Insider Daily. Seeking alpha as mystery continues. MS tell full DX12 support. Impossible with 7790
March 23rd, 2:43
Insider: Gpu is not standard gcn... hsa full support .. And the gpu is effectively dual logic. Multi threaded gpu core..
Each thread is 1.3tf 2.6 with out optimisation. Dp
Look here is the clue that matters most to everybody .. Its the accelerator it is the secret I do not know the real name for it.
Its not on the hot chips slides. Other then a red line from gpummu to cpu.. its not on the slides its been in the patents .. it all evolves around the oban..
I honestly cant go any further with this it is in your hands guys .. the whole soc is built on this.. it is why gpu can do the things it can why esram is so scaled. And why there is 36 micro processor that are no wear to be found its also the missing edram that is no where to be found. And its never been used ..
You know what to do.. if the community all works together they cant hide this
Misterx: So, configuration 3?
12/14 cus with smaller die space could not be more powerful than ps4 bigger CUs?
Misterx: No clears answer for config 1 or config 2 is the answer too. Config 1 is wrong?
Insider: Totally different gpu.. cu are way more powerful. . Waaaay more powerful
I need to find time to answer all the questions. . Woooooooorrrrkkkk
Misterx: Smaller die size(for GPU) but way more powerful(to PS4)? Chipworks photo shows all CUs?
Insider: Chip work does not show all cu.. they do not even bring cpu cu into account..
12/14 x 2x to 4x but it has to be engine specific. Like the gpu core logic cu are way beyond... this is not the same specification as ps4 off the shelf..
ms did the same thing with 360 architect
ps4 has an extra 4 cu for comput specific stuff .. but think kinect 2 cpu only x1 has it off board..
Misterx: "Each thread is 1.3tf 2.6 with out optimisation. Dp"
That means 2.6 DP? Am i read it correct? 2.6DP for 2 threads??? wow that a number...SP number could be 2.6 x 3-4?
Misterx: That is pretty big...not all CUs are on 1st layer...that means there is other CUs must be somewhere else...
Misterx: Still a little bit of mystery for me how 14CU on Chipwork diagram could be smaller in size, smaller overal square and be 2-3 more powerfull than PS4...
I still beleive more in Mictercteam digging about thouse 12 CUs on 1st layer are Dxcore only(for apps and UI). By HSA paradigm Dxcore is supporting GPU for CPU...i asked you many times about Dxcore and 1st layer...there was allways no clear answer from you...
should be NDA...but a hint that on 1st layer there is not all CUs means that there is another layer with CUs...
Also what about sdk hardware? why MS provided 2x7970 SDKs while developers could only use 1st 1.31TF SP layer?
Misterx: Also you told about hight frequincy for xbox one. Official numbers see much lower frequency...
Is turbo clock tech(future clock upgrade) is still possible and planned? that means that 1.3TF DP is not final number for real GPU?
Misterx: Is it save continue to say there is 1.3Tf SP DXcore(1st layer in HSA terms equals CPU) and 1.3Tf DP GPU inside Xbox One?
What do you mean by DP?
a) x2 = 2 treads
or
b) floating point double precision that equals x3x4 more power needed compared for SP?
Misterx: So basivally new info confirms all Mistercteam digging...because it is impossible to have on Chipwork diagrams to have smaller and less CUs and be more 2-4 times more powerfull than PS4...
Misterx: Another theory that it is 1.3TF DP we see on 1st layer of Chipworks but thouse little CUs could do 3-4 times more that bigger CUs of PS4. Something like engine specific that produces less idlle and stalls and do more work due to hardware help of 2nd layer. Also there is 2 of 1.3Tf DP cores...the second one on 2nd layer.
Now Xbox One uses only one 1.31Tf part of first layer without engine specific stuff and produces 1.31TF SP output. With engine specific tools and better management(use of all DX12 features)) it could produce 1.31TF DP values( or 1.31Tf SP x3-x4 with help of DX12). Stereo means there will be access to second 1.31 Tf DP part...with DX12 in engine that will allow to render 3-4 objects compared to 1 of standart GCN core.
It is something like this:
(1.31TF SP - mono driver + 1.31TF SP - stereo driver) x 3 or 4 by DX12 optimised engines
= (1.31SP-mono + 1.31SP-stereo) x3 or x 4 by DX12 special hardware to render 3 or 4 more objects because of:
- less stalls and idle
- new command buffer management
- with help of new memory management(like ESRAM, move engines, EDRAM and etc).
How that hardware is distributed by hardware layers is another question...and Mistercteam has all the answers. But seems like we have 2.6DP here...DP is not double precission of floating point calculations but it equals to that perfomance gain if use DX12 new features that allows to render 3-4 times more objects with the help of new co-processors and new management system while TF number stay the same.
Mystery continues...
But we can be sure that 7790 GPU as Digital Foundry and other clueeless journos told us could not have full DX12 hardware support.
By their logic al better than 7790 GPUs including PS4 77850 GPU is full DX12 compatable. By their logic that means that DX12 is nothing special and only reduces overlays.
There is still strong beleive in that and MS should show all DX12 features other than 50% less CPU Mantle like utilisation. MS haters think AMD made Mantle by their own while in reality it is MS who helped AMD with Mantle and gave this low level access from console Directx version.