Digital Foundry at Rezzed 2017: Future of Gaming Technology - Full next-gen consoles in about 3 years with AMD RyZen

Gets interesting at about 21 minutes in - Full next gen consoles (beyond PS4 Pro and beyond Scorpio)

will certainly include AMD RyZen CPU technology. Even though advances in silicon manufacturing process is slowing down, there will be enough of an advancement for console-class silicon to make another full generational leap over the base PS4 and Xbox One released in 2013 on 28nm, especially in CPU, which was one of the main things PS4 system architect Mark Cerny stated would define the difference between console generations, along with the type of RAM and storage.

The console manufacturer should make the next gen consoles somewhere in the next 4 years at least so they can at least recoup some of the production and research costs of those mid gen upgrade consoles.

Fall 2020 at the latest. That's like 3 and 1/2 years from now.

Also, the R&D of these mid generation upgraded consoles are not a huge expenditure for the console makers (less for PS4 Pro, a little more probably for Scorpio). AMD does most of R&D on the SoCs anyway. It's not like in the old days where consoles used in-house technology (PS2) or required a massive R&D effort (PS3's Cell processor).

The base consoles from 2013 were very conservative and used slightly custom PC technology (Jaguar CPU, and GCN 1.x GPU). The new upgraded versions for 2016 and 2017 aren't much of a departure, just significantly more powerful (especially Scorpio).

@scrollinglayers:

AMD already presented resolution guidance with RX 4XX series GPUs.

RX 4XX with GDDR5 256 bit memory bus, 8 and 7 numbers = 1440p resolution. PS4 Pro falls within this guidance.

Unlike PS4 Pro and RX 470/480 products, AMD has claimed "4K gaming" for Scorpio.

Scorpio's APU size is similar to XBO's APU size.

PS4 Pro's APU chip

PS4 Slim with 209 mm^2 size chip

http://news.mydrivers.com/1/507/507389.htm

仔细观察，PS4 Pro的APU晶体面积更大，通过卡尺测量其长宽分别为22.8×14.6毫米，面积约为321.9平方毫米，而CXD90043GB的面积则为212.5平方毫米。因此，PS4 Pro的APU晶体面积就是PS4-2000大约1.57倍。

Carefully observed, PS4 Pro APU crystal area larger, measured by the caliper length and width of 22.8 × 14.6 mm, an area of about 321.9 square millimeters, while the CXD90043GB area was 212.5 square millimeters.

Scorpio's cgi APU chip with 362 mm^2.

In terms of BOM cost, Scorpio's APU cost is about 12 percent higher than PS4 Pro.

XBO's APU size is 363 mm^2

PS4's APU size is 348 mm^2

Most of Scorpio's 6 TFLOPS improvement comes from Vega's 4X perf/watt.

Vega 10's 64 CU with 12 TFLOPS target would need about 1484 Mhz clock speed and 225 watts.

Scorpio's 6 TFLOPS is half of Vega 10's 12 TFLOPS, hence 225 watts would be around 112.5 watts.

Higher clock speed increases GPU's fix function hardware performance such as ROPS, tessellation/geometry, graphics command processor, delta memory compression/decompression, L2 cache and 'etc'. This is the NVIDIA's design approach for Maxwell v2 and Pascal GPUs.

http://www.eurogamer.net/articles/digitalfoundry-2016-xbox-one-project-scorpio-spec-analysis

DF's Richard Leadbetter claimed 50 CU to 60 CU GPU for Scorpio

Vega 10 has 64 NCU with ~525 mm^2 chip area size.

Polaris 10 with 36 CU and 232 mm^2 area size scaled to 60 CU version would be around 386.67 mm^2 area size.

DF's Richard Leadbetter is a fool to claim 60 CU GPU without factoring Scorpio's APU chip areas budget!

DF's 50 CU is not dividable by four CU lanes design rule.

This is Polaris 10/RX-480's block diagram.

Each CU lane is lead by 1 geometry engine and 1 rasterizer engine. Each CU lane has it's own RB aka ROPS.

DF's 50 CU claim doesn't fit this design rule.

DF's Richard Leadbetter is a fool to claim 50 CU GPU without factoring GCN's basic design scaling rules.

Hawaii XT's R9-390X follows GCN's basic design scaling rules with 44 CU.

If use DF's Richard Leadbetter's 850 Mhz GPU clock speed estimate with 44 CU... it yields 4.7 TFLOPS... Scorpio's GPU has 6 TFLOPS target.

-------------------

The new upgraded versions for 2016 and 2017 aren't much of a departure, just significantly more powerful (especially Scorpio)

The big departure for AMD GPU is gaining NVIDIA Maxwell/Pascal style "tile cache render" feature.

The above block diagram shows AMD's soon to be replaced GPU architecture.

The above block diagram shows AMD's Vega GPU architecture i.e. pixel engines has links to L2 cache.

The above block diagram shows NVIDIA's Maxwell/Pascal GPU architecture i.e. pixel engines has links to L2 cache.

So you're saying...

If Scorpio has Vega/Maxwell/Pascal's tile cache rendering and Pascal's pack math (just integers on NV side) advantages, expect near Maxwell/Pascal style GPU with 6 TFLOPS Fp32 results.

Get 1080 Ti and reduce it's Fp32 12.9 TFLOPS Fp32 down to 6 TFLOPS Fp32 and reduce it's physical memory bandwidth down to 320 GB/s.

GTX 1080 Ti already has 2X rate Int16 and 4X rate Int8 pack math features over it's single rate Fp32 processing just like AMD's Vega NCU.

GTX 1080/1070 already has 1.22X rate Int16 and 2.44X rate Int8 pack math features over it's single rate Fp32 processing.

If you want a Vega 10 level GPU without double rate Fp16 feature, refer to GTX 1080 TI since it has most of the Vega's marketing bullet point features.

If you want a Vega 11 level GPU without double rate Fp16 feature, refer to GTX 1070 since it has most of the Vega's marketing bullet point features.

Pure TFLOPS arguments are meaningless without factoring effective memory bandwidth.

Certainly one thing that Scorpio won't have, that full next-gen consoles (PS5, Xbox Next) will have to have, is High Bandwidth Memory in the form of either HBM2 or HBM3. Along with completely new CPUs, High Bandwidth Memory will distinguish future generation consoles from PS4 / PS4 Pro, Xbox One, Scorpio

HBM2 can reach 1 TB per second memory bandwidth with 32 GB memory.

HBM3 will be able to reach TB per second bandwidth, upto 64 GB.

Something else to consider, although HBM3 isn't due until 2019-2020, it's being designed to be cheaper to produce than HBM2.

Also, we have AMD's CPU, GPU roadmap, and advances in high-end APUs.

Right now this new "Mega APU" is currently in early design stages, with no planned release date. It is clear that this design uses a new GPU design that is beyond Vega, using a next-generation memory standard which offers advantages over both GDDR and HBM.

Building a large chip using several smaller CPU and GPU dies is a smart move from AMD, allowing them to create separate components on manufacturing processes that are optimised and best suited to each separate component and allows each constituent piece to be used in several different CPU, GPU or APU products.

For example, CPUs could be built on a performance optimised node, while the GPU clusters can be optimised for enhanced silicon density, with interposers being created using a cheaper process due to their simplistic functions that do not require cutting edge process technology.

This design method could be the future of how AMD creates all of their products, with both high-end and low-end GPUs being made from different numbers of the same chiplets and future consoles, desktop APUs and server products using many of the same CPU or GPU chiplets/components.

https://www.overclock3d.net/news/cpu_mainboard/amd_reveals_a_exascale_mega_apu_in_a_new_academic_paper/1