I don't see how a 500+mm^2 Vega doesn't easily beat a 1080 and at least compete with a Titan XP.
Note, that may be only if the game is programmed to use primitive shaders.
Man if Xbox Scorpio uses even a small Vega well optimized with the features we learned about today it could be an insanely cool native 4k gaming machine. I really think Sony messed up the timing here with their Polaris based PS4 Pro. However this thread is about Vega so I digress...Looking forward to the new Xbox scorpio games.. Seems like Vega is a keeper for the years to come.
Nope, it's HW based, not needed to have support in SW.Note, that may be only if the game is programmed to use primitive shaders.
the updated geometry engines will also feature one more advancement, which AMD is calling the primitive shader. A new shader stage that runs in place of the usual vertex and geometry shader path, the primitive shader allows for the high speed discarding of hidden/unnecessary primitives. Along with improving the total primitive rate, discarding primitives is the next best way to improve overall geometry performance, especially as game geometry gets increasingly fine, and very small, overdrawn triangles risk choking the GPU.
Only thing you need, is drivers. And thats it.Nope, it's HW based, not needed to have support in SW.
From Ryan's article:
Nope, it's HW based, not needed to have support in SW.
From Ryan's article:
Today's Radeon GPUs retain fixed-function geometry-processing hardware in their front ends, but the company has observed that more and more developers have been doing geometry processing in compute shaders...
To accomodate developers' increasing appetite for migrating geometry work to compute shaders, AMD is introducing a more programmable geometry pipeline stage in Vega that will run a new type of shader it calls a primitive shader. According to AMD corporate fellow Mike Mantor, primitive shaders will have "the same access that a compute shader would have to coordinate how you bring work into the shader." Mantor also says that primitive shaders will give developers access to all the data they need to effectively process geometry, as well.
Factually you are both correct. Primitive Shaders give much more control over geometry pipeline to developers. For this, you would need an update to the game, that is correct. However, Primitive Shader job is also to cull objects invisible to the observer before they are even drawn. For that, you will need only Driver update, its hardware stuff.From Tech Report's article:
(Emphasis is mine)
It sounds to me like developers need to due something to utilize this new shader type.
The new programmable geometry pipeline on Vega will offer up to 2x the peak throughput per clock compared to previous generations by utilizing a new “primitive shader.” This new shader combines the functions of vertex and geometry shader and, as AMD told it to me, “with the right knowledge” you can discard game based primitives at an incredible rate. This right knowledge though is the crucial component – it is something that has to be coded for directly and isn’t something that AMD or Vega will be able to do behind the scenes.
how important is this?So no better geometry performance than polaris.I dont like it because polaris is still very behind even GP106 in geometry performance...
That 2x geometry performance is only if developers optimize for vega...
Also 4x shader engines/geometry procesors are confirmed same as Fiji...
And most likely only 64rops.
Wow, nice geometry throughput increase... so no more geometry bottleneck on AMD cards?
how important is this?
Those chip resistor packages look like 1.4mm with a 2.0 mm pitch between packages. That puts my estimate at about 25x19. But the spacing is not even so its really not much more than a guess. I cant see it being much over 500 though. How do we know those dies are 11.87mm?
Looks like it might be another situation where the hardware is going to be ahead of the software. So the performance will be lacking of it's full potential for a good amount of time, but will give good performance over a much longer period of time. If they work with the major engines to get at least partial functionality this might be quicker than with GCN.
Did you get a chance to ask if adding support for some of these new features is something AMD can do for AAA titles using shader replacement in game ready drivers, or are they more fundamental than that?I chatted with Scott Wasson and other AMD reps yesterday at CES and have published my coverage here:
http://techbuyersguru.com/ces-2017-amds-ryzen-and-vega-revealed
While Scott provided me the full rundown on the architecture, I was able to glean a bit more on Vega based on demos and a bit of deduction.
Did you get a chance to ask if adding support for some of these new features is something AMD can do for AAA titles using shader replacement in game ready drivers, or are they more fundamental than that?
Since it appears no one has posted this, I went ahead and rotated the best image of vega I could find:
It is actually still off by a tiny amount, and it is also not exactly flat with respect to the camera lens, but its close enough for government work. My best pixel analysis yields a die size of 542 sq mm +/- 3.
I used both dies as a reference to get a more accurate result
Right die = (9*3*11.87/12)*(6.83*3*11.87/12) = 541.3
Left die = (9*3*11.87/12)*(6.86*3*11.87/12) = 543.7
Geometry bound processes will still be geometry bound. Those should be much faster now though.
Pretty important. The biggest lead nvidia has is probably in geometry throughput.
Most people are thinking that there are 4096 CU in this, based on information that's been provided about the FirePro adapters. The numbers honestly don't add up for me, TBH. The rational for that is the scaling that AMD got with Polaris. It wasn't as good as expected, but they still got 2304 CUs and a 256-bit bus into a 232mm² die. A 380X was 2048 shaders and a 256 (384) bit bus in a 352mm² die, so even dropping 33% of the physical MC, they packed 12.5% more shaders into 66% of the space. That's +90% CU per total die mm² going from GDDR5 GCN3 to GDDR5 Polaris. If Vega is even 500mm² and 4096 CUs, that would mean they've gone to in interface half as wide, and they still only managed to get the same number of shaders into 80% of the space, or +25% the CU/mm² going from HBM GCN3 to HBM2 Vega. It could be that all the changes they made with NCU really just take that much more space, but that would seem really poor IMO.how many compute units would be in this though. 14 nm 500+mm^2. not likely 4096 right?
in useful cases, not just "I want 64 x tessellation and i will have it!"