Let me guess, you need more?

Let me guess, you need more?

Homeless People Are Sexy Shirt $21.68

Ape Out Shirt $21.68

Homeless People Are Sexy Shirt $21.68

  1. 1 month ago
    Anonymous

    >gigabyte
    yeah I need something that isn't shit

    >t. bought a gigabyte vega56 gaming OC

    • 1 month ago
      Anonymous

      I need a medic bag

      Stfu lame ass homies

      • 1 month ago
        Anonymous

        your thread is bad and you should feel bad

        • 1 month ago
          Anonymous

          Nah my thread is good and your reply is bad and you should feel bad

  2. 1 month ago
    Anonymous

    I need a medic bag

  3. 1 month ago
    Anonymous

    no. there are no games worth to play after 2018...

  4. 1 month ago
    Anonymous

    >185W for comparable performance to a 1050ti

    • 1 month ago
      Anonymous

      gigabyte is TRASH

      Stfu gay ass homies

    • 1 month ago
      Anonymous

      Wrong homosexual, cope and seethe

  5. 1 month ago
    Anonymous

    gigabyte is TRASH

  6. 1 month ago
    Anonymous

    I need moar VRAM.

    • 1 month ago
      Anonymous

      The chinks got you covered
      good luck getting it to work with rocm but the ram is on the pcb

  7. 1 month ago
    Anonymous

    >build first rig in 2006
    >play 00s games fine
    >dies in 2011
    >replace it
    >try a bunch of new games, enjoy some of them
    >still mostly play the same 00s games
    >card dies in 2018
    >build a new 1060 rig
    >excited to try many newer games that I was filtered from
    >they suck
    >keep trying new games into 2020
    >can't find a single enjoyable game
    >go back to playing 00s games

    I don't think the fans on the card have turned on for years by now.

    • 1 month ago
      Anonymous

      kek, thats basically me.
      What game should you update for if you have a gtx 1070? The newest one I've played that doesn't run on ultra settings is Metro Exodus.

      • 1 month ago
        Anonymous

        >build first rig in 2006
        >play 00s games fine
        >dies in 2011
        >replace it
        >try a bunch of new games, enjoy some of them
        >still mostly play the same 00s games
        >card dies in 2018
        >build a new 1060 rig
        >excited to try many newer games that I was filtered from
        >they suck
        >keep trying new games into 2020
        >can't find a single enjoyable game
        >go back to playing 00s games

        I don't think the fans on the card have turned on for years by now.

        witcher 3 would smoke that card on any decent settings, so would any modern yakuza
        >insert good looking game here

        • 1 month ago
          Anonymous

          Witcher 3 came out in 2015, played it in my GTX 770 I had before. Yakuza lost judgement runs with 60fps+ in high settings.

          I'll probably habe to upgrade when I switch to a 1440p high refresh rate monitor, but for fhd 60fps the 1070 still seems to do fine.

          • 1 month ago
            Anonymous

            thats fair if you just talking med to high. I try to run every game on max settings/1440p.
            >witcher showing 55~80fps with 3080ti

  8. 1 month ago
    Anonymous

    I don't actually.

  9. 1 month ago
    Anonymous

    >yes

    • 1 month ago
      Anonymous

      I hope to god it wasn't $1676.61 for just the GPU

  10. 1 month ago
    Anonymous

    I need something that still gets new drivers. Like Maxwell which is older but Nvidia still supports.

  11. 1 month ago
    Anonymous

    actually frick you amd
    i went through fire and fricking ice to get theis fricking thing to work for what i need
    and only barely
    bc ayyymd didnt provide a handle for kernel auto-timeout settings

    fukc you aymmd.
    ill be buying ngreedia next time

    • 1 month ago
      Anonymous

      >ill be buying ngreedia next time
      or maybe not
      bc these things were cheap as dirt at a certain point and you can bypass the timeout by having multiple gpus
      but still

      frick you aymmd
      i lost literal fricking hundreds of hours on your stupid shit

    • 1 month ago
      Anonymous

      the AMD AMP opencl library is like properly abandoned, even the ROCm ones were last updated 2 years ago
      shoulda bought a card that supports HIP (AMD CUDA)
      rusticl is an option if you're on linux

      • 1 month ago
        Anonymous

        also apparently you're literally one generation out lol

        • 1 month ago
          Anonymous

          >i dont need more than that
          moreover, fun fact:
          a gpu from 2019 has as much computing power as a server FLOOR in 2004
          a wee bit' less ram, but a similar number crunching power

      • 1 month ago
        Anonymous

        hip is an intermediary layer.
        iirc its needed for opencl framework to even run on the damn card
        i might miss some things in my environment
        this whole environment is a fricking frankenstein of kernels and bits of firmware here and there

        i wasnt kidding when i said i lost hundreds of hours on this shit

        also apparently you're literally one generation out lol

        i dont need more than that
        thats why im concerned about low level control

        • 1 month ago
          Anonymous

          >hip is an intermediary layer.
          i've spent equally as long fricking around with this shit, it's not, or shouldn't be
          i've worked with the ROCm source repos, afaik hip isn't necessary for the opencl runtimes, i think the core driver for both is AMDKFD and the ROCk kernel thunks (and i think that's the same on windows, i haven't worked with the PAL runtimes that much because getting them to compile is a b***h and i'm still not even sure if it's possible)
          same thing as opencl, just a different runtime for compute kernels
          there's no intermediary layers like HSAIL (which i think used to be the case for HIP) or AMDIL any more, compute kernels are just ASM

          >i dont need more than that
          >thats why im concerned about low level control
          what you could do is look into the lower level aspects of how the opencl runtime interacts with AMDKFD
          alternative vulkan's gotten to the point where it's actually acceptable as a compute replacement, you have to use raw vulkan, all those opencl-to-vulkan layers use out of date shit before proper pointers were added to GLSL and SPIR-V

          • 1 month ago
            Anonymous

            also if you want actual proper low level control i think there might be tools for writing vulkan compute kernels in AMDGPU ASM
            there's certainly tools for disassembling them, one of the tools in the radeon gpu analyzer suite does that

          • 1 month ago
            Anonymous

            >you have to use raw vulkan
            solid gold info
            next step will be compute shaders then

            i shouldnt be thinking about any of this tho
            amd ships world class hardware but the support is the slimmest of all
            and by world class i mean just that
            rx590 has a theoretical max throughput of 250 GB/s
            sequential operations yield 225 in practice
            ->but non-sequential reads come at 180GB/s
            ->when im supposed to expect almost an order of magnitude of deifference between both
            ->polaris 30 EOL @ 2021 in practice.
            oficially dropped in 2022(? iirc)
            wtf ayymd?

          • 1 month ago
            Anonymous

            this is the ROCm opencl runtime (and the HIP runtime)'s source btw
            https://github.com/ROCm/clr
            and yea i was right about the heirarchy HIPAMD and amdocl are seperate and both depend on rocclr which depends on the kernel thunks which depends on AMDKFD

            you might even be able to modify this to support what you want

            >you have to use raw vulkan
            solid gold info
            next step will be compute shaders then

            i shouldnt be thinking about any of this tho
            amd ships world class hardware but the support is the slimmest of all
            and by world class i mean just that
            rx590 has a theoretical max throughput of 250 GB/s
            sequential operations yield 225 in practice
            ->but non-sequential reads come at 180GB/s
            ->when im supposed to expect almost an order of magnitude of deifference between both
            ->polaris 30 EOL @ 2021 in practice.
            oficially dropped in 2022(? iirc)
            wtf ayymd?

            you'll be wanting to use these, these are how pointers are exposed to GLSL
            https://github.com/KhronosGroup/GLSL/blob/main/extensions/ext/GLSL_EXT_buffer_reference.txt
            https://github.com/KhronosGroup/GLSL/blob/main/extensions/ext/GLSL_EXT_buffer_reference2.txt
            it's been made core as of vulkan 1.3 but like everything that doesn't mean it's not optional
            you don't even have to use descriptors (vulkan abstract resource handles) with them since you can just shove a vulkan buffer pointer you get from the API in a push constant and it works
            and be careful with vulkan, since everything is optional there's a certain subset of features which are good to turn on and some that are bad to turn on
            like you'll definitely want all the robustness (bounds checking) features turned off for runtime use, the subgroup features turned on (cross warp communication primitives), probably most of the atomic memory access related features turned on, the scoped memory access features turned on since that's how you access LDS and GDS memory on AMD cards inside shaders
            apparently the memory model (which is also somehow optional) is really important for good compute performance
            i wound up picking a superset of the vulkan desktop profile (https://vulkan.lunarg.com/doc/sdk/1.3.275.0/windows/profiles_definitions.html)
            if you have to use textures/images and your drivers support them use descriptor buffers
            they're super efficient on AMD hardware
            >abut the support is the slimmest of all
            it's bad
            explicit use of most of their advanced features like DPP (sharing inside of a warp) are inaccessible unless you write ASM even with HIP (there's compiler intrinsics but they suck)

          • 1 month ago
            Anonymous

            info duly saved.
            thanks anon
            amd invented a mountain, and its easier to climb it with a map

          • 1 month ago
            Anonymous

            i have a bit more for you
            i linked it before but this is the vulkan GLSL extensions repo which contains all the new bits added to glsl for vulkan
            https://github.com/KhronosGroup/GLSL/blob/main/extensions/
            unfortunately there's not list anywhere i can find of what regular GLSL extensions are supported (some of them are but which? who knows)

            the directx equivalents of these are apparently super important for good DXIL generation (read through this series of 7 blogposts if you've got the time, it's by a vkd3d-proton dev https://themaister.net/blog/2021/09/05/my-personal-hell-of-translating-dxil-to-spir-v-part-1/) so i assume they're for good SPIR-V generation, especially the first one and i don't think? they require driver support
            https://github.com/KhronosGroup/GLSL/blob/main/extensions/ext/GL_EXT_control_flow_attributes.txt
            https://github.com/KhronosGroup/GLSL/blob/main/extensions/ext/GL_EXT_control_flow_attributes2.txt
            https://github.com/KhronosGroup/GLSL/blob/main/extensions/ext/GL_EXT_maximal_reconvergence.txt

            you'll want to use an offline shader compiler (i think glslc is the proper one with the most features although a lot of people just use the reference impl glslang) or integrate something like shaderc into your program

            oh and the AMD coherent memory extension is super useful for jury rigging shit too
            it's slower but it means you don't have to worry about host-device synchronization with memory access to mappable allocated buffers
            HIP has a similar feature but i don't know if opencl ever got proper shared memory support

            and don't forget about the radeon GPU analyzer suite i mentioned

            also if you want actual proper low level control i think there might be tools for writing vulkan compute kernels in AMDGPU ASM
            there's certainly tools for disassembling them, one of the tools in the radeon gpu analyzer suite does that

            , you can disassemble serialized vulkan pipelines and it comes bundled with an offline pipeline compiler for regular compute pipelines as well

          • 1 month ago
            Anonymous

            >it comes bundled with an offline pipeline compiler for regular compute pipelines as well
            to clarify that's a full offline pipeline compiler
            so GLSL to AMDGPU binaries
            i think? it works with the mesa drivers?
            at least i hope it does, i've only used it on windows

          • 1 month ago
            Anonymous

            i will check that out
            its not a problem to make a win install

            i have a bit more for you
            i linked it before but this is the vulkan GLSL extensions repo which contains all the new bits added to glsl for vulkan
            https://github.com/KhronosGroup/GLSL/blob/main/extensions/
            unfortunately there's not list anywhere i can find of what regular GLSL extensions are supported (some of them are but which? who knows)

            the directx equivalents of these are apparently super important for good DXIL generation (read through this series of 7 blogposts if you've got the time, it's by a vkd3d-proton dev https://themaister.net/blog/2021/09/05/my-personal-hell-of-translating-dxil-to-spir-v-part-1/) so i assume they're for good SPIR-V generation, especially the first one and i don't think? they require driver support
            https://github.com/KhronosGroup/GLSL/blob/main/extensions/ext/GL_EXT_control_flow_attributes.txt
            https://github.com/KhronosGroup/GLSL/blob/main/extensions/ext/GL_EXT_control_flow_attributes2.txt
            https://github.com/KhronosGroup/GLSL/blob/main/extensions/ext/GL_EXT_maximal_reconvergence.txt

            you'll want to use an offline shader compiler (i think glslc is the proper one with the most features although a lot of people just use the reference impl glslang) or integrate something like shaderc into your program

            oh and the AMD coherent memory extension is super useful for jury rigging shit too
            it's slower but it means you don't have to worry about host-device synchronization with memory access to mappable allocated buffers
            HIP has a similar feature but i don't know if opencl ever got proper shared memory support

            and don't forget about the radeon GPU analyzer suite i mentioned [...], you can disassemble serialized vulkan pipelines and it comes bundled with an offline pipeline compiler for regular compute pipelines as well

            >big boy tools
            me likey.
            but also it makes me appreciate somewhat more the simplicity of opencl, ngl.
            its just
            ->grug makes C string
            ->sends to adequate function call
            simple as.

            but yeah, i guess its time to actually learn the ins and outs of the tech.
            GLSL is the classical way of doing things no? (and might come in handy outside of pure compute. like in case of plan b gayme)

          • 1 month ago
            Anonymous

            >->grug makes C string
            >->sends to adequate function call
            that's what shaderc is for
            aside from logical device setup (where you pick extensions and features, pic related is part of the logical device creation from my on-again off-again vulkan compute project, this took a lot of research to figure out and still isn't proper as i've got a bunch of unnecessary shit enabled) using vulkan for compute with an integrated online compiler library is very very easy
            descriptors (needed for images) and renderpasses (only needed for rendering) is where vulkan gets messy
            descriptor buffers kinda fix descriptors
            a lot of people complain about synchronization but if you've used opencl you're probably okay with the concept of granular sync using things like scoped memory barriers
            >>GLSL is the classical way of doing things no? (and might come in handy outside of pure compute. like in case of plan b gayme)
            yea
            you have the option of HLSL in vulkan now since MS opensourced dxc (the vulkan specific stuff is done through attributes) but i think SPIR-V generation is still all kinds of messed up
            and as of the last vulkan conference (vulkanized, there's a lot of useful info in them) they're trying to push slang because they realize they're dropped the ball hard with GLSL

            also here are some assorted docs
            https://vkguide.dev/
            https://vulkan-tutorial.com/
            https://github.com/KhronosGroup/Vulkan-Samples
            https://github.com/SaschaWillems/Vulkan
            alas there's not really a good place to learn pure compute vulkan

            they fairly recently revamped the spec into a proper documentation site:
            https://docs.vulkan.org/spec/latest/index.html

            also everyone uses the basic C API but for your sake use the C++ API, it's very helpful and even fixes some errata, you can turn off a lot of the bloat if need be
            i patched the generator to include C++20 using enum support for their flag types to make it easier to use but even without that it's very comfy

          • 1 month ago
            Anonymous

            here's the remaining part of it if you'd like to use it
            and this is the list of device extensions i've got enabled
            VK_KHR_GLOBAL_PRIORITY_EXTENSION_NAME,
            VK_EXT_SHADER_ATOMIC_FLOAT_EXTENSION_NAME,
            VK_EXT_SHADER_ATOMIC_FLOAT_2_EXTENSION_NAME,
            VK_KHR_SHADER_CLOCK_EXTENSION_NAME,
            VK_EXT_DESCRIPTOR_BUFFER_EXTENSION_NAME,
            VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME,
            VK_EXT_MEMORY_BUDGET_EXTENSION_NAME,
            VK_EXT_ROBUSTNESS_2_EXTENSION_NAME,
            VK_GOOGLE_DECORATE_STRING_EXTENSION_NAME,
            VK_GOOGLE_HLSL_FUNCTIONALITY1_EXTENSION_NAME,
            VK_GOOGLE_USER_TYPE_EXTENSION_NAME,
            VK_AMD_DEVICE_COHERENT_MEMORY_EXTENSION_NAME

            also the C++ API comes with something the C API doesn't, an extension dependency checker, <vulkan/vulkan_extension_inspection.hpp> is really important

          • 1 month ago
            Anonymous

            here's the remaining part of it if you'd like to use it
            and this is the list of device extensions i've got enabled
            VK_KHR_GLOBAL_PRIORITY_EXTENSION_NAME,
            VK_EXT_SHADER_ATOMIC_FLOAT_EXTENSION_NAME,
            VK_EXT_SHADER_ATOMIC_FLOAT_2_EXTENSION_NAME,
            VK_KHR_SHADER_CLOCK_EXTENSION_NAME,
            VK_EXT_DESCRIPTOR_BUFFER_EXTENSION_NAME,
            VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME,
            VK_EXT_MEMORY_BUDGET_EXTENSION_NAME,
            VK_EXT_ROBUSTNESS_2_EXTENSION_NAME,
            VK_GOOGLE_DECORATE_STRING_EXTENSION_NAME,
            VK_GOOGLE_HLSL_FUNCTIONALITY1_EXTENSION_NAME,
            VK_GOOGLE_USER_TYPE_EXTENSION_NAME,
            VK_AMD_DEVICE_COHERENT_MEMORY_EXTENSION_NAME

            also the C++ API comes with something the C API doesn't, an extension dependency checker, <vulkan/vulkan_extension_inspection.hpp> is really important

            extremely based
            currently reading vulkan features specs despite being slightly drunk
            you provided enough to get me a running start anon, god bless

          • 1 month ago
            Anonymous

            one more thing since this is AMD specific, something i only learned recently, and not very well documented, and i don't know if polaris has these (you'll have to check the ISA manual)
            https://github.com/KhronosGroup/GLSL/blob/main/extensions/khr/GL_KHR_memory_scope_semantics.txt
            this is the memory scope extension i mentioned and i'm not sure how true this is
            i think GDS memory (global small memory, might have special hardware acceleration for atomic ints) is accessed in shaders through the workgroupcoherent qualifier (the shaders in question where i learned this from were written in HLSL and annotated with groupshared and i think that's the equivalent)
            usually LDS (local small memory, might have special acceleration for atomic ints) is usually just accessed through small regions of shared memory but in theory subgroupcoherent should explicitly refer to it
            if you're interested in the nitty gritty LLVM has more info on how a lot of this stuff works at a very very low level
            https://llvm.org/docs/AMDGPUUsage.html

          • 1 month ago
            Anonymous

            how the f can you know all this, senpai?

          • 1 month ago
            Anonymous

            i read source code for fun, have worked somewhat with compiler internals (i maintained an experimental clang fork for myself for a while before the pace of upstream updates caused so many errors i just gave up), and something about low level GPGPU tickles my autism just right
            despite that i've not much to show for it lol, so many half started projects
            i'm just a hobbyist dev, the only related industry around where i live is military industrial and this isn't relevant to them

          • 1 month ago
            Anonymous

            although i suppose it's not like you see any professional devs working on AMD shit besides AMD and Valve developers
            i wonder if valve would hire me with the very limited credentials i have

          • 1 month ago
            Anonymous

            although i suppose it's not like you see any professional devs working on AMD shit besides AMD and Valve developers
            i wonder if valve would hire me with the very limited credentials i have

            also book piracy
            libgen has so much shit
            i downloaded basically every C++ book, every CUDA book, a bunch of opencl books i never read because i found out HIP could work on windows well before official release (although i've since switched to linux) if you knew your AMD driver internals and clang's source code
            i think some of their shit has moved to tor after they got blocked from a bunch of public IPFS instances when idiots on tiktok started shilling them

            nothing helped with the AMD specific shit though, that's all source code (the AMD vulkan ray tracing lib, llvm and clang's AMDGPU testing code, and the HIP and opencl device libraries in particular (https://github.com/ROCm/llvm-project/tree/amd-staging/amd/device-libs/))

            Buy the chink 5600xt 6gb for $130 or 5700xt 8gb for $150

            5700xt
            i think the 8 gigs of VRAM is fairly important and the more power is nice

          • 1 month ago
            Anonymous

            *https://github.com/ROCm/llvm-project/tree/amd-staging/amd/device-libs

          • 1 month ago
            Anonymous

            >that's all source code
            *and AMD's official docs
            they don't have much but
            https://gpuopen.com/learn/amdgcn-assembly/
            https://www.amd.com/content/dam/amd/en/documents/radeon-tech-docs/instruction-set-architectures/gcn3-instruction-set-architecture.pdf
            apparently these are the docs for polaris despite it being gcn4?
            maybe gcn4 was just a refresh?

  12. 1 month ago
    Anonymous

    >laughs in 7900xtx
    The 580 was a low tier card 4 years ago if you haven’t upgraded yet you are either poor, tech illiterate, or just a moron

  13. 1 month ago
    Anonymous

    I think mine has a bigger number. I don't know, I wasn't really paying attention. I just clicked buy!
    Also mine is a Sapphire, which I heard was good or whatever.

  14. 1 month ago
    Anonymous

    Intel needs to save us from these companies i cannot take more amd cards bricking from bad drivers and nvidia poomping their prices by 800% because of AI

    Intel might just be the last hope for solid midrange gpus that cost a midrange price

  15. 1 month ago
    Anonymous

    My photogrammetry projects are bottlenecked by my RX580 and camera+light rig. The GPU is the more economic investment of the two. Going to test processing times with an A5000 at work.

  16. 1 month ago
    Anonymous

    Buy the chink 5600xt 6gb for $130 or 5700xt 8gb for $150

  17. 1 month ago
    Anonymous

    I need less
    >t. Sapphire Pulse RX 570 8GB model

  18. 1 month ago
    Anonymous

    This card served me well and I have nothing but good things to say about it, if only AMD wasn't a b***h and supported it a bit more driver/software wise but it's time to let go of it I guess.

  19. 1 month ago
    Anonymous

    I need less

Your email address will not be published. Required fields are marked *