same microarchitecture. Like Ryzen and Epyc are very different processors, but they both use the same Zen 5 microarchitecture.
>>102398103I buy AMD out of principle. NVIDIA is too mainstream and gay
>>102400132Depends on whether it needs new data types or instructions, but given that FSR 3.1 runs on pretty much any GPU including ancient, low-end, non-AMD cards like the GTX 1050 I'd say it's likely that future versions will work on recent AMD cards.
>>102399966>it's ok for a CPU to idle at 40W despite the competition idling at 5W because... it just is okay?
>>102400702>competition idling at 5W because... it just is okay?More like 30-40W, read from the actual wall not silly digital probes. You think those bells and whistles on those kitchen sink motherboards come free? If you are really that anal about, just undervolt/underclock them like a laptop/ultraportable chip. The big take is that vast majority of desktop users, don't know or don't care to make the effort.
>>102400794AMD mobos with non APU systems have high idle power consumption because of the system chipset. It has virtually nothing to do with whatever extra features the mobo vendor chooses to include in the given SKU. APUs have very low idle power consumption.
>>102398103>What can they makeA business that isn't collapsing to the ground
>>102400819It is just a few Ws at worse you brainlet. A non-factor outside of ultraportables/laptop because of battery life. Ergo, APUs are tailored for said such platforms.The "mah idle power consumption" argument on desktop is very silly and little more retarded "environment-fags" trying to virtue signal.
>>102398103>can't make CPUs idle at low power consumptionno one can do that
>>102400877>112 watts IDLEwhat the fuck lmao
>>102400894HEDT always needs more juice
>>102400894Workstation platforms havetons of I/O connectivity and memory bandwidth nor its market care for idle as they are going mostly use for production.
>>102400871I don't believe that level of power consumption matters in any practical sense, but I am an autist about technical accuracy. I care about facts. We have to be good stewards of information for future generations or the internet becomes totally worthless.
>>102400921>Few Ws extra from the IF and chiplet overhead is "considered" to be high idle power consumption not the aggressive clockspeed and voltage profiles.There's a reason why laptop/ultraportable SKUs are not clock as high nor require as much volts.
>>102400955Clock speed has literally nothing to do with this. You can look at any HS/HX APU in a laptop, you can forcibly OC them depending on who made the board and what BIOS it ships with. These chips repackaged in the G series desktop chips are the same. It does not change idle power to anywhere near desktop CPU levels.
>>102398123*launches a quad core in 2024*
>>102401015Yes, clocksepeed is a large factor you dingus. You need more volts and power at clockspeed x. Most of the idle power consumption on desktop speeds comes from sustaining higher clockspeeds and voltages required for stability. Look at >>102400877 picture for exampleThat's why laptop/ultraportable SKUs are binned for lower clockspeeds and voltages and thier boost profiles are much more relax.IF overhead is just several more watts which shaves off minutes of precious battery life.
>>102401203clockspeed is a tiny factorit's the io diemy desktop mobile cpu is static overclocked at dangerous voltage level and the idle draw is less than 10W
>>102398103Linux drivers.
They have three times the market cap of Intel chud.
>>102401372Those aren't made by AMD
>>102401363He doesn't understand that the SOC is the problem because he's a retard. Let it go.
>>102400132they are probably going dual route like xess, one version for RDNA3 cards that will be better than FSR3 but worse than DLSS3 and other more robust version for RDNA4 that will be on par with DLSS3.
>>102401406Yes they are. AMD engineers are paid to develop AMDGPU and contribute upstream. It's not reverse engineered like Nouveau.
>>102398103they cant do shit but still make better cpus than intel
>>102401203>bloviating retard shitStop posting.
>>102401203What clock speed do you think these processors are running at while idling?
>>102401805This dipshit doesn't even know that pstates exist
>>102399766no it can't faggot
>>102399166gcn wasnt the problem, it was the atrocious software and poor technical decisions, notably with fiji, going heavy on shader and texture performance ,and not beefing up vertex performance , leading to bottlenecks in poorly optimized software(most software).with vega , picking 14nm glofo over 16nm tsmc, leading to VERY high power consumption at higher frequencies, and using hbm instead of gddr , in an attempt to reduce power consumption.these things probably would have happened regardless of architecture
>>102401917Half width HBM should have been the future. HBM PHY in a couple pads on die is immensely superior to a numerous channel PHY design all around the perimeter of a die. Packaging cost is lower than the added cost of the numerous GDDR chips.
>>102398192>DLSS requires thousands of hours of model training, per gameReally? I thought it was a general model using data from variety of games.
>>102401917they just overvolted the shit out of vega1.2V is ridiculousyou could drop 150mV and still hit advertised boost at half the power
>>102402039Nvidia had a blog post about it when it came out, trying to shill it to developers. Don't care to search for it now.
>>102402039DLSS is game specific.
>>102399535Interesting, my 14nm static OC intel is idling 30W (with Firefox open)
>>102402039DLSS1 required that, the newer versions are general models.
>>102398103>>can't make upscaling that doesn't look ugly as sinwell yeah, it's upscalingupscaling can not look anything other than worse than native
>>102402190DLSS looks as good as or better than native.
