Would a GTX 580 improve my current system?

Discussion in 'Videocards - NVIDIA GeForce' started by NeoEnigma, Oct 16, 2011.

  1. BlackZero

    BlackZero Ancient Guru

    Messages:
    8,879
    Likes Received:
    478
    GPU:
    RX Vega
    lol, ok didn't want to make this more complicated but now we agree that the memory bandwidth wouldn't change whether you had one truck or two as it would still take the same amount of time to fill it and return I'll now explain why I said two trucks.

    I said two trucks because i didn't want to drift away from the fact that in reality you do have two physical amounts of memory but the memory, or sand in this case, needs to be in seperate trucks so the the person shovelling has easy access to the sand without the other person getting in his way.

    Now as the memory is mirrored both trucks have sand rather than one having sand and the other gravel. This is because they must throw sand into the sea for half the day and gravel for the other half, the times that sand or gravel gets thrown in can not change but the amount of sand or gravel gettting thrown can change.

    It's the same with memory, even though you have 1 gb on each card the data stored on that 1gb must be the same so that the required data for that period is avaialbale to both cards and they can both use that same data and effectively process it twice and so render twice as many frames. but the essential data itself still remians identacle for both cards and can only be collected when both cards require it so is not collected twice as fast (memory bandwidth).
     
  2. kitch9

    kitch9 Ancient Guru

    Messages:
    1,889
    Likes Received:
    0
    GPU:
    XFX 7990 3GB
    To be fair thinking about it, both our analogies are crap, as we are treating the video memory as something that constantly gets depleted and needs to be refilled.

    When you play a game information such as textures, geometry ect ect are loaded into into video ram yes? This information is provided from system ram usually before the start of a level unless some kind of streaming technology is involved, but lets disregard streaming for now. This data being loaded has nothing to do with video card bandwidth, as this comes over the PCI-e bus can we agree on that?

    Once the game loads all the data the gpu will need to process each frame is in video memory which is eventually outputted to your screen.

    The GPU will treat the data in the video memory as a library, and it will look up what data it needs to build the frame based on what info it receives from the CPU. The data in the video memory will remain fairly constant, will not deplete and the GPU will keep referring to it as it needs to. The speed at what the GPU can refer to or modify this data is the part that is classed as bandwidth, do we agree on that?

    On that basis we have to treat the video ram more as a library full of books for the GPU to use as it sees fit, and the time it takes to read those books for the needed info as the bandwidth. Do we agree here? The GPU will also store anything its written in the library too.

    Now in a single gpu system you have one library and one GPU trying to solve a problem using info from the library with a fixed reading and writing speed (Bandwidth) and everything is fairly straight forward, yes?

    In a SLI system what we have now is two GPU's exactly the same as each other both working on the same problem. Each GPU has their own library, BUT each of their libraries has EXACTLY the same books and information in it.

    The benefits of this though is that each GPU can solve a different part of the problem AT THE SAME TIME, so whilst one GPU is looking up the info and reading about one thing (Bandwidth.) the other GPU is looking up and reading about another part of the problem (Bandwidth.) again. Therefore twice the amount of information can be read (and written.) at the same time, therefore doubling bandwidth.

    That's how I see it.
     
    Last edited: Oct 17, 2011
  3. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,963
    Likes Received:
    963
    GPU:
    GTX1080Ti
    Just as wrong as when they add the bus WIDTH together in SLI or add the GPU memory together to come up with a GTX 590 having 3072MB of video memory when clearly it only has a 1536MB frame buffer per GPU. Why did you need that repeated again? What you lied about was saying I agree with you that two GPU working together doubled the bandwidth. It's obvious from my post, I do not. It's amusing you took a couple of dozen sentences to try to explain what took me only a couple and you still have it wrong.
     
  4. BlackZero

    BlackZero Ancient Guru

    Messages:
    8,879
    Likes Received:
    478
    GPU:
    RX Vega
    For the most part that is actually very intelligently described and quite accurate however where you seem to miss a part is that not only do they both have the same books but they must also read the same book together and, even though they are reading different lines, the lines must be read in turn so it still takes them both the same amount of time to read the books as if it was a sinlge card reading it as they can't move to the next book until they have both finished reading that book.
     

  5. kitch9

    kitch9 Ancient Guru

    Messages:
    1,889
    Likes Received:
    0
    GPU:
    XFX 7990 3GB
    I'm not arguing about frame buffer size, I completely agree with you on that point and have never said otherwise.

    I am talking about how much data can be read or written to or from video ram in SLI, compared to a single GPU. Please see above post for what I mean.
     
  6. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,963
    Likes Received:
    963
    GPU:
    GTX1080Ti
    And that doesn't even get into render ahead, back buffers, buffer flipping etc ect...
     
  7. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,963
    Likes Received:
    963
    GPU:
    GTX1080Ti
    And you're not arguing that adding bus width is wrong as well. And now you're not even arguing the memory throughput is effectively doubled. In fact, you more or less conceded you're wrong. Any questions?
     
  8. kitch9

    kitch9 Ancient Guru

    Messages:
    1,889
    Likes Received:
    0
    GPU:
    XFX 7990 3GB
    I disagree, each card works on alternate frames / half of the frame based on what the CPU requested from the GPU driver. Each GPU can read what it wants from its own library based on what its asked to process buy the GPU driver.

    Unless you have info to the contrary, but I certainly have never read anything like you describe.



    Are your eyes painted on? I've been arguing effective memory bandwidth is doubled with SLI and nothing I have posted contradicts that?


    Hence in most cases SLI doesn't quite scale 100% usually but nearer 95-98% in an SLI optimised title, but sometimes you can over 100% SLI scaling as bandwidth constraints can actually be freed in certain conditions.
     
    Last edited: Oct 17, 2011
  9. BlackZero

    BlackZero Ancient Guru

    Messages:
    8,879
    Likes Received:
    478
    GPU:
    RX Vega
    Well I can't school you any further, you can continue to see it how you like but it doesn't change the facts.
     
  10. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,963
    Likes Received:
    963
    GPU:
    GTX1080Ti

    And you're wrong. Recall your original claim: 'SLI effectively doubles available bandwidth'. The contents of each GPU are identical. So there isn't a effective doubling of bandwidth as each GPU gets the exact same information loaded into memory at the same time. It's amazing to me how something so simple can be screwed up time and time again. But I'm 100% certain you'll make the same screw up again. And I'm just as certain any further posts would be a complete waste of time.
     
    Last edited: Oct 17, 2011

  11. kitch9

    kitch9 Ancient Guru

    Messages:
    1,889
    Likes Received:
    0
    GPU:
    XFX 7990 3GB
    You haven't schooled me in anything, the "facts" you mention should be well documented as its a MAJOR hindrance to how SLI would work. I can't find anything which says what you have said, so please feel free to prove me wrong.

    Dude, I'm not denying each video card loads the same data, but thats not bandwidth. Bandwidth is the speed at which the GPU can READ OR WRITE the data to or from video ram

    In SLI you have 2 GPUs each with the same data in their video ram libraries, but each can read or write their own data independently of each other to their respective video ram effectively doubling the throughput. (Bandwidth.)
     
    Last edited: Oct 17, 2011
  12. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,963
    Likes Received:
    963
    GPU:
    GTX1080Ti
    It's well documented that in SLI and Crossfire that the memory contents are mirrored. Derp.
     
  13. kitch9

    kitch9 Ancient Guru

    Messages:
    1,889
    Likes Received:
    0
    GPU:
    XFX 7990 3GB
    I know dude, that's the point!?!?!
     
  14. BlackZero

    BlackZero Ancient Guru

    Messages:
    8,879
    Likes Received:
    478
    GPU:
    RX Vega
    My very first post tried to make this clear but it had very little effect, evidently, so no matter what is said it's not going to make any difference. I mean how difficult is it to understand that the memory is mirrored so both cards have to load the same data and it's the memory bandwidth doing the loading.
     
  15. Præses

    Præses Master Guru

    Messages:
    941
    Likes Received:
    13
    GPU:
    Strix GTX 970 4GB
    OP: Try OC'ing your CPU first before you fork out money for anything new. If OC'ing your CPU helps, then you know where the problem lies.

    Arguing people: Get your own thread :p
     

  16. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,963
    Likes Received:
    963
    GPU:
    GTX1080Ti
    The point is, as Nvidia says:

    A note on GPU Memory in SLI
    In all SLI-rendering modes all the graphics API resources (such as buffers or textures) that would normally be expected to be placed in GPU memory are automatically replicated in the memory of all the GPUs in the SLI configuration. This means that on an SLI system with two 512MB video cards, there is still only 512MB of onboard video memory available to the application. Any data update performed from the CPU on a resource placed in GPU memory (for example, dynamic texture updates) will usually require the update to be broadcast other GPUs. This can introduce a performance penalty depending on the size and characteristics of the data. Other performance considerations are covered in the section on SLI performance.

    http://developer.download.nvidia.com/whitepapers/2011/SLI_Best_Practices_2011_Feb.pdf

    Class dismissed.
     
    Last edited: Oct 17, 2011
  17. kitch9

    kitch9 Ancient Guru

    Messages:
    1,889
    Likes Received:
    0
    GPU:
    XFX 7990 3GB
    Awesome, you've found a vague line that says that using dynamic textures can affect performance of SLI, but mentions nothing about choking bandwidth, and I would guess the bandwidth requirements of performing such a task would be very little.

    So I ask:

    A:) Is every game made up of entirely dynamic assets? I would guess not.

    B:) How much difference does it actually make to memory bandwidth as the line you posted mentions nothing about it?
     
  18. kitch9

    kitch9 Ancient Guru

    Messages:
    1,889
    Likes Received:
    0
    GPU:
    XFX 7990 3GB
    OK, from the very document you posted:

    In fact reading further it even states in their document that they go out of their way to avoid cross GPU communication:

    I think you should call the class back......
     
    Last edited: Oct 17, 2011
  19. XP-200

    XP-200 Ancient Guru

    Messages:
    5,481
    Likes Received:
    976
    GPU:
    Zotac GTX 1080 Mini
    What was the question? LOL
     
  20. kitch9

    kitch9 Ancient Guru

    Messages:
    1,889
    Likes Received:
    0
    GPU:
    XFX 7990 3GB
    Lol, don't ask....
     

Share This Page