NVIDIA's Fermi: Architected for Tesla, 3 Billion Transistors in 2010
by Anand Lal Shimpi on September 30, 2009 12:00 AM EST- Posted in
- GPUs
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.
Facebook
Twitter
RSS
415 Comments
View All Comments
SiliconDoc - Friday, October 2, 2009 - link
Yeah, of course, 3 million T core, and it doesn't need much bandiwth, and it won't be used, perhaps, or probably, because the GPU designers haven't a clue, and of course, you do.---
Another amazing clown with a red nose. You people really should stop typing stupid stuff like that.
Calin - Sunday, October 4, 2009 - link
There was no performance improvement from increasing the bandwidth of the Athlon64 processors due to the move to DDR2 memory (a theoretical doubling of performance, with about one and a half more measured bandwidth in the first generation of the processor).I might not have a clue, but do you?
SiliconDoc - Wednesday, September 30, 2009 - link
Very interesting of course, and so with your theory, it could also be LOWER, with slower ddr5, but the fact REMAINS, 240 has ALREADY BEEN LEAKED.so we know what it is, and Anand KNOWS IT'S MORE, BUT NOT DOUBLE AND SAYS SO !
---
SO WHAT WE HAVE IS A CONVENIENT COVER UP. PERIOD!
--
JUST NOT AS STUPID AS YOU, THAT'S ALL.
dragunover - Thursday, October 1, 2009 - link
"SO WHAT WE HAVE IS A CONVENIENT COVER UP. PERIOD!--
JUST NOT AS STUPID AS YOU, THAT'S ALL. "
I disregarded anything you said before and after that.
SiliconDoc - Friday, October 2, 2009 - link
I'm certain you attempted it, as you no doubt love to wallow in blissful ignorance and denial just by mere unchangeable habit.But having read it, you have nowhere to go. Your mind is already irreparably altered. Congratulations.
jonGhast - Wednesday, September 30, 2009 - link
True, you seem to be a whole new level of stupid.mapesdhs - Wednesday, September 30, 2009 - link
I have a theory on that one: faulty keyboard. Every time he hits Shift
to get upper case, his keyboard is zapping out brain cells with EMF bursts. :D
Best to just ignore & not reply IMO.
Ian.
SiliconDoc - Thursday, October 1, 2009 - link
Well you can't ignore the discussion, and as far as that goes, you want everyone else to do your will, as you beg for it, with your empty advice, which is by the way, all you provided in the last thread.So you quack around telling others not to participate. That's your whole gig mary.
I find the stupidity level amazing, as most of you can only spew in and beg eachother not to comment, and by the quality of the comments that actually try a counter, I certainly cannot blame you, for begging others to surrender ahead of time.
You notice how many new names are here ? lol
It's clear who and what it is that responds, and what level of conduct they are all about.
Now let's see the Sybillic mapes sign off on his/her own glorious advice, which she/he failed to follow already.
Silverel - Thursday, October 1, 2009 - link
Hey buddy. Take a pill and relax.Your product isn't on the market yet. There's no recommendation to be heeded. It doesn't matter how fast it is, you can't have one. No one can. If someone wants to get their next-gen performance on now, ATI is what you'll buy.
It's okay though. Not everyone is going to run out and grab an ATI card. There's plenty of people that will wait for hell to freeze over for nVidia to release a new card. Personally, I'll take the Q1 2010 release date as semi-fact, but hell freezing over is my fallback date.
So feel free to continue ranting like you've run out of meds. Insult anyone who has made a comment contrary to your own. It's not really doing you any favors, but that's okay. I'm sure it's having an effect on the opinions of people sitting on the fence. Ah, and just to dig the finger in the wound, you can consider me an ATI shill, just for good measure.
SiliconDoc - Thursday, October 1, 2009 - link
Gosh you';re a sweetheart, too, and wrong ! WRONG ! WRONG !the red bloviator> " . If someone wants to get their next-gen performance on now, ATI is what you'll buy. "
Gee, did you lose it that easily, was your heart beating so hard, were you sweaty, and upset, and out of control, and decided it was great to tell me to settle down, when your brain was on FART ?
Please see the GTX295 that BEATS the 5870. The 5870 is PRIOR GEN performance - as ati's own 4870x2/or CF is equal.
Golly, what a deal, another red rooster massive ruse.