Results 1 to 8 of 8

Thread: Nvidia GF100 Architecture

  1. #1
    Join Date
    Nov 2008
    Posts
    162

    Nvidia GF100 Architecture

    If we look back to the summer of 2009, Nvidia was undoubtedly the great king of the graphics market. It had been a good while as well, AMD/ATI was long a small notch behind. But suddenly on 23 September 2009 everything changed. AMD released the long-awaited 5-series. In a short time, Nvidia was really set aside and AMD knocked out one good card after another. Production problems at 40 nm technology at TSMC did not do things that much better for either party. Anyway, Nvidia lost his reputation, and they have not yet pulled them up again.

    During CES in Las Vegas, Nvidia presented their new core, GF100. GF100 is the codename for the processor that will be the Nvidia's answer to the great AMD leap that has been included in DirectX 11 world. GF100 is based on Fermi-architecture, and boasts many exciting features. This graphics processor matches Radeon HD 5870. It is therefore the complete kernel for all the new video cards. A number of degraded versions of this core will also be launched as a more economically friendly video card.

    In this article we will look in depth on GF100. We will look at the features it offers and how it can match AMD's equivalent card. Precise specifications of the new video cards are still kept strictly secret, and little is certain right here.

  2. #2
    Join Date
    Nov 2008
    Posts
    162

    Re: Nvidia GF100 Architecture

    Known specs

    There is not even a full specification list on GF100-core, but some figures we've gotten in.
    • 3 billion transistors
    • 512 CUDA processors
    • 16 geometry devices
    • 4 raster devices
    • 64 texture units
    • 48 ROPS
    • 40 nm
    • 384-bit GDDR5 memory
    • DirectX 11


    Nvidia may now boasts the 512 shader processors. Shader processors might think you are, it does not have the list provided above? Yes, Nvidia has just decided to give this new, cool name - CUDA.

    Something that is often a good indication for the performance and a video card can deliver, the number of transistors. RV870 (ATI's latest core) had 2.15 billion. On a very small piece, Nvidia has crafted 3 billion transistors.

    AMD could with their core boasts 32 ROPS and Nvidia is now making out with 48 pieces. They also run the memory bandwidth of 384-bit. This is often the major concerned when it comes to overclocking, this is a point where Nvidia has often been a great advance. Several of the cards in the 2xx series had a high memory bandwidth.

    In general we can say that Nvidia has increased the number of devices on most points, if we compare with the RV870. In addition, they claim that the way the kernel is constructed to be more efficient, which in turn leads to a higher performance even though it is not necessarily more of the individual units.

  3. #3
    Join Date
    Nov 2008
    Posts
    162

    Re: Nvidia GF100 Architecture

    Core

    Nvidia must have planned that the power of their new graphics processor will be directed and processed. If you encounter a few terms you might not understand completely, I hope the following pages will explain a great deal.

    Were you a little dizzy when you saw the picture above? Well, I hope I can get lost mostly down for you. Things are not necessarily as complicated as it seems. But, by all means, it is complicated.



    The graph you see over a simplified block diagram, for what is known as GPCArchitecture. What you see above is the host interface, also known as gigathread engine to Nvidia.

    Broken down, it is here to talk about the six memory controllers, the L2 cache of 786 kb and four GPC packets. The latter consists again of four polymorph ROPS engines, Shader processor and several other goodies. One of the focus areas of the GF100 is geometric processing and tessellation. It should create a more realistic 3D image that you may recognize from films. We will go deeper into exactly this later, but simply put, it is here talking about massive amounts of detail, not just a flat image of something that actually is curved.

    Graphics such as the GTX 280 and GTX 285 came oath 30 ROPS, and it is now talking of the 48 such units. These works mainly with mixing pixels, which most people will recognize as AA. This is the feature that smooths out the edges in 3D graphical models, the feature that prevents your game looks like the new kit from Lego.

    There are 16 green packets in the picture above, this is the shader processors. There are four such in each GPC packet, each of which consists of 32 processors. There are 512 pieces in total, more than doubling from the previous generation of video cards. Moreover, it is baked in two layers of cache, L1 and L2. Both above and below these packages include raster and polymorph engines. These works in practice as small processors, like the processor on the main card.

    Polymorph engine was added to the graphics cards as part of future development. Tessellation is something we shall come back to, but this feature alone offers more job for the video card. More power in some places in the core was required. Here is Nvidia's core efficiency into play. Rather than put this engine as a completely separate entity, its now several smaller units embedded in the overall core.

    How, then, I hope at least you've got the big picture to look like something other than a big chaos. We can explain the depth of many pages how this is related, but we do not see the big point in it. What we should do, however, is to explain to you how you can benefit from this core structure.

  4. #4
    Join Date
    Nov 2008
    Posts
    162

    Re: Nvidia GF100 Architecture

    DirectX 11

    I feel almost a little funny to think back on the DirectX world. Both Version 8 and 9 offered a host of new features and opportunities. Version 8 introduced the vertex and pixel shader in 2000. Two years later, DX9, which meant Shader Version 2.0. This was significantly greater capacity.

    DX10 however, was more than less dropped by many developers. The majority of games today still based on DX9, although many also come with support for DX10. The reason that this upgrade almost went into oblivion, is probably too simple. Microsoft did namely the genius of the decade and only implement support with Windows Vista, something we all know was not a unique popular operating system. Today however, Windows 7 is in place, with full support for DX11.

    GF100 comes as expected with the full support for DX11, everything else had just been mega stupid. Nvidia, however, is not first in the world on this, AMD had it in place six months before Nvidia had the opportunity. Let's look at exactly what this entails. Here we just instantly go through some of the most important.

    Tessellation

    Tessellation is without doubt the most exciting of DirectX 11.

    Multi-Threading

    This is the function that allows users with multi-core CPUs to have a higher speed in the game. This function will scale the graphics for several of the cores, while developers can leverage the power of a more efficient way. Your game will get a higher FPS, while the reproduction of detail is taken care of.

    Directcompute

    CPU and GPU will work in the years ahead more and more together. Previously, this was two completely different components with each task. Today, including Directcompute work together. Game developer can exploit the power of the CPU to speed in both games, office tasks and sub-editing.

    Shader Model 5.0

    Shader was first introduced in DX8 in 2000, and it has now come a long way. DX11 offers version 5.0 of this feature, which essentially makes things easier for developers. It does, however, not contain benefits for you, it does. When things go easier for developers, they can make more graphic detail in less time. It is difficult to evoke "bugs" because a number of parameters are determined for each other.

  5. #5
    Join Date
    Nov 2008
    Posts
    162

    Re: Nvidia GF100 Architecture

    Tessellation

    Tessellation has been present for some time already, but it was not until AMD came on the pitch with DirectX 11 that this was something many would actually use. We will now go a little into why this sweet word is something you really should worry about in the future.



    Let us first take a concrete example. The photo above is taken from the game Far Cry 2. This is a game with fairly good graphics and bright colors.

    First, why do you think that this person uses hat? Because the character creators wanted it? Well, it might be a part of the truth, but what really is the big reason is that woman's hair is the only one that no graphics card will not be able to put together, without that she would look untidy. They did this character hat, so your graphics card will drop and constant work with the blonde hair blowing in the wind.

    See the little metal roof. This should really be wavy. Although it might look like, it is not wavy. It is flat, with it's colored to give it some depth. This is also an example of what game makers do to save your graphics card, while giving you a somewhat realistic picture

    Tessellation is a feature that will make 3D objects more detailed and natural. It will get rid of the problem that 3D-dimensional surfaces should not be created on a surface with a colored surface. It will provide an actual depth, without requiring much from the graphics processor.



    The picture above shows how a character in any game is created. A variety of geometric shapes put together, creating rough sketch to the final grade. Graphics processor jobs a bit, and creates a softer surface that you can look at the character in the middle. Even if things here are a bit exaggerated, it still looks like we are 10 years back in time. The chart on the right hand, is something we can all accept. Until DX11 with tessellation came on the court, it would unfairly require much power from the graphics card in order to achieve this end result. Tessellation, however, does this job much more efficiently, almost without affecting the game's overall performance.

    The thing with tessellation is that it is now integrated into the GPU. AMD had this in place in their 5-series cards, and GF100 also comes with support for this. In short words tessellation gives you lots more detail in a 3D-image "on the fly".

    Tessellation breaks the surface of an object down to the numerous small geometric shapes. This is not something that just want to give great ladies a more realistic hair, or more detailed monsters. Nature Scenes will also have a significantly increased sense of things, yes, it is natural.

  6. #6
    Join Date
    Nov 2008
    Posts
    162

    Re: Nvidia GF100 Architecture

    Image quality

    The term AA or anti-aliasing if you will, should be familiar to many. This is a feature that has been adopted across a number of years now, and are often like "4xAA". This is the feature that smooths out the edges, make things look smoother and not just "choppy" at each transition between pixels.

    So, when the head is not something new, why do we mention it? Well, Nvidia has added a new mode right here. They have called the Accelerated Jittered Sampa, which we will mention that AJS forward.



    Pictured above are used tessellation on the door, because the bullet holes are actually bent inwards. Look a little closer at the gun, the edges are jagged. AJS is not in the image here, just normal AA. How much AA in question we do not know, but it does not change the situation. The AJS will do actually creates more noise in the picture. It spattered with similar colors around the edges, as the screen gets it to look smoother.

    Nvidia has implemented this under something called 32xCSAA, or just 32xAA if you like. This is an incredibly high figure, now I see myself more than satisfied enough with 4xAA in most games. Everyone knows that the more anti-aliasing you drive on, the greater the performance loss. With this new mode, it should however be only talking about a small performance impact, in relation to the graphics you'll see.





    To the top we see the GTX 280 with 16xCSAA, the later is GF100 with 32xCSAA. Although this is not a result we have gained, it is clear that the GF100 will offer a significantly better form of AA, without slaughter performance in the game.

    Multiple monitors?

    Until last fall had been a normal limitation on the video cards. You could not connect to more than two monitors. Well, it's not the whole truth, you could connect many screens to the Matrox card and similar. The problem here was that performance was never close to something you could play.

    With AMD's "Eyefinity" this was possible. Eyefinity lets you connect up to 6 monitors, each with a resolution of 2560 x 1600 pixels. If you wanted to extend these to a massive image, or six individual screens were completely up to you.

    Eyefinity was actually a little response on the visual front, because Nvidia has already released 3D support with their cards. This can not be separated with AMD, and now it appears that Nvidia really going to get many points right here. They call it "3D Surround". Although the name indicates 3 dimensions, you can also take advantage of this technology to the normal 2D use.



    3D Surround will indeed have a maximum support for three monitors. Under normal use, both playing for office tasks, the three screens each have a maximum resolution of 2560 x 1600 pixels. If you use the 3D technology, the resolution is limited to three displays of 1920 x 1080 pixels. Why image height is set to 1080 and not 1200 pixels, we do not know, but the latter had been preferred. The image can be spread composed of the three screens, alternatively, screens be individual devices.

    A slight disadvantage of Nvidia technology is that three screens require two video cards. AMD lets you connect to 3 or 6 monitors on a single card, but Nvidia require you to use two video cards running in SLI. Three 30" monitors are expensive enough in itself, and that you then need to add thousands of patches to the pot for an extra video card could well have been avoided. But so far things are now.

  7. #7
    Join Date
    Nov 2008
    Posts
    162

    Re: Nvidia GF100 Architecture

    Conclusion

    Everything we have discussed here today about upcoming Nvidia graphics processor, GF100 on the Fermi architecture is based on information they have officially released. We, or anyone else for that matter, have not yet gotten their hands around one of the new graphics cards. It means so that the end product is not tested in an objective manner, but we will say that it definitely means that the future is exciting.

    The question now is whether they will be able to take up the battle against AMD when it comes to practical performance. All functions and theories we have seen today looks promising, but it hardly helps if things do not perform in practice.

  8. #8
    EmmaMadeleine Guest

    Re: Nvidia GF100 Architecture

    thnxx software guy

Similar Threads

  1. The change in Nvidia Architecture
    By SinarB in forum Monitor & Video Cards
    Replies: 4
    Last Post: 14-08-2011, 08:28 PM
  2. Fujitsu Celsius R550 does not boot with NVidia GF100 graphics cards
    By Malcolm l in forum Monitor & Video Cards
    Replies: 6
    Last Post: 15-06-2011, 09:15 PM
  3. Technical details of Nvidia GF100
    By Lalana in forum Monitor & Video Cards
    Replies: 9
    Last Post: 25-02-2011, 07:32 PM
  4. NVIDIA GF100 vs GeForce GTX 480/470/465
    By MacKenze in forum Monitor & Video Cards
    Replies: 4
    Last Post: 23-12-2010, 02:18 AM
  5. About Nvidia GF100
    By Lawford in forum Hardware Peripherals
    Replies: 4
    Last Post: 20-04-2010, 06:08 AM

Tags for this Thread

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
Page generated in 1,711,714,025.20378 seconds with 17 queries