Architecting Fermi: More Than 2x GT200

NVIDIA keeps referring to Fermi as a brand new architecture, while calling GT200 (and RV870) bigger versions of their predecessors with a few added features. Marginalizing the efforts required to build any multi-billion transistor chip is just silly, to an extent all of these GPUs have been significantly redesigned.

At a high level, Fermi doesn't look much different than a bigger GT200. NVIDIA is committed to its scalar architecture for the foreseeable future. In fact, its one op per clock per core philosophy comes from a basic desire to execute single threaded programs as quickly as possible. Remember, these are compute and graphics chips. NVIDIA sees no benefit in building a 16-wide or 5-wide core as the basis of its architectures, although we may see a bit more flexibility at the core level in the future.

Despite the similarities, large parts of the architecture have evolved. The redesign happened at low as the core level. NVIDIA used to call these SPs (Streaming Processors), now they call them CUDA Cores, I’m going to call them cores.

All of the processing done at the core level is now to IEEE spec. That’s IEEE-754 2008 for floating point math (same as RV870/5870) and full 32-bit for integers. In the past 32-bit integer multiplies had to be emulated, the hardware could only do 24-bit integer muls. That silliness is now gone. Fused Multiply Add is also included. The goal was to avoid doing any cheesy tricks to implement math. Everything should be industry standards compliant and give you the results that you’d expect.

Double precision floating point (FP64) performance is improved tremendously. Peak 64-bit FP execution rate is now 1/2 of 32-bit FP, it used to be 1/8 (AMD's is 1/5). Wow.

NVIDIA isn’t disclosing clock speeds yet, so we don’t know exactly what that rate is yet.

In G80 and GT200 NVIDIA grouped eight cores into what it called an SM. With Fermi, you get 32 cores per SM.

The high end single-GPU Fermi configuration will have 16 SMs. That’s fewer SMs than GT200, but more cores. 512 to be exact. Fermi has more than twice the core count of the GeForce GTX 285.

  Fermi GT200 G80
Cores 512 240 128
Memory Interface 384-bit GDDR5 512-bit GDDR3 384-bit GDDR3


In addition to the cores, each SM has a Special Function Unit (SFU) used for transcendental math and interpolation. In GT200 this SFU had two pipelines, in Fermi it has four. While NVIDIA increased general math horsepower by 4x per SM, SFU resources only doubled.

The infamous missing MUL has been pulled out of the SFU, we shouldn’t have to quote peak single and dual-issue arithmetic rates any longer for NVIDIA GPUs.

NVIDIA organizes these SMs into TPCs, but the exact hierarchy isn’t being disclosed today. With the launch's Tesla focus we also don't know specific on ROPs, texture filtering or anything else related to 3D graphics. Boo.

A Real Cache Hierarchy

Each SM in GT200 had 16KB of shared memory that could be used by all of the cores. This wasn’t a cache, but rather software managed memory. The application would have to knowingly move data in and out of it. The benefit here is predictability, you always know if something is in shared memory because you put it there. The downside is it doesn’t work so well if the application isn’t very predictable.

Branch heavy applications and many of the general purpose compute applications that NVIDIA is going after need a real cache. So with Fermi at 40nm, NVIDIA gave them a real cache.

Attached to each SM is 64KB of configurable memory. It can be partitioned as 16KB/48KB or 48KB/16KB; one partition is shared memory, the other partition is an L1 cache. The 16KB minimum partition means that applications written for GT200 that require 16KB of shared memory will still work just fine on Fermi. If your app prefers shared memory, it gets 3x the space in Fermi. If your application could really benefit from a cache, Fermi now delivers that as well. GT200 did have an L1 texture cache (one per TPC), but the cache was mostly useless when the GPU ran in compute mode.

The entire chip shares a 768KB L2 cache. The result is a reduced penalty for doing an atomic memory op, Fermi is 5 - 20x faster here than GT200.

A Different Sort of Launch A More Efficient Architecture


View All Comments

  • Gary Key - Monday, October 05, 2009 - link

    He is gone as are a couple of others. We have a new comments system in final development now that should take care of this problem in the future. Reply
  • Voo - Saturday, October 03, 2009 - link

    You may overseen it, but there was a edit by an administrator to one of his posts which did exactly what you want ;) Reply
  • james jwb - Sunday, October 04, 2009 - link

    that's good to hear :) Reply
  • Hxx - Friday, October 02, 2009 - link

    By the looks of it, Nvidia doesn't have much going on for this year. If they loose the DX11 boat against ATI then I will pity their stockholders. About the only thing that makes those green cards attractive is their Physics spiel. Now if ATI would hurry up and do somethin with that Havoc, then dark days will await Nvidia. One way or the other, its a win-win for the consumer. I just wish their AMD division would fare just as well against intel. Reply
  • Zool - Friday, October 02, 2009 - link

    I dont wont to be too pesimistic but availability in Q1 2010 is lame late. Windows 7 will come out soon so people will surely want to upgrade to dx11 till christmas. Also OEM market which is actualy the most profitable. Dell, HP and others will hawe windows 7 systems and they will of course need dx11 cards till christmas.(amd will hawe hopefully all models out till that time)
    Than of course dx11 games that will come out in future can be optimized for radeon 5K now while for gt300 we dont even know the graphic specs and the only working silicon dont even resemble to a card.
    Very bad timing for nvidia this time that will give amd a huge advantage.
  • Zool - Friday, October 02, 2009 - link

    Actualy this could hapen if u merge a super gpgpu tesla card and a GPU and want to sell it as one("because designing GPUs this big is "fucking hard"). Average people (maybe 95% of all) dont even know what Megabyte or bit is not even GPGPU. They will want to buy a graphic card not cuda card.
    If amd and microsoft will make heawy DX11 pr than even the rest of nvidias gpus wont sell.
  • PorscheRacer - Friday, October 02, 2009 - link

    As with anything hardware, you need the killer software to have consumers want it. DX11 is out now, so we have Windows 7 (which most people are taking a liking to, even gamers) and you have a few upcoming games that people look to be interested in. For GPGPU and all that, well... What do we have as a seriously awesome application that consumers want and feel they need to go out and buy a GPU for? Some do that for F@H and the like, and a few for transcoding video, but what else is there? Until we see that, it's going to be ahrd to convince consumers to buy that GPU. As it is, most feel IGP is good enough for them... Reply
  • PorscheRacer - Friday, October 02, 2009 - link

    Actually, thinking about this... Maybe if they were able to put a small portion of this into IGP, and include some good software with it, maybe the average consumer could see the benefits easier and quicker and be inclined to go for that step up to a dedicated GPU? Reply
  • RXR - Friday, October 02, 2009 - link

    DocSilicon, you are one funny as hell mental patient to be!. I really hope you dont get banned. You just made reading the comments a whole lot more fun. Plus, it's win win. You get to satisfy your need to go completely postal at everyone, and we get a funny sideshow.

  • - Friday, October 02, 2009 - link

    Great words but nothing behind! Fermis is Nvidias Prescott or should I say much like the last Voodoo chip that never really appeared on the market? Too many transistors are not good ... Reply

Log in

Don't have an account? Sign up now