The only reasons why there's no GP102 for gaming, yields of a bigger ~450mm2 chip will be worse.
If Polaris 10 doesn't get GDDR5X, its bottlenecked. Then a 2500-2800sp cut down Vega 11 part is going to be much faster. Not to mention it could have another TMU/ROP layout as well.
Vega 11 is the competitor to GP104. Polaris 10 is just too small to compete and it will face off versus GP106 as expected.
If GP104 really only has 2560 shaders, Polaris 10 might actually beat it.
But you just wait, when Pascal GTX debuts, the VR hype from NV is gonna be all aboard that latency train.
I think if GP100 launch is any indication, Nvidia may be trying to win this generation with clockspeed. 2560 cores but if base clock is 1.5 or 1.6ghz you are looking at 60-65% faster than a stock 980, and about as fast as an overclocked 1.3-1.4ghz 980ti.
980TI is only 22% faster than GTX980.If Gp104 is 65% faster than GTX980 then it will be 40% faster than 980TI.
i really dont know why everyobne overestimate 980Ti performance.Its really not some miracle card.
Its just 20-23% faster than GTX980.
Btw if its 40% faster than GTX980Ti then 980Ti cant match it even at 1500Mhz.
Btw i also think polaris 10=Gp106
gp104 is 300mm2 SKu.Polaris cant match it with 232mm2, but it will probably beat 200mm2 GP106/1060
We don't REALLY know that polaris 10 is 232mm^2. I mean I would say that Polaris 10 is actually NOT the chip you are talking about, and rather that one, on the linkedin profile, is something that will never come to retail, especially since it was posted publicly. That's my opinion on that matter.
Plus it is a denser process AMD is using too.
So you can tell VR gamers to not move their heads until they are at a graphics draw call completion?
Don't be silly man.
That Async Timewarp needs to fire as soon as people move their heads, that's the entire point of it. You cannot control when people move to fall in-line to prevent stalls of Async Timewarp. -_-
But you just wait, when Pascal GTX debuts, the VR hype from NV is gonna be all aboard that latency train. We can come back and discuss how you are wrong, again.
Plus everybody keeps forgetting about the primitive discard accelerator which ought to help with memory bandwidth as well. I think a 232mm2 Polaris can be close to a 294mm2 Pascal.
you can do it in the way Nvidia is doing here by preempting it, which basically just means that it gets inserted into the rendering pipeline at the most opportune time (instead of having to wait for another draw call to finish first.
The standard VR pipeline from input in (when you move your head) to photons out (when you see the action occur in-game) is about 57 milliseconds (ms). However, for a good VR experience, this latency should be under 20ms.
Combined, and with the addition of further NVIDIA-developer tweaks, the VR pipeline is now only 25ms.
Which wasn't possible on Maxwell and older, because it has to wait for graphics to finish first before being able to context switch the pipeline to handle compute.
With Pascal, that's an instantaneous change very much like GCN.
It works best if the compute timewarp can run in parallel, as Async Compute.
It works okay if the timewarp can go on a priority context and there's no delay for context switching. This is Pascal's uarch change, based on that article.
It works the least well the current way.
Why do I say this is "basic" Async Compute?
Because currently NV GPUs actually take a performance hit when AC is run. This is again because of their slow context switch. It causes stalls in the pipeline, wasting time where no work can be done.
The change in Pascal means even if they cannot run graphics + compute in parallel, devs calling for Async Compute, or even general games that use a lot of compute, won't cause stalls. In theory, it should behave like GCN where the graphics/compute context switch is fast.
Which wasn't possible on Maxwell and older, because it has to wait for graphics to finish first before being able to context switch the pipeline to handle compute.
13ms comes from the 75hz display. :\This is why NV say Maxwell at the BEST is only capable of 25ms motion to photon latency via async timewarp. Still above the 20ms recommended.
http://www.geforce.com/whats-new/ar...us-the-only-choice-for-virtual-reality-gaming
This is possible today, too - after a draw call.Basically with Pascal, it's much faster, what they call "fine-grained preemption", more flexible, anytime. And compute on priority can even override the current graphics task if that is needed.
Just Pascal no longer incurs a performance penalty from async compute, doesn't mean that it is supporting async compute, not even a "basic" async compute. The whole point of async compute is to improve performance, not simply maintain a status quo in performance.
Pascal no longer incurring a performance penalty, simply means that they have fixed their "workaround" (preemption) for not having async compute support. Of course instruction level pre-emption will likely be useful in many other areas as well, so it's not just a fix for lack of async support.
@sontin
Read and learn man. You come up with such random stuff that goes against what even NVIDIA tells developers about what their hardware is capable of.
@sontin
Read and learn man. You come up with such random stuff that goes against what even NVIDIA tells developers about what their hardware is capable of.
https://developer.nvidia.com/sites/...works/vr/GameWorks_VR_2015_Final_handouts.pdf
^ NV claims it supports priority context...
^ They even say their priority context takes over the whole GPU and preempts whatever it's working on to switch to the new task... LOL
^ Except it can't. It's not actually a priority context at all. It gets stuck in traffic like everything else.
It's the same fault they had when they claim their hardware supports Async Compute... except it can't.
Pascal changes this, it's a major change for them. Pascal has real priority preemption and instant context switching of graphics/compute rendering.
Celebrate it instead of talking silly like Maxwell can do it when NV clearly says it cannot.
Need I remind you, you were so insistent that Maxwell also supports Async Compute. There were lots of threads where I tried to educate you otherwise, but nope, you were determined... and wrong.
And in none of these slides is mentioned that preemption has a penatly.
You are the one who doesnt care what is said and making fanfiction. Stop it. :thumbsdown:
Nvidia is making up fanfiction?? What??
It's $ per wafer they can get. If a bunch of dies goes to $12k Tesla, a bunch more goes to $5K Quadros, the rest can be $1.5K GTX Titans. Overall they will make a lot per wafer that way. Telsa/Quadro in effect, subsidizes the existence of a huge high-end gaming chip from NV. Like it has for many generations.
There's only so far you can cut though before it starts to not make sense.
The P100 is 56 SMs. I bet they could sell a 50-52 SM cheaper Tesla and then 45-50 Quadros sometime in early 2017.
Theoretically the full GP104 is 40? (2560). I guess doing 2560 cores with high clock speeds shouldn't be a surprise but at 300 mm2 it's way too big for the price range anticipated. This node is going to be all about transistor usage efficiency and it looks bad for nVidia right now. Maybe that's why Volta also has shown up on the radar so soon because nVidia knows that Pascal is basically Fermi 2.0.