NVIDIA GeForce RTX 5080 / RTX 5090 Linux Gaming Benchmarks

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
  • WannaBeOCer
    Senior Member
    • Jun 2020
    • 309

    #51
    Originally posted by mSparks View Post

    its not so much the disk speed, even fairly detailed high res texture files can compress down to a few 10s of MBs (once they are created), the problems come from decompressing them and their mipmaps, because they need decompressing to their actual data at some point to actually be used for something, and taking a texture, rather than a computed approach, even 24GB is only capable of holding a few dozen hires textures in uncompressed form at any one moment, anything more than that requires a ram juggling act that makes even the most famous circus act look amateur, that seems to have reached the point with 4K it is actually requiring more computation than raytracing.

    Titan only has about 100 GB/s total bandwidth, the RTX5090 is rocking 1,792 GB/s

    PS5 is 440GB/s

    main problem for developers is the PS5/Xbox series X/PCs with GPUs better than Titan RTX markets are still tiny, my guess is GTA7 will be the first game since this revolution started that can count on selling new consoles rather than selling into the fractured and overwhelmingly low end console/PC markets.

    Assuming it actually has good gameplay, which in 2025 is probably a big assumption.
    The Titan RTX memory bandwidth is 672GB/s, again textures isn't an issue. Games have been texture streaming from storage for years. It's mainly been a storage issue which was resolved with Direct storage.

    The Titan RTX is faster than all current generation consoles. It's a bit quicker than a RTX 3070 with more VRAM.
    Last edited by WannaBeOCer; 02 February 2025, 04:09 PM.

    Comment

    • mSparks
      Senior Member
      • Oct 2007
      • 2102

      #52
      Originally posted by WannaBeOCer View Post

      The Titan RTX memory bandwidth is 672GB/s,
      OK,
      But that is still only a few hundred uncompressed high res textures a second
      and only marginally more than a PS5.
      Last edited by mSparks; 02 February 2025, 04:45 PM.

      Comment

      • WannaBeOCer
        Senior Member
        • Jun 2020
        • 309

        #53
        Originally posted by mSparks View Post

        OK,
        But that is still only a few hundred uncompressed high res textures a second
        and only marginally more than a PS5.
        If memory bandwidth was that important the Radeon VII would have crushed the RTX 2080. Mine is actually at 806.4GB/s since I overclocked my memory from 1750Mhz to 2100Mhz even with that large overclock games don't get a large FPS increase. When I overclock the core from 1770Mhz to 2060Mhz though that's a different story.

        Comment

        • mSparks
          Senior Member
          • Oct 2007
          • 2102

          #54
          Originally posted by WannaBeOCer View Post

          If memory bandwidth was that important the Radeon VII would have crushed the RTX 2080.
          Its a combination of everything - e.g. you still need enough silicon to decompress the textures on the GPU to make use of that bandwidth (still never actually found out where that happens, might be CPU.... but there are formats like dxt which have "wide GPU hardware decompression support"). gonna try deepseep... ohhh a TMU function...
          Originally posted by WannaBeOCer View Post
          overclocked my memory from 1750Mhz to 2100Mhz even with that large overclock games don't get a large FPS increase. When I overclock the core from 1770Mhz to 2060Mhz though that's a different story.
          Or the TMUs are on the core clock...

          There is loads and loads of very custom silicon in modern cards and I dont pretend to know what the point of half of it does. there is also no linear scaling anymore, e.g. you run at 1 texture a second at 500GB/s, 2, 3 then suddenly 5 texture a second and the whole pipeline drops to 50GB/s on that one bottleneck you can only track down with something like nsight (The #1 reason imho nvidia GPUs have been dominating similar AMD silicon, debugging tools are important dammit.... and the success of any "real" development for either still needs a driver developers direct dial number when cutting edge developers find their bugs...)

          Comment

          Working...
          X