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
shotage - Thursday, October 1, 2009 - link
lol*shakes head*
palladium - Thursday, October 1, 2009 - link
Ahh, he said a 9800 GTX + GDDR5 = 4870 !blindbox - Thursday, October 1, 2009 - link
Ooops, I think I need to speak something on topic at least. Anyone could tell me if OpenCL SDK is out yet? Or DirectCompute too? It has been over a year since GPU computing was announced and nothing useful for the consumers (I don't call folding for consumers).habibo - Thursday, October 1, 2009 - link
Yes, both OpenCL and DirectCompute are available for development. It will take time for developers to release applications that use these APIs.There are already consumer applications that use CUDA, although these are mostly video encoding, Folding@Home/SETI@home, and PhysX-based games. Possibly not too exciting to you, but hopefully more will be coming as GPU computing gains traction.
PorscheRacer - Thursday, October 1, 2009 - link
Does anyone know if the 5000 series support hardware virtualisation? I think this will be the killer feature once AMD's 800 series chipsets debut here shortly. Being able to virtualise the GPU and other hardware with your virtual machines is the last stop to pure bliss.dgz - Thursday, October 1, 2009 - link
I am also curious. Right now only nVidia's Quadro cards support this.The thing is, though, that your CPU and chipset also have to support what Intel calls VT-d.
Being able to play 3D games in virtual OS with little to no performance would be great and useful.
Not going to happen soon, though. It's also funny that virtually no one Lynnfield mentioned the lack of VT-d in 750 in his "deep" review. Huge disappointment.
wifiwolf - Thursday, October 1, 2009 - link
If there's any technology that seams to scratch that virtualization, i think this new gt300 is the one. When reading about nvidia making the card compute oriented it just drove my mind to that thought. Hope i'm right. To be fair with amd, i think their doubled stream processors could be a step forward in that direction too, coupled with dx11 direct compute. Virtual machines just need to acknowledge the cards and capabilities.dgz - Friday, October 2, 2009 - link
They already do. vmware and vbox have such capabilities. Not everything is possible atm, though.dgz - Thursday, October 1, 2009 - link
oops, I meant "little to no performance penalty" :)sigmatau - Thursday, October 1, 2009 - link
According to the super troll who keeps screeching about bandwidth, then the GT300 must be a lesser card since it doesn't have 512 bit connection like the GT200.LOL @ Trolls.