Jen-Hsun, however, wasn’t done. Waiting until the end of his presentation, he also announced that we’re launching the NVIDIA GTX 1070 as well, with the “Founders Edition,” available June 10th for $449, and custom boards from partners expected to start at $379.
The NVIDIA GeForce GTX 1080 “Founders Edition” shown tonight will be available on May 27 for $699. It will be available from ASUS, Colorful, EVGA, Gainward, Galaxy, Gigabyte, Innovision 3D, MSI, NVIDIA. Palit, PNY and Zotac. Custom boards from partners will vary by region and pricing is expected to start at $599.
...but anyone wanting 2GHz stock GPU clocks must have surely been disappointed by this :hmm:
Nope, a viewport does NOT allow to render something from different point of views. It can only translate and stretch vertices on the image plane (it's a simple 2D transformation), you can't change perspective with it. Single pass stereo has to be a completely new feature on Pascal, totally independent from rendering to multiple viewports.seems they just bumped it up to 16 from maxwell's 9.
Nothing much on the feature front at all. Just more shaders and higher clocks on the hardware front.
Yes, realistically like me, however quite a few (the usual suspects) were predicting 2GHz clocks & great(er) efficiency at the same time.Were there many?
Most people expected MUCH lower clocks, or high clocks with a much higher power consumption...
How can you tell just by looking at it, that it isn't gimped?
It still a wait and see ballgame here, we don't have the full details yet, so, going by how something looks is just speculation.
Lol. Man I'm getting old.
Nope, a viewport does NOT allow to render something from different point of views. It can only translate and stretch vertices on the image plane (it's a simple 2D transformation), you can't change perspective with it. Single pass stereo has to be a completely new feature on Pascal, totally independent from rendering to multiple viewports.
A typical VR renderer only renders to 1 viewport per eye, but you have to submit geometry twice to the GPU (once per eye). On Pascal you cut the geometry workload in half by rendering both left and right eye in a single pass.
Nope, a viewport does NOT allow to render something from different point of views. It can only translate and stretch vertices on the image plane (it's a simple 2D transformation), you can't change perspective with it. Single pass stereo has to be a completely new feature on Pascal, totally independent from rendering to multiple viewports.
A typical VR renderer only renders to 1 viewport per eye, but you have to submit geometry twice to the GPU (once per eye). On Pascal you cut the geometry workload in half by rendering both left and right eye in a single pass.
Yes, realistically like me, however quite a few (the usual suspects) were predicting 2GHz clocks & great(er) efficiency at the same time.
Suffice to say that never happened, 16nm FF+ is good but not that good
Just an fyi.. Msi has dropped prices of its 980ti in many markets. I can see in Australia the prices are lowered by 150AUD. anything happening in the US market?
https://www.pccasegear.com/products/32234/msi-geforce-gtx-980-ti-gaming-6gb
That's incorrect. MRTs allow to render to up to 8 render targets simultaneously, but you are rendering the very same geometry with the very same shader (the shader can output different data to each render target).MRT has been possible for a long time, and that demo they did, didn't really show anything that wasn't currently possible.
You can have different shaders (even async shaders) on each rendering target as well, so I am really curious as to the specifics of what exactly is different.
Newegg had a Zotac 980 Ti for $550 a couple days ago, I think that's the lowest it's gone...
If you are doing the two projection at the same time *by definition* you are not sending the geometry twice to the GPU Projecting a vertex is super cheap (just a vector-matrix multiplications and a bunch of divisions) and Pascal is not saving you anything here since it will still have to project each vertex onto each eye. The saving comes from not having to send megabytes of geometry to the GPU twice and not having to transform all of it twice (only position needs to be transformed twice, all the rest (normals, UV coordinates, bones, etc..) gets transformed once.Got to wait until more information comes out and its actually used. It sounds like the key is not rendering the scene twice but doing the projections at the same time. Sounds like a software feature that can be exploited by a GPU that can do multiple viewports. It does not seem like maxwell could not. it seems more a VRworks improvement. They can't cut out the actual rendering, just do it at once. - layman
It is, though you'd have to cite Intel's CPUs for something similar in the FF space. For instance most Skylake (non K) CPU's have stock clocks closer to 3GHz & a TDP of 65W, the 6700K is 4GHz stock with 91W TDP.Hm, it is not really easy to compare but:
GP104 has 10% less transistor than GM200 and running at a 60% higher base clock with 20% less power (TDP 180W vs realworld of 225W).
So it provides a huge leap forward from 28nm.
The point is expecting 2GHz clocks without a major regression in efficiency is foolish even if it is 16nm FF+, there are very real & hard limits which cannot be overcome even with a node shrink. Going forward we may (possibly) even get a regression of stock clocks on 10nm, but that debate is for another time.
New asynchronous compute advances improve efficiency and gaming performance.
Were there many?
Most people expected MUCH lower clocks, or high clocks with a much higher power consumption...