I am not exactly sure what you are trying to say here. I'd say you are willfully ignoring what is written in the DF coverage. Their info is there, draw whatever conclusion you want from it, it still doesn't change what is written there.
ID buffer and DR FP16
- Thread starter Jamend
- Start date
-
- Tags
- id software buffa
I am not exactly sure what you are trying to say here. I'd say you are willfully ignoring what is written in the DF coverage. Their info is there, draw whatever conclusion you want from it, it still doesn't change what is written there.
This is a mixture of an observation and an evaluation here. We definitely don't know if one leads to the other. Some other things you should have observed is that MS is heavily marketing Scorpio as a 4K console. They are doing their best to separate themselves from their nearest competition.
One could easily believe that MS would rather stay silent on the topic of ID Buffer since it shows a lack of raw power to drive 4K, which has been their focal point for their marketing campaign.
I'm with the other school of thought here. I've never known a marketing campaign exclude a check-box feature their rivals have to try and make themselves look better. Marketing 101 is 'we have everything they have and even more', especially when you cost more. MS's history to me suggests they'd take the same tech and brand it with some fancy name if it was in there, rather than leave Sony with an unchecked PR volley that undermines XB1X's superiority. Fancy branding as an image enhancement co-processor or such-like could shift the focus of the ID buffer to non-upscaling functions (where of course it'll actually find all its use
).
I'm just providing perspective here as to why they may not market it. As many have been ripping into MS marketing of true 4K, I don't know if you want to also advertise the ability to upscale/checkerboard. Which is it really?I'm with the other school of thought here. I've never known a marketing campaign exclude a check-box feature their rivals have to try and make themselves look better. Marketing 101 is 'we have everything they have and even more', especially when you cost more. MS's history to me suggests they'd take the same tech and brand it with some fancy name if it was in there, rather than leave Sony with an unchecked PR volley that undermines XB1X's superiority. Fancy branding as an image enhancement co-processor or such-like could shift the focus of the ID buffer to non-upscaling functions (where of course it'll actually find all its use
They have done a decent job explaining what Scorpio brings to the table. And I think it's resonated with the crowd.
It's hard to fault their marketing when it's been successful. once again I'm not sure if 1X has the hardware capabiltiy either, I say no for different reasons posted earlier. I'm just not putting "lack of marketing checkbox" as my main driver for that reason.
Actually they did, or more specifically, Andrew Goosen did, and its in the first post that I quoted. Its in the very article you quoted.
It's not my main driver either. Shouldn't be anyone's. Given an ambiguous situation (Goossen says, "We have hardware techniques for making checkerboarding very efficient," not just 'hardware', which is weird phrasing, and Sony apparently have the patent on an ID buffer), we look to other indicators, which seem to weigh on the negative side IMO. Thus in the balance I think it sides fairly comfortably one side. I think historically every time we've debated over whether a feature is present in a console or not based on similar ambiguous data, it's proven absent, though I'm not keeping score.
Silent_Buddha
Legend
I'm with the other school of thought here. I've never known a marketing campaign exclude a check-box feature their rivals have to try and make themselves look better. Marketing 101 is 'we have everything they have and even more', especially when you cost more. MS's history to me suggests they'd take the same tech and brand it with some fancy name if it was in there, rather than leave Sony with an unchecked PR volley that undermines XB1X's superiority. Fancy branding as an image enhancement co-processor or such-like could shift the focus of the ID buffer to non-upscaling functions (where of course it'll actually find all its use
One thing comes to mind immediately here. NVidia's tiled rendering. They basically said little to nothing about it, and it was up to consumers and tech sites to figure out what was happening on their hardware. It was something that gave them a competitive advantage WRT to their competition, but they kept quiet about it so their competition couldn't start R&D on something similar. Because of this AMD are significantly behind WRT to tiled rendering and the benefits it can provide.
Regards,
SB
Anarchist4000
Veteran
Or it functions extremely similarly to Imagination's TBDR and patents. Coming from Nvidia's work on mobile at the time. Not a huge stretch considering they won't open up Maxwell drivers on Linux. Could be more legal than competitive, but even to this day the feature isn't acknowledged to the best of my knowledge.
Some of the commentary after the trianglebin test was disclosed indicated that there were suspicions. I think one of the early Techreport reviews of Maxwell had a somewhat obscure set of binning puns that might have alluded to this. How much some of those more connected in the industry may have known is unclear.
Nvidia's choice may not have been broadcast for consumers, but even so it's not like there wasn't a range of tiled renderers (deferred or not) from IMG, ARM, and finally Qualcomm's Adreno.
The last case specifically meant AMD would have been aware of this direction. The patent for AMD's binning rasterizer was filed in 2013, before Maxwell's launch.
The two data points indicate to me that there was sufficient visibility for AMD to have the information it needed in this direction.
Nvidia admitted to the tiling after the trianglebin article. Some of their engineers confirmed it.
If there are patent concerns, they would be from a smaller section of IMG's portfolio, since there is no hidden surface removal and rasterization wasn't deferred to a second pass.
That would be a secret sauce. In this case Sony have already let slip what the tech is, rather than MS having implemented it and keeping quiet so Sony don't find out. Also, in the GPU space nVidia are competing with benchmarks from their rival, so don't need to explain the tech as long as they get the scores. In the console space, DF et al might report that Console A always gets 30% more FPS than Console B after they release, but prior to that it's marketing mud-slinging that helps win sales, and even after the fact. The magic of Cell helped sustain PS3 interest even when it was reported the system as a whole was producing inferior versions of many cross-platform titles.
I suppose the real analogy here would be AMD introducing tiled rendering and catching up with nVidia, but not telling anyone they've made added this feature leaving the public believing AMD were at a technical disadvantage.
You're right, I missed it.
Since FP16 is extremely difficult to use for a Compiler it does not improve the performance as much as some might think. The instructions must always have both operants in the same register and the same operation must be performed on both.
In addition, a lot is only FP32 and then has to be converted from FP16 to FP32 and back. Which again requires more instructions and registers. The register is quite a problem with GPUs. If FP16 has to be converted to FP32 back and forth the register location advantage is gone.
On current GPUs the benefit is very low while future GPUs could support FP16 more efficiently.
In addition, a lot is only FP32 and then has to be converted from FP16 to FP32 and back. Which again requires more instructions and registers. The register is quite a problem with GPUs. If FP16 has to be converted to FP32 back and forth the register location advantage is gone.
On current GPUs the benefit is very low while future GPUs could support FP16 more efficiently.
Last edited:
Late reply, but minor point: The quad isn't stretched into a rectangle, but actually a slanted rectangle. It also flips slant if you invert the checkerboard pattern each frame.
That's assuming you use the even/odd CB scheme (You could do something like what shadow fall did in it's multiplayer, with skipped rows or columns of pixels, but this is almost always worse off).
