Announcement

Collapse
No announcement yet.

AMD Announces The Radeon RX 7700 XT & RX 7800 XT Graphics Cards

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #41
    Originally posted by Mike Frett View Post

    I think you mean the price will go up.
    No, I do not. We already saw this exact situation play out with the $900 7900 XT and $1000 7900 XTX. It's even proportional (divide those by 2).

    The 7800 XT will hold around that $500 price. It appears to be a little faster than the 6800 XT which is going for around $520 right now. 7700 XT will drop to around $400-420 relatively quickly, and later in its life down to $300-350 where the 6700 XT is today.

    Comment


    • #42
      Originally posted by drakonas777 View Post
      Thanks to such internet clowns as JaysTwoCents who made such bullshit as water cooling in electronics a almost "desirable" norm.
      I'd argue that laptops have gone even battier. They seem to have opted for a roaring fan system as well as scorching the table underneath.

      Comment


      • #43
        Not a good upgrade for those with a 6700 or 6800 GPU. Probably only fractionally faster.

        I've heard the 8000 series will also have similar issues at the top end however.

        We may be waiting for 9000 series for some serious gains with AMD cards.

        My only real regret with the 4090 purchase I made is that NVIDIA's driver quality in 2023 has been pretty terrible, but they are attempting to fix all the little annoying issues...

        Comment


        • #44
          Originally posted by avis View Post

          It's inevitable. In the past you could shrink transistors and get performance for free. For the past couple of years, transistors haven't really shrunk much but OEMs need to release products with increased performance, so they've had no choice but to use bigger and bigger dies which results in more power consumption and much beefier coolers, i.e. substantially larger cards.

          We need new physics or completely new materials (photonics?) to tackle the issue.
          Sometimes you can be surprisingly rational good post

          Comment


          • #45
            It is certainly it bad, but unfortunately not a worthy upgrade from the 6 series. VRAM is tight in SDXL and it doesn't address the issue.

            Comment


            • #46
              Originally posted by artivision View Post
              That is the way you count it: 6K cores * 4xFP32 * 2.5Ghz = 60Tflops
              I couldn't get 60Tflops to anything tho. I guess is 30Tflops + 15Tiops + Your Imagination
              Ahh, I understand - you meant "4x FP32 operations per clock per SP" (using FMA and dual issue) not "4x FP32 performance of previous generation" ?

              If so, here is an example of getting 60+ TF - scroll down to "Full GPU Throughput - Vulcan" about half way down - 62.9TF on a 7900XTX.

              Editor’s Note (6/14/2023): We have a new article that reevaluates the cache latency of Navi 31, so please refer to that article for some new latency data. RDNA 3 represents the third iteratio…


              The author made a reddit post later mentioning that they now understood the Nemes' benchmarks were running in Wave64 mode, which apparently makes it easier to occupy the second ALU in each SP:

              Last edited by bridgman; 27 August 2023, 05:33 PM.
              Test signature

              Comment


              • #47
                Originally posted by bridgman View Post

                Ahh, I understand - you meant "4x FP32 operations per clock per SP" (using FMA and dual issue) not "4x FP32 performance of previous generation" ?

                If so, here is an example of getting 60+ TF - scroll down to "Full GPU Throughput - Vulcan" about half way down - 62.9TF on a 7900XTX.

                Editor’s Note (6/14/2023): We have a new article that reevaluates the cache latency of Navi 31, so please refer to that article for some new latency data. RDNA 3 represents the third iteratio…


                The author made a reddit post later mentioning that they now understood the Nemes' benchmarks were running in Wave64 mode, which apparently makes it easier to occupy the second ALU in each SP:

                https://www.reddit.com/r/hardware/co...g_amds_rdna_3/
                Come on man, you base your opinion on one benchmark than not even the tester knows what he is doing. We have run a bazillion benchmarks our selves and i can tell you that is only 30Tflops, especially in OpenCL. If you try to hackish fill anything more in FP (like replacing with hand made shaders), you end up with distorted graphics after a wile. Your second FPU is hardware damaged man. Didn't you learn nothing from Vega? When you have bad Alpha silicone, its' the best to have a rapid plan B. In your situation you could drop Navi3 for Navi2.5 with 6 months extra work.

                With shader engines from Navi2 and Asic + Neural + More Registers from Navi3 at 5nm. 300mm2 chips -5nm it's easily 8K Navi 2 Cores and 30% more performance than today at the same power. Also you didn't learn anything from consoles. Making an APU with unified DDR6 memory will raise the graphics performance about 25%, because another overhead Read/Write barrier is dropped. So soldering on the 300mm2 chip 2 Ryzen chiplets with one DDR6 channel each, will send the competition out of the window. You just need to lower the lithography by 10%, 5.5GhzMax(5GhzActual_undervolted) on the Cpu and 2.75GhzMax(2.5GhzActual_undervolted) on the Gpu (power cut 40%). Synchronous 20Gbps 1.15v undervolted memory can be used, 6+2 or 4+2 dual side channels, 48-64GB costing 200-250 bucks. Instead you raised the Gpu lithography +20% for a 3.7(3.5)Ghz target to make fuzz.

                Other lithographies and products should exist. Amd should only sell Apu's and must be on everything, even on TVs. Instead you cut Handheld Apus with the same lithography as desktop parts. Cut them with mobile lithography and get 2x the performance per watt, also use more memory channels because they are starving. It's very disturbing how your management keeps a multitrillion potential from happening.

                Comment


                • #48
                  It's always fun to read how fucking javascript kids explain how to run a semicon corporation in their 13-year old understanding of hardware business LOL

                  Comment


                  • #49
                    Originally posted by drakonas777 View Post
                    It's always fun to read how fucking javascript kids explain how to run a semicon corporation in their 13-year old understanding of hardware business LOL
                    When they are computer engineers they can.

                    Comment


                    • #50
                      Originally posted by artivision View Post

                      When they are computer engineers they can.
                      As an engineer you have no fucking clue how semicon business works. That can be easily inferred from your childishly arrogant, one-sided superficial opinion on how AMD should handle things.

                      Comment

                      Working...
                      X