GF100’s Gaming Architecture

Although it feels like ages ago, NVIDIA announced the Fermi architecture back in September of 2009, focusing on the compute abilities of the GPU that would be GF100. Today’s announcement is about filling in the blanks – where does the graphics hardware fit in to the design that NVIDIA revealed earlier.

As a quick refresher, the GF100 is composed of 512 SPs, which NVIDIA now calls CUDA cores. Each CUDA core is organized in to a block of 32 cores which we call a Streaming Multiprocessor (SM), which in turn are laid out so that 4 of them are in a Graphics Processing Cluster (GPC). Finally, there are 4 GPCs in a single GF100 chip. 32x4x4 = 512 CUDA cores.

  GF100 GT200x2 (GTX 295) GT200 (GTX 285) G92 (9800+ GTX)
Stream Processors 512 2 x 240 240 128
Texture Address / Filtering 64/256 2 x 80 / 80 80 / 80 64 / 64
ROPs 48 2x 28 32 16
Core Clock ? 576MHz 648MHz 738MHz
Shader Clock ? 1242MHz 1476MHz 1836MHz
Memory Clock ? GDDR5 999MHz (1998MHz data rate) GDDR3 1242MHz (2484MHz data rate) GDDR3 1100MHz (2200MHz data rate) GDDR3
Memory Bus Width 384-bit 2 x 448-bit 512-bit 256-bit
Frame Buffer ? 2 x 896MB 1GB 512MB
Transistor Count 3B 2 x 1.4B 1.4B 754M
Manufacturing Process TSMC 40nm TSMC 55nm TSMC 55nm TSMC 55nm
Price Point $? $500 $400 $150 - 200

When NVIDIA first unveiled GF100, we did not know anything about the ROPs, texture units, or any of the fixed-function graphics units that are customary in a GPU. Today we now have that information, and can discuss where it goes.

While GF100 resembles GT200 in a number of ways as a compute GPU, as a gaming GPU it’s very close to being a complete departure from GT200. The big change here is that the single block of fixed-function hardware as we know it is gone. Virtually every piece of hardware has been split up and moved down a level, and can now be found as part of a GPC or as a SM. NVIDIA’s GF100 block diagram does a good job of showing this.

What’s left of the fixed function hardware is the GigaThread Engine, which is what NVIDIA is calling their scheduler. To understand where our fixed function hardware went, we must meet the newest execution units: The PolyMorph Engine and the Raster Engine.

We’ll start with the Raster Engine. The Raster Engine is the combination of all the raster related hardware that we would have found previously in the fixed function pipeline. It does edge/triangle setup, rasterization, and z-culling in a pipelined manner. Each GPC has its own Raster Engine. NVIDIA has very little to say about the Raster Engine, but very little has changed from GT200 besides the fact that there are now 4 of them (one for each GPC) as opposed to 1 larger block. Each rasterizer can do 8 pixels per clock, for a total of 32 pixels per clock over the entirety of GF100.

The much more important of the two new execution units is the PolyMorph Engine, which is what NVIDIA is calling the execution unit that handles geometry for GF100. The PolyMorph Engine is responsible for handling Vertex Fetch, Tessellation, Viewport Transform, Attribute Setup, and Stream Output. It’s here where the Tessellator is located, one of the biggest changes that DX11 is bringing to GPU design. Unlike the Raster Engine, each SM has a PolyMorph Engine, giving the GF100 16 PolyMorph Engines altogether.

While the PolyMoprh Engine may sound simple in its description, don’t let it fool you. NVIDIA didn’t just move their geometry hardware to a different place, clone it 15 times, and call it a day. This was previously fixed-function hardware where a single unit sat in a pipeline and did its share of the work. By splitting up the fixed-function pipeline like this, NVIDIA in actuality created a lot of work for themselves. Why? Out of order execution.

OoO is something we usually reserve for CPUs, where high-end CPUs are built to execute instructions out of order in order to extract more performance out of them through instruction level parallelism. OoO is very hard to accomplish, because you can only execute certain instructions ahead of other ones while maintaining the correct result for your data. Execute an add instruction that relies on a previous operation before that’s done, and you have problems. GF100 isn’t a full OoO design, so we’re not going to cover OoO in-depth here, but if you’d like to know more please see this article.

At any rate, internally each PolyMorph Engine is still a simple in-order design. NVIDIA hasn’t gone so far as to make a PolyMorph Engine an OoO design – but because there are 16 of them when there used to be just 1, OoO hazards can occur just as they would in an OoO executing CPU. NVIDIA now has to keep track of what each PolyMorph Engine is doing in respect to the other 15, and put the brakes on any of them that get too far ahead in order to maintain the integrity of results.

To resolve the hazards of OoO, GF100 has a private communication channel just for the PolyMorph Engines that allows them to stay together on a task in spite of being spread apart. The fact of the matter is that all of the work that goes in to making a design like this work correctly is an immense amount of effort, and NVIDIA’s engineers are quite proud of this effort. They have taken the monolithic design of prior GPUs’ geometry units, and made it parallel. We can’t overstate how much of an engineering challenge this is.

However all of this work came at a cost, and not just the significant engineering resources NVIDIA threw at GF100. The other cost was time – we believe that the PolyMorph Engine is the single biggest reason that GF100 didn’t make it out last year. It’s the single biggest redesign of any component in GF100, and is something that NVIDIA had to start virtually from scratch on. When NVIDIA told us that designing a big GPU is hard, this is what they had in mind.

Now why did NVIDIA put themselves through all of this? Because in their eyes, they had to. The use of a fixed-function pipeline in their eyes was a poor choice given the geometric complexity that a tessellator would create, and hence the entire pipeline needed to be rebalanced. By moving to the parallel design of the PolyMorph Engine, NVIDIA’s geometry hardware is no longer bound by any limits of the pipelined fixed-function design (such as bottlenecks in one stage of the pipeline), and for better or for worse, they can scale their geometry and raster abilities with the size of the chip. A smaller GF100 derivative will not have as many PolyMorph or Raster units as GF100, and as a result won’t have the same level of performance; G92 derivatives and AMD’s designs both maintain the same fixed function pipeline through all chips, always offering the same level of performance.

Speaking of performance, we’ll get to that in a bit, but for now we’ll finish our discussion of GF100’s graphics units. Each SM has 4 texture units, which have changes somewhat from the GT200. In GT200, the texture units were at the TPC level; here they are per SM. The texture units are also affected by the cache reorganization of GF100; for L1 cache they still have the same 12KB of texture cache per set of 4 texture units, while they all share GF100’s 768KB L2 cache, which is now used by all execution units and not just textures. Each unit can compute 1 texture address and fetch 4 texture samples per clock, a significantly higher texture fetch rate than on the GT200. Furthermore in conjunction with DX11, NVIDIA’s texture units now support DX11’s new compressed texture formats. Finally, texture units are now tied to the shader clock and not the core clock. They run at 1/2 the shader clock speed of GF100.

Last but not least, this brings us to the ROPs. The ROPs have been reorganized, there are now 48 of them in 6 parttions of 8, and a 64bit memory channel serving each partition. The ROPs now share the L2 cache with the rest of GF100, while under GT200 they had their own L2 cache. Each ROP can do 1 regular 32bit pixel per clock, 1 FP16 pixel over 2 clocks, or 1 FP32 pixel over 4 clocks, giving the GF100 the ability to retire 48 regular pixels per clock. The ROPs are clocked together with the L2 cache.

That leaves us on a final note: clocks. The core clock has been virtually done away with on GF100, as almost every unit now operates at or on a fraction of the shader clock. Only the ROPs and L2 cache operate on a different clock, which is best described as what’s left of the core clock. The shader clock now drives the majority of the chip, including the shaders, the texture units, and the new PolyMorph and Raster Engines. Specifically, the texture units, PolyMorph Engine, and Raster Engine all run at 1/2 shader clock (which NVIDIA is tentatively calling the "GPC Clock"), while the L1 cache and the shaders themselves run at the full shader clock. Don’t be surprised if GF100 overclocking is different from GT200 overclocking as a result.

Index Why NVIDIA Is Focused On Geometry
POST A COMMENT

115 Comments

View All Comments

  • Stas - Tuesday, January 19, 2010 - link

    all that hype just sounds awful for nVidia. I hope they don't leave us for good. I like AMD but I like competition more :) Reply
  • SmCaudata - Monday, January 18, 2010 - link

    The 50% larger die size will kill them. Even if the reports of lower yields are false they will have to get a much smaller profit margin on their cards than AMD to stay competetive. As it is the 5870 can run nearly any game on a 30" monitor with everything turned up at a playable rate. The target audience for anything more than a 5870 is absurdly small. If Nvidia does not release a mainstream card the only people that are going to buy this beast are the people that have been looking for a reason not to buy and AMD card all along.

    In the end I think Nvidia will loose even more market share this generation. Across the board AMD is the fastest card at every price point. That will not change and with the dual GPU card already out from ATI it will be a long time before Nvidia has the highest performing card because I doubt they will release a dual GPU card at launch if they are having thermal issues with a single GPU card.

    BTW... I've only ever owned Nvidia cards but that will likely change at my next system build even after this "information."
    Reply
  • Yojimbo - Monday, January 18, 2010 - link

    what do you mean by "information"? Reply
  • SmCaudata - Monday, January 18, 2010 - link

    Heh. Just that it was hyped up so much and we really didn't get much other than some architectural changes. I suppose that maybe this is really interesting to some, but I've seen a lot of hardware underperform early spec based guesses.

    The Anandtech article was great. The information revealed by Nvidia was just okay.
    Reply
  • qwertymac93 - Monday, January 18, 2010 - link

    I really hope fermi doesn't turn into "nvidias 2900xt". late, hot, and expensive. while i doubt it will be slow by any stretch of the imagination, i hope it isn't TOO hot and heavy to be feasible. i like amd, but nvidia failing is not good for anybody. higher prices(as we've seen) and slower advancements in technology hurt EVERYONE. Reply
  • alvin3486 - Monday, January 18, 2010 - link

    Nvidia GF100 pulls 280W and is unmanufacturable , details it wont talk about publicly



    Reply
  • swaaye - Monday, January 18, 2010 - link

    Remember that they talked all about how wondrous NV30 was going to be too. This is marketing folks. They can have the most amazing eye popping theoretical paper specs in the universe, but if it can't be turned into something affordable and highly competitive, it simply doesn't matter.

    Put another way, they haven't been delaying it because it's so awesome the world isn't ready for it. Look deeper. :D
    Reply
  • blowfish - Monday, January 18, 2010 - link

    This was a great read, but it made my head hurt!

    I wonder how it will scale, since the bulk of the market is for more mainstream cards. (the article mentioned lesser derivatives having less polymorph engines)

    Can't wait to see reviews of actual hardware.
    Reply
  • Zool - Monday, January 18, 2010 - link

    Iam still curious why is nvidia pushing this geometry so hard. With 850 Mhz the cypress should be able to make 850mil polygons/s with one triangel/clock speed. Now thats 14 mil per single frame max at 60fps which is quite unrealistic. Thats more than 7 triangels per single pixel in 1920*1050. Making that amount of geometry in single pixel is quite waste and also botlenecks performance. U just wont see the diference.
    Thats why amd/ati is pushing also adaptive tesselation which can reduce the tesselation level with copute shader lod to fit a reasonable amount of triangels per pixel.
    I can push teselation factor to 14 in the dx9 ATI tesselation sdk demo and reach 100fps or put it on 3 and reach 700+ fps with almost zero difference.

    Reply
  • Zool - Tuesday, January 19, 2010 - link

    Also want to note that just tesselation is not enough and u always use displacement mapping too. Not to mention u change the whole rendering scene to more shader demanding(shadows,lightning) so to much tesselation (like in uniengine heaven on almost everything, when without tesselation even stairs are flat) can realy make big shader hit.
    If u compare the graphic quality before tesselation and after in uniengine heaven i would rather ask what the hell is taken away that much performance without tesselation as everything looks so flat like in a 10y old engine.
    The increased geometry setup should bring litle to no performance advantage for gf100, the main fps push are the much more eficience shaders with new cache architecture and the more than double the shaders of course.
    Reply

Log in

Don't have an account? Sign up now