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
A More Efficient Architecture
GPUs, like CPUs, work on streams of instructions called threads. While high end CPUs work on as many as 8 complicated threads at a time, GPUs handle many more threads in parallel.
The table below shows just how many threads each generation of NVIDIA GPU can have in flight at the same time:
| Fermi | GT200 | G80 | |
| Max Threads in Flight | 24576 | 30720 | 12288 |
Fermi can't actually support as many threads in parallel as GT200. NVIDIA found that the majority of compute cases were bound by shared memory size, not thread count in GT200. Thus thread count went down, and shared memory size went up in Fermi.
NVIDIA groups 32 threads into a unit called a warp (taken from the looming term warp, referring to a group of parallel threads). In GT200 and G80, half of a warp was issued to an SM every clock cycle. In other words, it takes two clocks to issue a full 32 threads to a single SM.
In previous architectures, the SM dispatch logic was closely coupled to the execution hardware. If you sent threads to the SFU, the entire SM couldn't issue new instructions until those instructions were done executing. If the only execution units in use were in your SFUs, the vast majority of your SM in GT200/G80 went unused. That's terrible for efficiency.
Fermi fixes this. There are two independent dispatch units at the front end of each SM in Fermi. These units are completely decoupled from the rest of the SM. Each dispatch unit can select and issue half of a warp every clock cycle. The threads can be from different warps in order to optimize the chance of finding independent operations.
There's a full crossbar between the dispatch units and the execution hardware in the SM. Each unit can dispatch threads to any group of units within the SM (with some limitations).
The inflexibility of NVIDIA's threading architecture is that every thread in the warp must be executing the same instruction at the same time. If they are, then you get full utilization of your resources. If they aren't, then some units go idle.
A single SM can execute:
| Fermi | FP32 | FP64 | INT | SFU | LD/ST |
| Ops per clock | 32 | 16 | 32 | 4 | 16 |
If you're executing FP64 instructions the entire SM can only run at 16 ops per clock. You can't dual issue FP64 and SFU operations.
The good news is that the SFU doesn't tie up the entire SM anymore. One dispatch unit can send 16 threads to the array of cores, while another can send 16 threads to the SFU. After two clocks, the dispatchers are free to send another pair of half-warps out again. As I mentioned before, in GT200/G80 the entire SM was tied up for a full 8 cycles after an SFU issue.
The flexibility is nice, or rather, the inflexibility of GT200/G80 was horrible for efficiency and Fermi fixes that.
Facebook
Twitter
RSS
415 Comments
View All Comments
SiliconDoc - Wednesday, September 30, 2009 - link
Nice rebuttal to page 2: " Another kind of LAUNCH "--
write it down, nvidia launched today....(according to lunatic lying red roosters)
tamalero - Wednesday, September 30, 2009 - link
weird.. they still said its "coming soon", I dont see any GF300 firm chips.when ATI said "we present the 5870" they were already on newegg.com
Silicon, let's face it, you're the biggest pro-nvidia troll I've ever seen.
SiliconDoc - Wednesday, September 30, 2009 - link
You are also the person that went into a tirade about nvidia not replacing laptop gpu's with the faulty substrate and instead puttig on a heftier fan.You waxed on about how much you hate nvidia, and how they harmed the children (you claimed to be a teacher of some sort) then you screeched about nvidia reps, wished violence upon them, and claimed you'd love to show them how to do their jobs correctly.
---
That's YOU tamalero.
--
Now it's pretty amazing I tell the simple plain truth, you deny it a week late, lying for ati, have you public hate and rage on this board for nvidia, and yet claim it is I that is a fanboy.
--
One Q, has your raging hatred for nvidia receded, or does lying about the 5870 release give you a sense of vengeful pleasure ?
tamalero - Friday, October 2, 2009 - link
what truth?you're just inventing random crap your brain somehow imagines in illusions.
and what the hell are you talking about?
I never claimed to be a "teacher", wished violence? what the hell are you smoking?
harmed the children.. jesuchrist... are you on some sort of scientologist brainwashing group ?
SiliconDoc - Friday, October 2, 2009 - link
Since you have lied, I will get the link and your quotes.SiliconDoc - Wednesday, September 30, 2009 - link
no they wre not already on newegg - listed and greyed out- the first one available in a trickle -and only today have those listed appeared available, before that it was on for a few seconds, card gone - all GREYED OUT again.---
Sept. 23rd was launch, this is 7 days later.
They were a WEEK of paper. (no one can fairly count a sickly 1,2 or half dozen trickle)
tamalero - Friday, October 2, 2009 - link
they were grey, because they sold out, note..., there were on amazon and tigerdirect.com as well. I woudlnt be surprised if newwave and other sites had the 5870 as well.you're just a person with mental problems who cant really accept anything outside your tiny world.
SiliconDoc - Friday, October 2, 2009 - link
TigerDirect was pre-order, as well as Amazon was reserve - you just haven't got clue one.fikimiki - Friday, October 2, 2009 - link
In Poland, (it is Europe cause you don't know for sure)it is available in shops.
Also you can grab one from newegg.com
bobvodka - Thursday, October 1, 2009 - link
I woke up on HD5870 launch day.I logged onto a website in the UK.
I ordered an HD5870.
It shipped the same day.
I had it the next day and have been enjoying it ever since.
Looks like a non-paper launch to me.