1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

AMD Announces RX 5000 Series Graphics processors at Computex - Demos RX 5700

Discussion in 'Frontpage news' started by Hilbert Hagedoorn, May 27, 2019.

  1. Fox2232

    Fox2232 Ancient Guru

    Messages:
    9,469
    Likes Received:
    2,082
    GPU:
    5700XT+AW@240Hz
    Backwards compatibility is achieved through 2x~4x performance. No reason to sit on old technology for that. Especially if it is not power efficient like Vega.

    In other words, with modern non-GCN GPU backwards compatibility is granted. What they need is relevance for multi-platform market for upcoming years. (Consoles which are not left in dust by PCs within one year.)

    And it is apparent that idea of : "And in a few years or so, they'll pick the next proven tech to go with." is not from this reality. One thing is same HW with bit higher clock and better power efficiency, and another is new bigger upgrade. As that turns Console into those All-in-One systems which can't be upgraded on users demand, but user has to upgrade as there is next big thing already. (This would kill consoles, that's why there were always many years in between next big upgrade.)
     
  2. vbetts

    vbetts Don Vincenzo Staff Member

    Messages:
    14,510
    Likes Received:
    1,060
    GPU:
    RTX 2070FE
    The architecture of the core itself is GCN based that's on a smaller node and faster clocks. Here is also an article from April 2019 about the AMD drivers found for Navi in Linux.

    https://wccftech.com/amds-navi-gpus-confirmed-to-retain-gcn-design/

    The git repositories were also recently updated to reflect Navi code.
     
  3. Denial

    Denial Ancient Guru

    Messages:
    12,216
    Likes Received:
    1,390
    GPU:
    EVGA 1080Ti
    Eh, that's like calling Vega 64 a rebrand because Instinct MI25 or vice versa. Rebrand has a negative connotation due to it typically meaning it's a previous generation card with a new name. Vega VII and Instinct MI60/50 all came out roughly the same time. They are on a new node from Vega 64 and have a few features that Vega 64 didn't have.
     
    Koniakki likes this.
  4. Astyanax

    Astyanax Ancient Guru

    Messages:
    2,466
    Likes Received:
    610
    GPU:
    GTX 1080ti

    There is what looks to be GCN structure, but it is not 'gcn based'.

    RDNA has a reworked VLIW engine that retains compatibility with gcn optimized applications whilst being better built for gaming application.
     
    MonstroMart and Fox2232 like this.

  5. MonstroMart

    MonstroMart Master Guru

    Messages:
    488
    Likes Received:
    135
    GPU:
    ASUS GTX 1070 Strix
    I think you guys are mixing instructions set and architecture. I might be wrong i did my computer engineer 22 years ago when the PII was just released and I have been a web developer for the last 12 years. The new hardware is way over my head. But you can have a totally new architecture performing way better than the old one on the same instructions set. That's not a trouble at all. I don't see how some unfinished linux drivers could prove it's the same architecture on 7nm. Again i'm outdated but outside of the instructions set i don't see how it can prove anything.
     
    Last edited: May 30, 2019
    carnivore likes this.
  6. BlackZero

    BlackZero Ancient Guru

    Messages:
    8,880
    Likes Received:
    468
    GPU:
    RX Vega
    The architecture is defined by the instruction set, as that is what it is designed to run. Being a GCN based architecture would imply hardware optimisation, and significant changes to architecture would require equivalent changes to the instruction set.

    However, if no new instruction set changes are necessitated by significant advancements, the architecture will remain focused on optimisations and improvements for the same instruction set. Consequently, If the instruction set remains fit for purpose, the architecture would continue to evolve around that same instruction set. Similarly, software would follow the same basic principle, and remain focused on the instruction set.

    Being perceived as old does not imply not being fit for purpose. In fact, changes at the architectural level are largely all improvements and additions at the instruction set level, where the latest addition defines the architecture.

    Simply put, you can’t just create an instruction set when there’s no technical advancements.
     
  7. SweenJM

    SweenJM Master Guru

    Messages:
    353
    Likes Received:
    134
    GPU:
    Sapphire 590 nitro
    Navi Vega with HBM3
    [​IMG]
     
    Clawedge likes this.
  8. Evildead666

    Evildead666 Maha Guru

    Messages:
    1,082
    Likes Received:
    169
    GPU:
    Vega64/EKWB/Noctua
    PS2->PS3->PS4 were all very different hardware. Backwards compatibility was sometimes hardware, sometimes software.
    With the PS4 upwards, they will choose the path with the least cost.
    It will be x86 CPUs and Navi-for-PS5/XB is likely a mix of our-Navi and some Next years cards stuff (RT stuff+others).
    Within a year or two desktops will be walking all over them again, and they might bring out the 'Pro' model, with maybe up-rated graphics again, if needed.

    I'm not saying our-Navi is entirely GCN, but a mix of GCN and what will come late next year some time (but not the RT part apparently).
    For the consoles, i will be interested to see what differences Sony/MS made between their models, if any.
     
    Last edited: May 30, 2019
  9. Digilator

    Digilator Member

    Messages:
    26
    Likes Received:
    5
    GPU:
    XFX RX480 GTR
    Do you mean the average gaming PC will be walking all over them? I doubt it.
     
  10. Evildead666

    Evildead666 Maha Guru

    Messages:
    1,082
    Likes Received:
    169
    GPU:
    Vega64/EKWB/Noctua
    No, High end PC's. ;)
    Theres the huge wattage advantage of a PC that consoles dont get to have.
     

  11. Fox2232

    Fox2232 Ancient Guru

    Messages:
    9,469
    Likes Received:
    2,082
    GPU:
    5700XT+AW@240Hz
    Long time ago AMD stated that Navi has one similarity with GCN. And that's it is pretending to be GCN. With GCN, you have each CU consisting of 64SP and 4TMUs.
    This is no longer fixed ratio for Navi, but AMD stated that 1st Navi should keep this ratio for all those games which are optimized to use GPUs like that.
    You probably not seen it, or misunderstood it.

    Then that notion that Navi is part GCN part something that will come next year. How does such statement even work? What does it imply?
    1) Did AMD made already that next thing and decided to hybridize it with Old GCN just for laughs?
    2) Apparently you understand that "Next Thing" will be evolution of Navi. But if Navi is GCN based, then so is that "Next Thing".
     
  12. Evildead666

    Evildead666 Maha Guru

    Messages:
    1,082
    Likes Received:
    169
    GPU:
    Vega64/EKWB/Noctua
    They've tested next gen tech on current silicon before,and not activated, or used it.
    Im not saying GCN in absolute terms, of course not. How else do we describe it ?
    What will come next year may have no relation to GCN at all, chiplets and all that.
    Navi, as I see it, is the bridge.

    As ive mentioned before, im more interested in how the console people are going to have RT tech, especially when Navi purportedly doesnt.
     
  13. Fox2232

    Fox2232 Ancient Guru

    Messages:
    9,469
    Likes Received:
    2,082
    GPU:
    5700XT+AW@240Hz
    You have no clue what they did test on Vega and that it is not what is in Navi. Actually you have no clue what is being called "GCN".

    So, please. Go look at what is VLIW5 that has been replaced by VLIW4 and GCN. That part of GPU classifies as GCN code name.

    Your confusion is so high that you may as well see no difference between yourself and cancer in you. Secondly, not doing RT does not make it GCN same way as doing RT on older nVidia's HW does not turn them into Turing.

    Then you again pull this out of not knowing what AMD says. They always say: "RT is important part of technology, but it is not time, yet."
    When they deliver, it will have nothing to do with being GCN or not.
    (You will be surprised how "small" part of GPU makes GCN.)
     
  14. Evildead666

    Evildead666 Maha Guru

    Messages:
    1,082
    Likes Received:
    169
    GPU:
    Vega64/EKWB/Noctua
    I now have absolutely no idea what you are going on about.
    I really DO hope you aren't wishing cancer on me.
    I'm out of this conversation.
     
  15. Fox2232

    Fox2232 Ancient Guru

    Messages:
    9,469
    Likes Received:
    2,082
    GPU:
    5700XT+AW@240Hz
    No, I am not wishing cancer on you. But you have no idea what is GCN. And What AMD changed it with in Navi. It is funny and sad that someone with so strong opinion did not care to understand topic.

    Size of GCN CU = 4SIMD + 4TMUs.
    When AMD moved from fixed VLIW5 to GCN, their "shaders" gained in average 20% per cycle. (By removal of inefficiency in terms of problematic scheduling.)
    Here is little explanation in changes as you now may have glimpse of parts.

    [​IMG]
    How it does it simply.
    How it does it detailed.

    In 1st picture in 1st link to Hilbert's article, you can see one GCN Compute unite. That diagram fundamentally remained same for all GCN GPUs. (Vega is still 4x SIMD-16 just able to do better scheduling.) Navi as non-GCN has that diagram bit different. And it is no longer bound to 4 TMUs per CU. (While it remains same for backward compatibility.)
    And that SIMD may be actually SuperSIMD which AMD had filled in Nov 2016 and went public just a year ago.
    That's where AMD may claim another 25% IPC now for graphical purpose.
     
    Last edited: May 31, 2019

  16. Rich_Guy

    Rich_Guy Ancient Guru

    Messages:
    12,216
    Likes Received:
    406
    GPU:
    MSI 2070S X-Trio
  17. Fox2232

    Fox2232 Ancient Guru

    Messages:
    9,469
    Likes Received:
    2,082
    GPU:
    5700XT+AW@240Hz
    They are demented. Navi is Navi and bigger one will be same as smaller one. It has configurable 2x or 4x S-SIMD per CU as written in patent. 1st uses 4x configuration for backwards compatibility with software. And it does not make it GCN. They just throw it vaguely around without any details as of why they think that. (And yes, they think. They do not know.)
     
    __hollywood|meo likes this.
  18. Denial

    Denial Ancient Guru

    Messages:
    12,216
    Likes Received:
    1,390
    GPU:
    EVGA 1080Ti
    __hollywood|meo and Undying like this.
  19. MonstroMart

    MonstroMart Master Guru

    Messages:
    488
    Likes Received:
    135
    GPU:
    ASUS GTX 1070 Strix
    It's because of the unfinished linux drivers code. It proves it is 100% GCN 6th generation. Which it probably does but that's not implying what some people think it is imo (it being a refinement of Vega).
     
    Last edited: Jun 2, 2019
  20. MonstroMart

    MonstroMart Master Guru

    Messages:
    488
    Likes Received:
    135
    GPU:
    ASUS GTX 1070 Strix
    That's pretty much what i've been saying for 3 pages (minus some confusion about what an architecture is because of me being out of touch) and that's pretty much the only thing these drivers prove.

    Anyway we will see i guess there's no point arguing anymore at this point. Some people seem to believe it 100% a refinement of vega at the hardware level (implementation of the instruction set) while some people report AMD says it's a new implementation or at least partially a new one. We will see.

    I'm french and did my computer engineer in 1997 and have not been working in the field for many many years but doing some research after replying in this thread i just realized people refer to ISA as just architecture and the realization of an ISA at the hardware level as an implementation. When i was younger people would often where i live (there was no internet) refer to ISA as simply instruction set and the implementation as micro-architecture which might have confused me a little with architecture etc ... (there was no gpu like we know today back then).
     
    Last edited: Jun 2, 2019

Share This Page