• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

[REVISED] Sony & Microsoft 10th-Gen Speculation: SONY EDITION

Will Sony release PS6 in JP, do something else for JP, or Skip JP Altogether?

  • Yes, Japanese release day-and-date

    Votes: 21 70.0%
  • Yes, Japanese release but months after America/Europe

    Votes: 6 20.0%
  • No Japanese release, some other device for Japanese market instead (portable)

    Votes: 0 0.0%
  • No JP release, no JP device, PC/mobile stream client for Japan

    Votes: 0 0.0%
  • No JP release, no JP device, PC/mobile stream & NATIVE play client for Japan

    Votes: 0 0.0%
  • No JP release, no JP device, no JP PC/mobile ports/service, JP abandoned :(

    Votes: 3 10.0%

  • Total voters
    30
  • Poll closed .
NOTE: This thread is EXCLUSIVELY for Sony discussion; for Microsoft discussion here is the companion thread!!!

---

Picking back up from the mid-gen refreshes speculation, here's revised 10th-gen speculation. This thread will focus on the Sony side. If any questions over how I arrive at some things (like HBM4), ask them in comments. Most of what you think is random, uses real data trends to extrapolate from. At end of day, this only speculation.

Long read, have been warned. Have fun!

--------

69309_01_new-playstation-5-renders-show-radical-different-design-again_full.jpg

(Concept render of a PS6. Glad the old thread still had this linked ;) )

[SONY PLAYSTATION 6]

>More or less continues traditional console design philosophy

>Comes with an Entry-level PSVR Gen 3 helmet for standard SKU

>Revised DualSense controller, now full VR-capable. Unique three-piece design, revamped haptics,
positional/motion/acceleration/proximity detection. Center piece usable as a room-scale calibration
and tracking module (extras can be purchased separately; wireless connection to mesh grid
network for expanded, scaled tracking)

>Significantly revamped OS for VR-native navigation, features, etc.

>Optional VR controls will also be available for users who prefer it​

amd-ryzen-678_678x452.jpg

[CPU]

>CORES: 8 (CPU block 1), 8 (CPU block 2)

>THREADS: 16 (CPU block 1), 8 (CPU block 2)

>CLOCK: 5.2 GHz (CPU block 1), 5.2 GHz (CPU block 2)

>L1$: 64 KB (per core, CPU B1), 32 KB (per core, CPU B2)

>L2$: 1 MB (per core, CPU B1), 1 MB (per CCX, CPU B2)

>L3$: 8 MB (all, CPU B1), 4 MB (CPU B2)

>L4$: 8 MB (shared, CPU B1 & B2)

>>NOTES: CPU B2 is repurposed integration of PlayStation Stream CPU. PlayStation Stream CPU is more advanced (i.e features) processor vs PlayStation Fold, still 8C/8T design, lower-clocked in PlayStation Stream. Design necessary as PS Stream provides native play of PS4 titles requiring 8C/8T setup with similar L2$ and L3$ implementation.

B2 cores based on lower-scale Zen-ARM variants; have certain reduced functionality (reduced FP and AVX 256) vs. B1 cores that are based on regular design. Cores B1 & B2 operate in HMP (Heterogeneous Multi-Processing) Mode.​

237170-radeon-rdna-chip-1260x709.png

[GPU]

>ARCHITECTURE: RDNA

>GENERATION: 7

>PROCESS: N3P

>CONFIG: Chiplet

>SHADER ENGINES: 2

>SHADER ARRAYS (PER SE): 2

>COMPUTE UNITS: 40 (20 per Shader Engine, 10 per Shader Array)

>COMPUTE UNIT CONFIGURATION: Dual CU

>SHADER CORES (PER CU): 128

>SHADER CORES (TOTAL): 5120

>ROPs: 128

>TMUs: 8 (per CU), 320 (total)

>ALUs/SHADER UNITS: 5120

>STATE MODES (IF ANY): 2; FULL (40 CUs), HALF (20 CUs)​

>WHAT IS STATE MODE: A way to dynamically shift power across sections of the GPU so that a given power draw profile can​
be maintained while increasing specification of performance for active parts of GPU. "Inactive" portion essentially goes into​
Standby Mode, data in cache is kept resident as required. Since Inactive portion is still technically active (just at much lower​
clocks, 25 MHz), specific low-impact tasks WRT data management can still be done on Inactive portion if desired.​

>CLOCK FREQUENCIES: 3901.396 MHz (FULL), 4659.96 MHz (HALF)

>IPC: 2

>IPS: 6724.427 million IPS (FULL), 9319.9236 million IPS (HALF)

>THEORETICAL FLOATING POINT OPERATIONS PER SECOND: 39.95 TF (FULL), 23.859 TF (HALF)

># PRIM UNITS: 4 (2 each SE, 1 each SA)

># PRIMs/CLOCK PER UNIT: 4

># TRI RAS/CLOCK PER UNIT: 2

>PRIMITIVES PER CLOCK: 16

>TRIANGLES PER CLOCK: 8

>GEOMETRY CULLING RATE: 62.422 billion (FULL), 37.279 billion (HALF)

>TRIANGLE RASTERIZATION RATE: 31.211 billion (FULL), 18.639 billion (HALF)

>PIXEL FILL RATE: 499.378 Gpixels/s (FULL), 298.237 Gpixels/s (HALF)

>TEXTURE FILL RATE: 1248.446 Gtexels/s (FULL), 745.5938 Gtexels/s (HALF)

>THREAD DEPLOYMENT RATE: 40,960 (FULL), 20,480 (HALF)

>CACHE CAPACITY:

>L0$: 16 KB per CU (640 KB total)

>L1$: 64 KB per dual-CU/WGP (1.28 MB total)

>L2$: 512 KB per Array (2 MB total)

>L3$: 16 MB

>TOTAL: 19.92 MB

>CACHE BANDWIDTH:

>L0$: 39.55 TB/s (FULL), 23.62 TB/s (HALF)

>L1$: 27.685 TB/s (FULL), 16.534 TB/s (HALF) (L0$ * .70)

>L2$: 16.611 TB/s (FULL), 9.92 TB/s (HALF) (L1$ * .60)

>L3$: 8.3 TB/s (FULL), 4.96 TB/s (HALF) (L1$ * .30)

>TOTAL: 88.829 TB/s (FULL), 55.034 TB/s (HALF)​

[TASK ACCELERATION ENGINE]
**Heavily modified CU for various hardware-based RT, ML, IU, AI tasks

**Each Shader Engine is paired with a Task Acceleration Engine

**TAEs can house either 1 or 2 Task Acceleration Arrays

**TAAs can be of any given size when it comes to TAUs; however ideal
balanced designs generally call for TAAs that are 50% the core unit
count of a design's Shader Array (i.e a design with 5 Dual CU Shader
Arrays (10 CUs) having 5 TAU TAAs).

**TAUs can be thought of as pseudo-FPGA cores; each unit has internal
fixed-function hardware as well as a modified frontend compared
to normal CUs. TAUs also can range from having the following:

>2,000 To 8,000 logic cells

>16 Kb (2 KB) to 128 Kb (16 KB) distributed RAM (LUT) (acts as L0$)

>64 Kb (8 KB) to 512 Kb (64 KB) BRAM (acts as L1$)

>Internal Command Sync & Program Interface Configuration core
(directed by application to program the FPGA logic into
desired program state, manage unit state)

>Compute Thread Controller core (this is what programmers
interface with in dispatching code for the configured units
in the TAA to execute)

>1 Mb (128 KB) to 4 Mb (512 KB) SRAM (L2$, shared with all units in TAA;
meant for processed data locality storage)

>Shared 8 MB XIP (Execute-in-Place) Resource State Preset-allocated
MRAM (holds the microcode in bit-addressable/byte-addressable
format for immediate configuration)*

* = Will be able to be updated with future firmware changes

* = Shared between both TAAs

* = Simply used to configure the FPGA logic & fixed-function
aggregate unit states to specific profile presets (RT, ML, AI,
or IU) by storing the relevant set-up and configure/compile data

># TAEs: 2 (1 per SA)

># TASK ACCELERATION UNITS (TAU) PER TAE: 5

>LOGIC CELLS: 5000 (per TAU); 25,000 (per TAE), 50,000 (total)

>LUT RAM (L0$): 12 KB (per TAU), 48 KB (per TAE), 96 KB (total)

>BRAM (L1$): 48 KB (per TAU), 240 KB (per TAE), 480 KB (total)

>SRAM (L2$): 384 KB (per TAE), 768 KB (total)

>MRAM: 8 MB (total)

># CU: 1 (PER DUAL CU IN SA), 5 (PER SA), 10 (TOTAL)​

[AUDIO]

>ARCHITECTURE: Tempest Audio (Tempest Audio Engine Next (TAEN))

>GENERATION: 2

>DSPs: 2

>PROGRAMMABLE LOGIC: 1

>Modified single CU core

>3.9 GHz (fixed clock)

>8 KB L0$, 32 KB L1$, 128 KB L2$DFD

>No L3$

>~ 1 TF (998.75 GFLOPs)

>Can leverage up to 40 GB/s of system bandwidth​

hbm_678_678x452.png

[MEMORY]

>TYPE: HBM4E

>GENERATION: 1

>MODULE CAPACITY: 2 GB

>MODULE AMOUNT: 16x

>CONFIGURATION: 1X 16-HI

>I/O PIN BANDWIDTH: 6.2 Gbps (775 MB/s)

>I/O PIN COUNT: 128

>BUS WIDTH: 2048-bit

>MODULE BANDWIDTH: 99.2 GB/s

>TOTAL BANDWIDTH: 1587.2 GB/s (1.5872 TB/s)

>TOTAL CAPACITY: 32 GB​

What_is_an_SSD-Thumb.png

[I/O COMPRESSION/DECOMPRESSION/MEMORY CONTROLLER]

[SSD]

>FORM FACTOR: M.2

>INTERFACE: PCIe-CXL

>INTERFACE GENERATION: 6.0

>CONNECTION SLOT(S): 1x (x4)

>NAND TYPE: ZNAND

>NAND MODULE CAPACITY: 2 Tb (256 GB)

>NAND MODULE BANDWIDTH: 2 GB/s

>NAND MODULE LATENCY: 1.8 ms

>NAND MODULE CHANNELS: 4 (per module), 32 (total)

># NAND MODULES: 8x​
>DEVICE BANDWIDTH: 16 GB/s

>MAXIMUM SUPPORTED BANDWIDTH: 32 GB/s

>DRIVE CAPACITY: 2 TB (2048 GB)

>MAXIMUM SUPPORTED DRIVE CAPACITY: 16 TB (16384 GB)​
[FLASH MEMORY CONTROLLER]

>STORAGE CACHE: IFC-SRAM, 384 MB

>STATUS: Internal cache control, GPU L4$ (shared cache)

**Specific implementation of Infinity Cache level for PlayStation 6

**Can replicate an approach similar to NVRAM without need reworking
of underlying memory sub-systems outside of virtual dynamic partition
scaling of the cache pool.

**Decompressor can stream through 96 GB of storage data into 384 MB cache
pool (would require 250 data block-buffer write cycles per second to the cache);
GPU can then access as an L4$ type Infinity Cache via an IFC cache
controller fabric (shared with the decompression hardware).

**Decompression block & flash memory controller logic sorts management of
cache coherency routines.​

>INTERFACE SUPPORT: x4 (full) M.2 PCIe-CXL 6.0 NVMe, x2 (half) M.2 PCIe-CXL 6.0 NVMe (either/or)

>BANDWIDTH CAPACITY (CACHE): 4150 GB/s (4.15 TB/s)

>LATENCY THRESHOLD: Sub-100 ns​

[COMPRESSOR/DECOMPRESSOR/MANAGEMENT]

>DECOMPRESSION BANDWIDTH LIMIT (HARDWARE): 96 GB/s

>DECOMPRESSION RATIO: 6:1

>COMPRESSION BANDWIDTH LIMIT (HARDWARE): 96 GB/s

>COMPRESSION RATIO: 6:1

># ENGINES: 4

1: IFC-SRAM Data Management Engine (DME)

2: RAM Data Management Engine (DME)

3: Network Data Management Engine (DME)

**Management for data over wifi and ethernet ports

4: Overhead/General Management Core Engine

**Orchestrates comms of the other three DMEs​

ws5F2PiSBx2qQ8KTQJToiG-1200-80.jpg

[PERIPHERAL SUPPORT]

>USB GENERATION SUPPORT: GEN 1, GEN 2, GEN 3, GEN 4

>USB-C SUPPORT: YES

># USB PORTS: 4

>FRONT: Custom USB Gen 4 Type-C 3x2 Alt-Mode data port (VR), 1x regular USB Gen 4 Type-C 2x2 port
(controllers, charging kits, etc.)

>BACK: 1x USB Gen 4 Type-A 2x2 port, 1x USB Gen 3 port​

>MICRO SD SUPPORT: YES

*Used for physical media delivery (128 GB/192 GB Game Cards). Managed by FMC & Decompressor.

*Game Card Bandwidth: 60 MB/s​

># MICRO SD SLOTS: 1 (FRONT, UHS-III, SDHC class)

>THUNDERBOLT SUPPORT: NO​

ServeImage

[WIFI/ETHERNET/BLUETOOTH]

[ETHERNET]

>IMPLEMENTATION: Single Port

>ETHERNET PHYSICAL LAYER: 25 Gbit

>NAME: 25GBASE

>STANDARD: 802.3bq-2016

>CONNECTOR: RJ45

>LINE CODE: 64b/66b

>TRANSCEIVER MODULE
: SFP28

>RECOMMENDED CABLE: 25GBASE-CR, 25GBASE-CR-S

>BANDWIDTH: 3.125 GB/s​
[WIFI]

>IMPLEMENTATION: Single Module

>NAME: Wifi 7

>STANDARD: 801.11acx

>ENCODE RATE: 8b/10b encoding

>MU-MIMO STREAMS: 5

>BANDWIDTH: 2925.427 Mbps per MIMO, 14.627 Gbps theoretical, 1.828 GB/s converted​

PSVR-sony-e1555441323440.jpg

[VR SUPPORT]

>1x Entry-Level PSVR Gen 3 headset included with PS6

>Headsets also compatible with PC as VR peripheral units

>Includes wired (USB Gen 4 3x2 split as 4x USB Gen 3-speed Type-C ports), proprietary
connection w/ safety mechanisms to secure connection line

>Included VR helmet supports "2K" (1080p) 120 Hz

>10th gen PSVR headsets come in 3 model variants:

>ENTRY: Supports up to 2K (1080p) 120 Hz, per eye. $129.99 MSRP (1x with every PS6)

>Bandwidth rate: 8 BPC/RGB/No Compression/No Timing (6.37 Gbit/s (796 MB/s) * 2 = 12.74 Gbit/s (1.5925 GB/s))
lower end, 8 BPC/RGB/No Compression/CVT Timing (7.27 Gbit/s (908.75 MB/s) * 2 = 14.54 Gbit/s (1.8175 GB/s))
higher end.

>W/ USB 4 Gen 3x2 128/132b Encoding: 13.138 Gbit/s (1.642 GB/s) lower end, 14.994 Gbit/s
(1.874 GB/s) higher end

>W/ WiFi 7E: 13.66 Gbit/s (1.707 GB/s)(lower end), 15.59 Gbit/s (1.94 GB/s)
(higher end)(96% utilization on wireless module only); 16.67 Gbit/s (2.08 GB/s)
(lower end), 19.49 Gbit/s (2.436 GB/s)(higher end)(75% utilization on standard WiFi 7E
transmission devices).

>Built for uncompressed 2K 120 Hz per eye

>Built-in speakers

>Built-in entry level microphone

>Internal battery; can charge through the USB port when in wired mode. Can
also charge wireless via wireless transmission unit's charging station feature
(sold separately).

>Comes with 9' USB Type-C cable for wired play, charging, and data transmission

>160 degree FOV

>Built-in 2x 320 MHz bandwidth MU-MIMO antenna streams​
>ENTHUSIAST: Supports up to native QHD (1440p) 90 Hz per eye. $199.99 MSRP

>Bandwidth rate: 10 BPC/RGB/No Compression/No Timing (9.95 Gbit/s (1.24 GB/s) * 2 = 19.9 Gbit/s (2.48 GB/s))
lower end, 10 BPC/RGB/No Compression/CVT-RB Timing (11.04 Gbit/s (1.38 GB/s( * 2 = 22.08 Gbit/s (2.76 GB/s))
higher end.

>With USB 4 Gen 3x2 128/132b Encoding: 20.52 Gbit/s (2.56 GB/s) lower end, 22.77 Gbit/s
(2.846 GB/s) higher end

>W/ WiFi 7E: 20.696 Gbit/s (2.587 GB/s)(lower end), 22.96 Gbit/s (2.87 GB/s)
(higher end)(96% utilization on wireless module only); 24.875 Gbit/s (3.109 GB/s)
(lower end), 27.6 Gbit/s (3.45 GB/s)(higher end)(75% utilization on standard WiFi 7E
transmission devices).

>Built-in speakers

>Built-in upgraded microphone

>Internal battery; charge through USB port in wired mode. Can
also charge wireless if used with wireless transmission unit via
proximity with unit charge station functionality (sold separately).

>3.5 hours on full charge for wireless play.

>Comes with 12' USB Type-C cable for wired play, charging, and data transmission

>160 degree FOV

>Built-in 2x 320 MHz bandwidth MU-MIMO antenna streams​
>PERFORMANCE: Supports up to native 4K 90 Hz per eye. $399.99 MSRP

>Bandwidth rate: 10 BPC/YCbCr 4:4:2/No Compression/No Timing (14.93 Gbit/s (1.866 GB/s) * 2 = 29.86 Gbit/s (3.73
GB/s)) lower end, 10 BPC/YCbCr 4:4:2/No Compression/CVT-RB(16.23 Gbit/s (2.028 GB/s( * 2 = 32.46 Gbit/s (4.05
GB/s)) higher end.

>W/ USB 4 Gen 3x2 128/132b Encoding: 30.79 Gbit/s (3.849 GB/s) lower end, 33.474 Gbit/s
(4.184 GB/s) higher end

>W/ WiFi 7E: 31.0544 Gbit/s (3.8818 GB/s)(lower end), 33.75 Gbit/s (4.21 GB/s)
(higher end)(96% utilization on wireless module only); 37.325 Gbit/s (4.66 GB/s)
(lower end), 40.575 Gbit/s (5.0718 GB/s)(higher end)(75% utilization on standard WiFi 7E
transmission devices)*

* = 4K 90 Hz native refresh rate not provided with standard
WiFi 7E transmission devices​

>Built-in speakers​
>Built-in upgraded microphone​
>Internal battery; charge through USB port in wired mode. Can​
also charge wireless if used with wireless transmission unit via​
proximity with unit charge station functionality (sold separately).​
>3.5 hours on full charge for wireless play.​
>Comes with 12' USB Type-C cable for wired play, charging, and data transmission​
>160 degree FOV​
>Built-in 2x 320 MHz bandwidth MU-MIMO antenna streams​
>Optional wireless module sold separately; plugs into system's USB 4 Gen 4 3x2 port.
Wifi 7-based, following specs:

>Five MU-MIMO antenna streams

>7 GHz band (7234 MHz)

>Compatible with 320 MHz channel width

>2925.427 Mbps per MIMO

>14.627 Gbps theoretical

>1.828 GB/s converted

>~ 1.75 GB/s effective (96% utilization)

>Can support Entry headsets (1x, ALL. 2x, DSC 3, ALL), Enthusiast (1x, 10 BPC/RGB/No Compression/No Timing,
lower end), Entry headset (3x, ALL @ 90 Hz, DSC 3), (1x Entry (10 BPC/RGB/DCS 3/CVT-RB) + 1x Enthusiast
(10 BPC/RGB/DCS 3/CVT-RB))

>Eight-core/Eight-thread high-end ZEN-ARM processor block (repurposed PS6 CPU B2)

>Modified/repurposed GPU CU core

>Built-in compression hardware with DSC 3.0x support

>512 MB LPDDR5 RAM

>1 GB NAND

>MSRP $129.99​

>Second version, performance-level wireless VR module, has specs of:

>Ten MU-MIMO antenna streams

>Simultaneous dual-band (7 GHz (7234 MHz) + 7 GHz (7234 MHz))

>Compatible with 320 MHz channel width

>2925.4275064 Mbps per MIMO

>29.254275064 Gbps theoretical

>3.656784383 GB/s converted

>~ 3.51 GB/s effective (96% utilization)

>Supports Entry headsets (1x, ALL. 4x, DSC 3, ALL), Enthusiast (3x, 10 BPC/RGB/DCS 3/CVT-RB,
higher end), Professional headset (2x @ 90 Hz/DSC 3/CVT-RB, higher end. 3x, 8 BPC/YCbCr/DSC 3/
CVT-RB, mid-low end; 4x, 8 BPC/YCbCr/DSC 3/No Timing, lower end)

>Eight-core/eight-thread ZEN-ARM processor block (repurposed CPU B2 of PS6)

>Modified/repurposed GPU CU core

>1 GB LPDDR5 RAM

>2 GB NAND

>MSRP $249.99​

85

[PRICING]

>PS6: $499.99

>PS6 WIRELESS VR SKU: $599 (includes PS6, Entry VR headset, and Wifi Client Basic)​

>PSVR GEN 3 PERFORMANCE HEADSET: $349

>PSVR GEN 3 ENTHUSIAST HEADSET: $199

>PSVR GEN 3 ENTRY HEADSET: $129

>PS.STREAM: $199

>PS.STREAM & PSVR GEN 3 ENTRY HEADSET SKU: $299​

>DUALSENSE 2 CONTROLLER: $69

>PSVR WIFI CLIENT BASIC: $129

>PSVR WIFI CLIENT ENHANCED: $249​

[RELEASE]

>November 2027 or March 2028
 
(con't)

So, that's what I think a 10th-gen PlayStation console/platform could look like from the hardware side of things. There's obviously the chance this doesn't play out, but I tried my best to look at various data trends and extrapolate any patterns from them, and keeping in mind the general business strategies these companies seem they want to pursue, their markets of interest, and tried timing things out as best as able.

If you all have any ideas for what YOU think Sony'll do for 10th-gen system designs (or if you already shared them in one of the other threads like this I did), please post/re-post them so we can discuss. If you think it's too early to talk 10th-gen, keep in mind these companies are ALREADY in planning stages for future system designs, so if they feel it's a good time to start brainstorming, so can we :goog_wink:

PLEASE keep this one focused solely on Sony and post any Microsoft-related stuff in the Microsoft thread. Also again, if you want me to explain ANYTHING with what I posted, just ask and I can do so; I've been working on this stuff on-and-off for the past several months in my spare time, I have enough notes to pull from ;)

Interested to see what thoughts and ideas you have 👍
 
Last edited:

TLZ

Banned
Yea definitely in Japan. I cannot see them dropping Japan. No way. PS5 is sold out there and they're doing lotteries. Even if it wasn't, I still don't see them dropping PS6 at all. Sony is still a Japanese company. Heck even MS selling 2 consoles per week is still doing it there.
 
Just a guess, but I think both companies will move to services rather than hardware.
By 2027/2028? No hardware options at all? I dunno; there's still room for growth in visual fidelity, and there's still a lot to do with VR & AR. Plus expecting internet infrastructure and ISP regulations to 100% tighten up by 2027/2028 is strongly optimistic.

But we'll see what happens; for sure they will rely very strongly on services no matter what tho.

Yea definitely in Japan. I cannot see them dropping Japan. No way. PS5 is sold out there and they're doing lotteries. Even if it wasn't, I still don't see them dropping PS6 at all. Sony is still a Japanese company. Heck even MS selling 2 consoles per week is still doing it there.
That's true but I'm still curious how long sales will remain strong (ish?) in Japan, given some of the growing sentiment of frustration with some Japanese gamers, and Nintendo gearing up to storm it again with new games and the next Switch. I hope MS is able to either improve hardware sales there or see growth through GamePass & Xcloud on mobile, which seems like the more likely option for them in that kind of region.

I kind of think the bigger factor that determines what Sony does with Japan depends on software attach rates for PS5 relative to hardware sales, because it's the software (and services) where the companies make the money. If the ratio's bad, at the very least they might reconsider what type of hardware approach they go for Japan w/ PS6, especially if BOM and distribution costs for the console are too high relative to the profits through software sales in the region.

lol already?

;) Why not? I bet c'ha Mark Cerny and the rest of the team are already brainstorming stuff for PS6 as we wait to get lucky securing a PS5 xD
 
Last edited:

tillbot8

Banned
By 2027/2028? No hardware options at all? I dunno; there's still room for growth in visual fidelity, and there's still a lot to do with VR & AR. Plus expecting internet infrastructure and ISP regulations to 100% tighten up by 2027/2028 is strongly optimistic.

But we'll see what happens; for sure they will rely very strongly on services no matter what tho.


That's true but I'm still curious how long sales will remain strong (ish?) in Japan, given some of the growing sentiment of frustration with some Japanese gamers, and Nintendo gearing up to storm it again with new games and the next Switch. I hope MS is able to either improve hardware sales there or see growth through GamePass & Xcloud on mobile, which seems like the more likely option for them in that kind of region.

I kind of think the bigger factor that determines what Sony does with Japan depends on software attach rates for PS5 relative to hardware sales, because it's the software (and services) where the companies make the money. If the ratio's bad, at the very least they might reconsider what type of hardware approach they go for Japan w/ PS6, especially if BOM and distribution costs for the console are too high relative to the profits through software sales in the region.



;) Why not? I bet c'ha Mark Cerny and the rest of the team are already brainstorming stuff for PS6 as we wait to get lucky securing a PS5 xD
Yeah, you are right. Really appreciate the effort of the thread btw, very good read :)
 
Yeah, you are right. Really appreciate the effort of the thread btw, very good read :)
I've been told multiple times I am crazy for writing these but some folks don't understand I just write most of this over a period of weeks.

Glad you've been having fun with what's here; we have to wait a long time to see what exactly these companies do for their next gen of hardware, but hopefully at least some of these ideas become real, in one way or another.
 
It’s hard to even consider what the next generation of consoles will be like:
  • VR is starting to mature with Quest 2 and Half Life Alyx in 2020 - it’s hard to imagine consoles not designed around VR in 2028.
  • If tradition and history holds, these SoCs will be manufactured on an approximately 1nm process. Crazy to think about. 5-8X the performance, the ray tracing, DLSS 5.0 or whatever. Wild stuff ahead If the pace continues.
  • Streaming sucks for hardcore fans but it‘s going to completely upend the long tail of the console market (at least for most genres).
My guess is that there will be console hardware but that the experience will be more different than the 360 launch is (blades and all) compared to the current console experience.
 
It’s hard to even consider what the next generation of consoles will be like:
  • VR is starting to mature with Quest 2 and Half Life Alyx in 2020 - it’s hard to imagine consoles not designed around VR in 2028.
  • If tradition and history holds, these SoCs will be manufactured on an approximately 1nm process. Crazy to think about. 5-8X the performance, the ray tracing, DLSS 5.0 or whatever. Wild stuff ahead If the pace continues.
  • Streaming sucks for hardcore fans but it‘s going to completely upend the long tail of the console market (at least for most genres).
My guess is that there will be console hardware but that the experience will be more different than the 360 launch is (blades and all) compared to the current console experience.

Personally I am hoping VR becomes a default standard, because that is what's really going to drive adoption rates. Remember when analog sticks and rumble where optional peripherals? Not a lot of games were designed around them. Then when 6th-gen consoles made them more of a standard those sort of things became incorporated more tightly into game design across the industry.

VR headsets of good quality at the cheap should be readily capable of production within the next few years. I'm talking entry-level headsets a little the price of a controller; that way if the BOM is also low enough, include a headset standard with all units. It could also encourage innovations with the UI features/experience, even encourage integration of VR with regular home routines possibly.

This gen is so boring we're already discussing next gen :messenger_tears_of_joy:

Only because most people still can't secure a console. I'm only planning on picking them up near the end of the year, when supply should be regular and easily found.

Mate, you have too much free time, but you use it in a good way. :lollipop_grinning_sweat:
Just biding some time while these consoles continue to act like phantoms for the masses waiting ;)
 

Ev1L AuRoN

Member
If Sony launches a compelling hardware I'll buy it, if services are the future, I'll stick with PC gaming until I die. I'm not against services and cloud gaming, and I will use them for sure, but I prefer to run my games locally, and PC will always allow me to do so, when the day come and games become available only by the cloud, I'll be happy with my memories and emulators on my PC.
 

yurinka

Member
I think PS6 will release worldwide November 2027 with the latest AMD CPU and GPU tech focused on getting a huge ass improvement of real time global illumination and raytracing to target photorealism with games featuring native 4K 60fps but would feature some cool tech to scale it to 8K with no visible artifacts with something that improves DLSS and some stuff to interpolate frames and push them to 120fps.

It would also include the lastest wifi, LAN tech, maybe even connection 5G and a dedicated decompression chip for Remote Play, cloud gaming, video streaming and wireless VR.

It may use PSVR3 controllers as main console controllers. Controllers that will be also used to play in phones or tablets for remote play or cloud gaming. There will be PSVR3 but maybe not at launch. It will also feature AR and will work standalone.

PSN store will be available for PC and mobile featuring there both Sony and 3rd party games. Full BC for all previous gens including PSP and Vita.

There won't be a next gen Xbox, MS will go full 3rd party in consoles. Sony will publish their own games on PC and Nintendo platforms after timed exclusivity on PS, in a few cases like remakes, remasters or GaaS day one or almost, in other cases years later.

Game subscriptions, cloud gaming and VR will have increased a lot their market and importance but game and add-on sales will continue being the king. F2P and GaaS will be way more important than already are now and may be the last console with disc if it features it will be mostly for BC because at launch the physical sales will be almost zero. Digital games, digital games add-ons (DLC, season passes, IAP), mods and UGC will be NFTs meaning that there will be an official 2nd hand market of all these digital items. Play to earn games will pass F2P as the top grossing gaming business model during this generation but paid games will still exist even if being a way smaller part of the market than they are now. Play 2 earn and F2P won't negatively affect gameplay, after some awful implementations they will keep iterating and will reach a point where both players and devs will be happy.

Crossplay, cross buy and cross save will be shared across all platforms: PS, PC, mobile, cloud and maybe even rival consoles. You'll be able to play via streaming all the digital PSN games you buy, not only the ones rented in the subscription. But you'll need to pay that subscription to play them via streaming in any device.

They won't have their own portable because won't need it because PC handelds will dominate the AA/AAA portable market and will merge with modern smartphones and tablets, also capable or running natively AA/AAA games. Both PC and mobile will feature almost the same games than console (in case of mobile, not the high end ones or maybe yes but with lower settings).

Nintendo fans will claim that the Switch 2 succesor invented 4K, VR, proper online multiplayer and cloud gaming.
 
Last edited:

jhjfss

Member
There won't be a next gen Xbox, MS will go full 3rd party in consoles. Sony will publish their own games on PC and Nintendo platforms after timed exclusivity on PS, in a few cases like remakes, remasters or GaaS day one or almost, in other cases years later.
Not this nonsense again🙄
 
Last edited:

VFXVeteran

Banned
I really don't understand the OP's over the top specs for a $500 piece of hardware. There becomes a point where power will level off and will simply be too expensive to add features in an exponential way. This happens with every generation speculation thread. The hardware is always much much less than what people are hoping for. Can't we be "conservative" for once instead of being overly aggressive with the predictions?
 
Last edited:
There won't be a next gen Xbox, MS will go full 3rd party in consoles. Sony will publish their own games on PC and Nintendo platforms after timed exclusivity on PS, in a few cases like remakes, remasters or GaaS day one or almost, in other cases years later.

Not this nonsense again🙄

Damn I'm late :p

Anyway, I think you're both right. TBH I think 10th gen Sony and Microsoft will have wildly divergent systems, not just in terms of certain specifications but in key features they'll push as well. I think after four generations of trying to compete with Sony on power and only clearly having the advantage in one of them (I don't think much will change between PS5 & Series X performance wise for the rest of this generation tbh, unless Mesh Shaders take off and Sony screwed up using super-outdated Primitive Shaders with no alterations...which has effectively a 0% chance of having happened), Microsoft isn't going to really pursue that going forward.

Not that their 10th-gen system won't offer comparable performance to Sony's: it will. But it'll do it by different means and with a much different marketing hook. And conversely, I think Sony will try building the ultimate console that fully addresses arguably the biggest issue with computing performance: data locality, something they've already started exploring with PS5's SSD I/O but will IMO perfect with PS6's design.

I also can see Microsoft deepening a partnership with Valve for 10th-gen in some way, and Sony finally having a storefront/launcher on PC they can own and manage in such a way to make Day 1 PC ports of PS games doable without impacting console sales too much. That being said...

...tbh I'm just gonna pretend this thread don't really exist xD; it's been more than a year, I've basically reneged on everything in the OP :p .

I really don't understand the OP's over the top specs for a $500 piece of hardware. There becomes a point where power will level off and will simply be too expensive to add features in an exponential way. This happens with every generation speculation thread. The hardware is always much much less than what people are hoping for. Can't we be "conservative" for once instead of being overly aggressive with the predictions?

If you think some of that was over the top, you're gonna hate where I think 10th-gen systems will land at now 😁
 

Sega Orphan

Banned
Sony's next console (either PS5 pro or PS6)

Zen 3-4 CPU 8 core, 16 thread @ 4.5ghz Variable
AMD RDNA 4 GPU 72 CU @ 2.5ghz Variable
24gb GDDR6 RAM @ 936gbs (384 bit bus)
Mesh Shaders, VRS, Ray Tracing and lower precision ML abilities. (Int8 and Int4)
2tb SSD @ 5.5 - 10gbs


Rational
Current games do not fully exploit multiple core CPUs. Higher clocks and IPC gains of newer Zen tech will improve the CPU performance by a factor of 2.

Inline with Sony's BC strategy the GPU will be a butterfly design with 72CUs. Higher clock speeds and performance gains due to newer RDNA architecture gives the GPU about 2.7 x improvement over the PS5.
Sony will adopt VRS and Mesh Shaders to be inline with PC and Xbox GPUs.
There will also be advances to lower precision abilities of RDNA cards and inline with that the next PS will have them as well.
24gb of RAM @ 936gbs will feed the GPU correctly.
SSD will double in size, however I wouldn't expect too much of a speed increase as there doesn't seem to be a need for that at this time.
 
Actually I think you will hate the actual outcome of your predictions because none of yours was accurate for this generation. :messenger_persevering:

This generation? PS5/Series X/Series S? Or mid-gen refreshes? I've had a change of mind on mid-gen refreshes TBH, I don't think Pro models are coming due to logistics reasons. But they can still refresh the current consoles keeping same core perf profile of the base systems and adding in a few new features on a newer RDNA spec, theoretically.

I guess we'll see what happens; really tho for immediate refreshes seems like Microsoft will probably focus on a smaller streaming-friendly box like an Apple TV; Sony would probably like to lower BOM for PS5 Digital to start producing it at real volumes.

Sony's next console (either PS5 pro or PS6)

Zen 3-4 CPU 8 core, 16 thread @ 4.5ghz Variable
AMD RDNA 4 GPU 72 CU @ 2.5ghz Variable
24gb GDDR6 RAM @ 936gbs (384 bit bus)
Mesh Shaders, VRS, Ray Tracing and lower precision ML abilities. (Int8 and Int4)
2tb SSD @ 5.5 - 10gbs


Rational
Current games do not fully exploit multiple core CPUs. Higher clocks and IPC gains of newer Zen tech will improve the CPU performance by a factor of 2.

Inline with Sony's BC strategy the GPU will be a butterfly design with 72CUs. Higher clock speeds and performance gains due to newer RDNA architecture gives the GPU about 2.7 x improvement over the PS5.
Sony will adopt VRS and Mesh Shaders to be inline with PC and Xbox GPUs.
There will also be advances to lower precision abilities of RDNA cards and inline with that the next PS will have them as well.
24gb of RAM @ 936gbs will feed the GPU correctly.
SSD will double in size, however I wouldn't expect too much of a speed increase as there doesn't seem to be a need for that at this time.

Nice speculations. Those specs on the GPU would put it right around a 6900 XT, but I think that'd probably be something closer to a Pro than PS6, personally. GDDR6 being another reason, and the SSD bandwidth being a third.

Technically speaking the PS5's Primitive Shaders are an updated version of the ones in the Vega GPUs IIRC, and should be functionally equivalent to Mesh Shaders, a few differences notwithstanding. At least that's what I would assume, I just can't picture them having not seen such an obvious advancement and not including it in PS5 due to oversight.

While I'm personally more bearish on mid-gen Pro-level refreshes now, if the gen happens to go on a bit longer, say to 2028 or even until 2030, then refreshes in 2024-2025 of that level could still be on the table. But I think the chip and wafer situation would need to improve to a point where it makes logistical sense for Sony & Microsoft to pursue, otherwise they could split production between more models in a way that's detrimental to their base unit numbers.
 
I’m looking forward to Sony’s next console, whether that will be PS5 Pro or PS6.

I hope there will be a PS5 Pro first as I have decided to get the 2nd iteration of each console generation from now on. It makes the most sense to me since games are in a cross gen transition phase until the latter half of a console’s lifecycle anyway. I just hope all future PlayStations will be backward compatible.
 
I’m looking forward to Sony’s next console, whether that will be PS5 Pro or PS6.

I hope there will be a PS5 Pro first as I have decided to get the 2nd iteration of each console generation from now on. It makes the most sense to me since games are in a cross gen transition phase until the latter half of a console’s lifecycle anyway. I just hope all future PlayStations will be backward compatible.

No reason for them not to be, unless they switch to something other than x86-64 again. I'm interested in how far RISC-V advances in the next five years; it can end up being a great ARM alternative and a cheaper option than ARM in the long run for chip designers, as well as greater flexibility in how designers can customize the microcode and features of the chip.

That said ARM still has the performance edge and it isn't prohibitively expensive compared to RISC-V, plus chip makers have more experience with ARM comparatively. There is still some performance impact in emulating x86/64 code on an ARM design but it's gotten better and better; Apple might be the best in this regard. Very possible that Sony (and Microsoft) could go with ARM-based CPUs next gen.

Also been thinking about how/if Intel can play into things; if they hit their roadmaps and some of their future CPU/GPU designs are as good as they claim, one of the platform holders could theoretically jump over to using an Intel GPU, one of the reasons being driven by them owning their own fabs (though again, they need to hit their roadmaps first).
 
Last edited:

Sega Orphan

Banned
This generation? PS5/Series X/Series S? Or mid-gen refreshes? I've had a change of mind on mid-gen refreshes TBH, I don't think Pro models are coming due to logistics reasons. But they can still refresh the current consoles keeping same core perf profile of the base systems and adding in a few new features on a newer RDNA spec, theoretically.

I guess we'll see what happens; really tho for immediate refreshes seems like Microsoft will probably focus on a smaller streaming-friendly box like an Apple TV; Sony would probably like to lower BOM for PS5 Digital to start producing it at real volumes.



Nice speculations. Those specs on the GPU would put it right around a 6900 XT, but I think that'd probably be something closer to a Pro than PS6, personally. GDDR6 being another reason, and the SSD bandwidth being a third.

Technically speaking the PS5's Primitive Shaders are an updated version of the ones in the Vega GPUs IIRC, and should be functionally equivalent to Mesh Shaders, a few differences notwithstanding. At least that's what I would assume, I just can't picture them having not seen such an obvious advancement and not including it in PS5 due to oversight.

While I'm personally more bearish on mid-gen Pro-level refreshes now, if the gen happens to go on a bit longer, say to 2028 or even until 2030, then refreshes in 2024-2025 of that level could still be on the table. But I think the chip and wafer situation would need to improve to a point where it makes logistical sense for Sony & Microsoft to pursue, otherwise they could split production between more models in a way that's detrimental to their base unit numbers.
The big issue for consoles in the future is heat. They have to have that 220w limit, and node shrinks isn't going to keep up. The gap between the top PC cards and the consoles is growing. The talk is that the next RDNA 3 cards will hit 70tflops. Nvidia are talking 800-900w cards.
By the time the PS6 comes out PC cards will be 150tflops or greater, and the PS6 might be 30.
 
This is all so tiresome... when will it end?
I'm still seeing threads right now with people talking about the "True RDNA2" advantages that the SeX have. It was reignited by FRS 2.0 and news that the SeX will support it. They have "doubts" that Sony will bring it to the PS5 or that the console even supports it... it's so ridiculous.
They're still repeating that nonsense about "ML", always forgetting that the only game thus far that took advantage of "ML" capabilities in these current consoles was a Sony exclusive.

It's just a waste of time trying to argue with them about this.
 
The big issue for consoles in the future is heat. They have to have that 220w limit, and node shrinks isn't going to keep up. The gap between the top PC cards and the consoles is growing. The talk is that the next RDNA 3 cards will hit 70tflops. Nvidia are talking 800-900w cards.
By the time the PS6 comes out PC cards will be 150tflops or greater, and the PS6 might be 30.

Personally I think PS6 will be closer to 40-50 TF when all is said and done, same with next Xbox, but you are right that TDP power limits are going to remain strict for consoles (if not get stricter) and node shrinks alone won't be enough. Architectural changes will help, but that also alone won't help.

Which is why I think 10th-gen will be the right time for consoles to finally ditch GDDR memory. It's a power hog compared to HBM, and requires more PCB real estate (you can't stack the dies). Plus, in the future it's questionable how well it will do performance-wise versus later HBM3 and HBM4 memories. But the biggest reason I think it'll finally be time to ditch it, and this is something I've had to do a lot of research on in spare time, is because one of the actual biggest ways consoles can squeeze more performance on tight power budgets will be with shifting towards Processing-In-Memory (PIM) technologies.

Samsung actually already have HBM2E and DDR PIM modules for AI customers; they pack PIM logic cores right in the memory dies to do in-memory calculations, that way systems don't need to spend nearly as much time moving data over the memory bus (which is the biggest contributor to power consumption in computational workloads). That can free up a lot of power to go towards other parts of the system, but more importantly, Samsung's PIM modules are already able to shift around 1 TF of compute in each module.

The only 'drawback' is memory capacity; HBM2E PIM modules have half the capacity of regular HBM2E modules, so stacks use a mix of the PIM modules (which can operate in either PIM mode or regular mode) and normal HBM2E for maximum 6 GB 8-Hi stacks. But HBM3 will be able to offer up to 16-Hi stacks and increase module capacities up to 4 GB (the current max is 2 GB); I don't know if there are any such plans in the works for GDDR7.

So basically, HBM just offers all-around better performance and more design flexibility, helps address data locality much better (nothing's preventing future HBM-PIM designs from implementing other types of logic into the memory modules themselves, for things like metadata), brings increased performance (systems with HBM2E-PIM memories have shown up to 2x performance uplift compared to those using non-PIM memories), much lower power consumption, better opportunities for 3D stacking in system designs....basically everything xD.

The only real issues are pricing (this seems to affect HBM in general, really) and number of manufacturers (HBM is made by only two companies, Samsung and SK Hynix. GDDR is made by both of them and also made by Micron. That said, pricing ultimately would not be too big a concern for Sony and Microsoft if they know they'll have to make smaller APUs anyway, and the HBM-PIM memories could shift a decent chunk of that power performance into the memory itself. They would just need to design intelligent architectures around it, which would come down to robust logic layers for the module stacks, and maybe active interposers.

The other thing that could help consoles in the next generation, IMO, is adopting OMI XSR. It's based on OpenCAPI, which IBM uses for its POWER chips, but OMI XSR is open-standard and available to all. It has extremely good low latency (practically on par with DDR) but what it really does that'd be so good for consoles is decouple the memory controllers into chiplet and use OMI XRS transceiver links to network them with the main processor die. You can even build small subsystems for the memory controller chiplets including near-memory processing buffers (basically a form of Processing-Near-Memory (PNM), which is similar to PIM but less integrated into the memory modules themselves) and whatnot. Each OMI XRS transceiver link is 64 GB/s in bandwidth, low-latency (much lower than PCIe, even lower than CXL 2.0 and 3.0, CCIX, Gen Z by very notable margins), and uses CRC. In a way it's kind of like Micron's HBC concept but refined/perfected, and a lot more flexible & scalable.

Oh, and like PIM memories, also helps in lowering power consumption a great deal, though that also depends on what device type is using it. I don't know if AMD have any investments in OMI XSR, because that could be a potential bottleneck, or a company like Sony would need to convince them some way to utilize it at least for a custom 10th-gen APU design (they could theoretically "just" use something like CXL 2.0 (preferably 3.0, because that allows for a mirrored/reversed memory buffer) as well, but that wouldn't allow for decoupling of memory controllers into chiplets, and would bring higher latency, power consumption, and lower peak bandwidths). OMI-XSR with layered CXL 3.0 would be great for future storage devices, for example.

That's my ideal vision for 10th-gen systems architecturally speaking, anyway. And I feel that Sony would potentially be the more likely to leverage more (or hopefully, all) of those ideas than Microsoft, because in some ways the SSD I/O they have implemented in PS5 is already a form of PNM (Processing-Near-Memory); Microsoft's I/O is something to this effect as well but not to the same degree, by the nature of what they want DirectStorage to be, so the CPU & GPU are still going to play a major part so you're dealing with majority of processing taking place further from the source of the memory (the SSD in this case), but it brings its own benefits in larger scalability & flexibility since it's less hardware-dependent.

Data locality and moving processing as close to data as possible really are going to be the future of significantly increasing performance in the face of smaller gains from node shrinks (which get more expensive the smaller the node, could be a motivator for Microsoft and Sony to stick with smaller GPU designs. And I'm sure Sony don't want a repeat of supply issues they've had with PS5 so far; a smaller APU helps with that) and pure GPU/CPU architecture improvements possibly not being enough to compensate for that. New ways of intelligently optimizing the memory systems in designs, especially in closed, embedded designs (i.e consoles), is going to be the key. Ditching antiquated memories like GDDR is going to be the key. Leveraging PIM & PNM-friendly memories and systems designed around them is going to be the key.

So that's where I see 10th-gen (hopefully) going, especially if it ends up being the last traditional console generation. Might as well build the best architecture you can, but again for reasons probably more related to business than anything else, I expect Sony and Microsoft to diverge a bit more and some of these things I've mentioned here being more likely Sony could end up adopting them conceptually into their next system than Microsoft. I've actually been brainstorming a lot what their 10th-gen lineup could look like, I'd be into sharing that in a future post sometime.
 

MikeM

Member
This is all so tiresome... when will it end?
I'm still seeing threads right now with people talking about the "True RDNA2" advantages that the SeX have. It was reignited by FRS 2.0 and news that the SeX will support it. They have "doubts" that Sony will bring it to the PS5 or that the console even supports it... it's so ridiculous.
They're still repeating that nonsense about "ML", always forgetting that the only game thus far that took advantage of "ML" capabilities in these current consoles was a Sony exclusive.

It's just a waste of time trying to argue with them about this.
Its funny because FSR 2.0 doesn’t even need full rdna2 to work. It works best on RDNA 2, but its not needed.
 
This is all so tiresome... when will it end?
I'm still seeing threads right now with people talking about the "True RDNA2" advantages that the SeX have. It was reignited by FRS 2.0 and news that the SeX will support it. They have "doubts" that Sony will bring it to the PS5 or that the console even supports it... it's so ridiculous.
They're still repeating that nonsense about "ML", always forgetting that the only game thus far that took advantage of "ML" capabilities in these current consoles was a Sony exclusive.

It's just a waste of time trying to argue with them about this.

I wouldn't worry about it too much nowadays; technological features don't mean much at the end of the day if you don't have games demonstrating what they can do in real-time. Of course the Series systems have games which show this (the best examples being Flight Sim and Forza Horizon 5), but there are more than enough games on PS5 look visually convincing enough, that to the vast majority the system would seem to at least have those same features or capabilities, if not even better alternatives.

I mean, just seeing something like HFW in motion, it doesn't matter what features it is or isn't using. Doesn't change the fact it's arguably the best looking game on the market at this time, nothing else has its level of fidelity and detail density, with that level of smooth performance. Even on PS4 (the version I'm playing through, at least for the meantime), it looks better than quite a few current-gen games on the new systems!

Again at the end of the day, the games talk, so I wouldn't sweat about people still hung up on trying to claim "RDNA 1.5", "no ML" etc. arguments because the games coming out certainly don't seem to be hurting whatsoever even if those arguments were by some wild off-chance true (which they aren't).
 
PSVR2 will prove VR is a successful gimmick.
Ive never even tried VR, butI hope it really takes off with ps vr 2. PsVR2 I'm pretty excited for. Ever since I was a kid I've imagined VR would be the future of gaming and based on everything I've heard people say, the tech is already pretty good.
 
Last edited:
Personally I think PS6 will be closer to 40-50 TF when all is said and done, same with next Xbox, but you are right that TDP power limits are going to remain strict for consoles (if not get stricter) and node shrinks alone won't be enough. Architectural changes will help, but that also alone won't help.

Which is why I think 10th-gen will be the right time for consoles to finally ditch GDDR memory. It's a power hog compared to HBM, and requires more PCB real estate (you can't stack the dies). Plus, in the future it's questionable how well it will do performance-wise versus later HBM3 and HBM4 memories. But the biggest reason I think it'll finally be time to ditch it, and this is something I've had to do a lot of research on in spare time, is because one of the actual biggest ways consoles can squeeze more performance on tight power budgets will be with shifting towards Processing-In-Memory (PIM) technologies.

Samsung actually already have HBM2E and DDR PIM modules for AI customers; they pack PIM logic cores right in the memory dies to do in-memory calculations, that way systems don't need to spend nearly as much time moving data over the memory bus (which is the biggest contributor to power consumption in computational workloads). That can free up a lot of power to go towards other parts of the system, but more importantly, Samsung's PIM modules are already able to shift around 1 TF of compute in each module.

The only 'drawback' is memory capacity; HBM2E PIM modules have half the capacity of regular HBM2E modules, so stacks use a mix of the PIM modules (which can operate in either PIM mode or regular mode) and normal HBM2E for maximum 6 GB 8-Hi stacks. But HBM3 will be able to offer up to 16-Hi stacks and increase module capacities up to 4 GB (the current max is 2 GB); I don't know if there are any such plans in the works for GDDR7.

So basically, HBM just offers all-around better performance and more design flexibility, helps address data locality much better (nothing's preventing future HBM-PIM designs from implementing other types of logic into the memory modules themselves, for things like metadata), brings increased performance (systems with HBM2E-PIM memories have shown up to 2x performance uplift compared to those using non-PIM memories), much lower power consumption, better opportunities for 3D stacking in system designs....basically everything xD.

The only real issues are pricing (this seems to affect HBM in general, really) and number of manufacturers (HBM is made by only two companies, Samsung and SK Hynix. GDDR is made by both of them and also made by Micron. That said, pricing ultimately would not be too big a concern for Sony and Microsoft if they know they'll have to make smaller APUs anyway, and the HBM-PIM memories could shift a decent chunk of that power performance into the memory itself. They would just need to design intelligent architectures around it, which would come down to robust logic layers for the module stacks, and maybe active interposers.

The other thing that could help consoles in the next generation, IMO, is adopting OMI XSR. It's based on OpenCAPI, which IBM uses for its POWER chips, but OMI XSR is open-standard and available to all. It has extremely good low latency (practically on par with DDR) but what it really does that'd be so good for consoles is decouple the memory controllers into chiplet and use OMI XRS transceiver links to network them with the main processor die. You can even build small subsystems for the memory controller chiplets including near-memory processing buffers (basically a form of Processing-Near-Memory (PNM), which is similar to PIM but less integrated into the memory modules themselves) and whatnot. Each OMI XRS transceiver link is 64 GB/s in bandwidth, low-latency (much lower than PCIe, even lower than CXL 2.0 and 3.0, CCIX, Gen Z by very notable margins), and uses CRC. In a way it's kind of like Micron's HBC concept but refined/perfected, and a lot more flexible & scalable.

Oh, and like PIM memories, also helps in lowering power consumption a great deal, though that also depends on what device type is using it. I don't know if AMD have any investments in OMI XSR, because that could be a potential bottleneck, or a company like Sony would need to convince them some way to utilize it at least for a custom 10th-gen APU design (they could theoretically "just" use something like CXL 2.0 (preferably 3.0, because that allows for a mirrored/reversed memory buffer) as well, but that wouldn't allow for decoupling of memory controllers into chiplets, and would bring higher latency, power consumption, and lower peak bandwidths). OMI-XSR with layered CXL 3.0 would be great for future storage devices, for example.

That's my ideal vision for 10th-gen systems architecturally speaking, anyway. And I feel that Sony would potentially be the more likely to leverage more (or hopefully, all) of those ideas than Microsoft, because in some ways the SSD I/O they have implemented in PS5 is already a form of PNM (Processing-Near-Memory); Microsoft's I/O is something to this effect as well but not to the same degree, by the nature of what they want DirectStorage to be, so the CPU & GPU are still going to play a major part so you're dealing with majority of processing taking place further from the source of the memory (the SSD in this case), but it brings its own benefits in larger scalability & flexibility since it's less hardware-dependent.

Data locality and moving processing as close to data as possible really are going to be the future of significantly increasing performance in the face of smaller gains from node shrinks (which get more expensive the smaller the node, could be a motivator for Microsoft and Sony to stick with smaller GPU designs. And I'm sure Sony don't want a repeat of supply issues they've had with PS5 so far; a smaller APU helps with that) and pure GPU/CPU architecture improvements possibly not being enough to compensate for that. New ways of intelligently optimizing the memory systems in designs, especially in closed, embedded designs (i.e consoles), is going to be the key. Ditching antiquated memories like GDDR is going to be the key. Leveraging PIM & PNM-friendly memories and systems designed around them is going to be the key.

So that's where I see 10th-gen (hopefully) going, especially if it ends up being the last traditional console generation. Might as well build the best architecture you can, but again for reasons probably more related to business than anything else, I expect Sony and Microsoft to diverge a bit more and some of these things I've mentioned here being more likely Sony could end up adopting them conceptually into their next system than Microsoft. I've actually been brainstorming a lot what their 10th-gen lineup could look like, I'd be into sharing that in a future post sometime.
Bring it on. That was a good read.
 
Didn't noticed that this thread is to discuss the next generation... it's too early to that.


I was seeing the numbers for the next generation of GPUs and man, the numbers are "oppressive".
The number of TFlops is jumping so fast! Next generation 20TF will be "midrange", when the next generation of consoles comes we should be around 50TF at least for a midrange. And this along with all these new ways of doing things that try to be more efficient and doing more with less.
 
Didn't noticed that this thread is to discuss the next generation... it's too early to that.


I was seeing the numbers for the next generation of GPUs and man, the numbers are "oppressive".
The number of TFlops is jumping so fast! Next generation 20TF will be "midrange", when the next generation of consoles comes we should be around 50TF at least for a midrange. And this along with all these new ways of doing things that try to be more efficient and doing more with less.

That's more or less the idea; play smarter, not harder. Definitely look into stuff like PIM, PNM, HBM, OMI XSR, CXL 2.0/3.0 etc.; those will be very important for 10th-gen systems I think (or hope).

Bring it on. That was a good read.

Thanks; I might do another post sometime with links to resources discussing things mentioned here for those who'd like to look into them.
 
Top Bottom