Design a console with better specs than PS4 and Xbone, constraints apply

That's a good question given that price, form factor, and TPD are the real limiting factors.

I think you could achieve something like the PS4 but with maybe 24 more CU's and higher clock for around 2.5 Tflops of performance.

I think both of the new consoles are at the tipping point of price after which costs would increase exponentially (assuming they want to maintain the small and quiet form factors) and it's not economically worth it to push it.
 
- SoC consisting of two AMD A6-5200, so 8x 2.0GHz Jaguars plus 4 CUs @ 600MHz dedicated to compute tasks.

- 1GHz discrete Pitcairn with dedicated 2GB GDDR5 @ 4.8GHz for 20 CUs, 2.56TFlops, 154GB/s

- 12GB of DDR3-1866 in a 192bit config for 48GB/s for the APUs and GPU

- GPU connects to the APU through a 32-lane PCI-Express bus for a 32GB/s access to the main memory.

- Controller that can split and attach to any tablet/smartphone by the sides.

- Dedicated 2.4/5GHz, 2x2 WiFi module for WiFi Direct and screencasting to android, ios or windows tablets/smartphones.


BoM would probably sit below $400, but quite a bit over the current models.
I guess this is what a next-gen console would look like if it was made under the same principles as the 2005 models.
 
Last edited by a moderator:
Im not sure why almost everyone would put eSRAM in their console in such quantity (and die space). It makes less and less sense to have that kind of memory, while machines gets more and more powerfull, easying up developpement is (should) be the only concern now (aka having one and unique memory pool and bandwidth to deal with). Unless it's interfaced for a lot of bandwidth that really makes a difference (x10 times we can talk).

Microsoft didn't put eSRAM because they thought it was great or fun, they would have been way too far behind in bandwidth without it, that's all and it was a relatively cheap way to work around this issue while keeping a huge amount of cheap memory (same might apply with Nintendo design)

SRAM takes less power than logic. So the higher the transistor budget is I would suppose using SRAM to fill the space instead of logic makes sense to me for TDP reasons. But your design should also take advantage of SRAM's high bandwidth possibilities.

Also, there s no way you can launch a console early next year on 20nm process.. it's not a limited edition graphic card to launch.. it's million of consoles and on a proven process.

Apple's A7 will be produced in 20nm early 2014. If it's good enough for Apple's volume expectations it should be good enough for MS's volume.
 
Apple's A7 will be produced in 20nm early 2014. If it's good enough for Apple's volume expectations it should be good enough for MS's volume.

No basis for this. 20nm won't be ready in any realistic time-frame. Furthermore, Apple does product launches in Fall and Spring. Their 20nm volume wouldn't be needed for actual product launch until April/May. I can guarantee you a 28nm A7 will appear for product launches this fall before then.
 
What is the time constraint? How far back can our time machine go for all these custom things that take years to design?
Does the up-front R&D cost count against me? Can I not develop software and system infrastructure, and pass the savings onto the hardware nobody can actually program?

For all the folks trying to customize an existing CPU core design, take note of the release date for the standard core, and then add 12-18 months to it.
 
Apple's A7 will be produced in 20nm early 2014. If it's good enough for Apple's volume expectations it should be good enough for MS's volume.

I would not assume that.

Yields don't scale linearly with chip size. While Apple may get great yields and volume for a chip around 100mm2, but MS/Sony could have horrific yields for chips that are 2-4X of that in size.

On a 300mm wafer, you could make around 600 100mm2 chips, but only around 200 for something in the 300mm2 size, before adding in yield effects.
 
I would not assume that.

Yields don't scale linearly with chip size. While Apple may get great yields and volume for a chip around 100mm2, but MS/Sony could have horrific yields for chips that are 2-4X of that in size.

You can build you chip with defects in mind. Allow one core to be fused off if defective, - same with GPU CUs. SRAM is trivially made redundant to the point that you get virtually perfect yield for it.

MS has reserved a CPU core to be disabled for yield purposes, Sony did the same with the PS3's SPUs, and will almost certainly do the same again (can't wait for the RAGE). I'm guessing both Sony and MS are allowing two CUs to be defective, ie. XB1 is effectively a 7790 with 2CUs disabled and PS4 is a GCN2 Pitcairn with two CUs disabled.

You can still get unlucky and have a defect in an IO driver or some complex interconnect, not easily made resistant, but overall yield should be good.

If MS' SOC is 400mm² they get around 130 dies out of wafer. Assuming each wafer costs $5000 and yield after packaging is 60% (probably low) you get a SOC price around $65.

Cheers
 
You can build you chip with defects in mind. Allow one core to be fused off if defective, - same with GPU CUs. SRAM is trivially made redundant to the point that you get virtually perfect yield for it.

MS has reserved a CPU core to be disabled for yield purposes, Sony did the same with the PS3's SPUs, and will almost certainly do the same again (can't wait for the RAGE). I'm guessing both Sony and MS are allowing two CUs to be defective, ie. XB1 is effectively a 7790 with 2CUs disabled and PS4 is a GCN2 Pitcairn with two CUs disabled.

You can still get unlucky and have a defect in an IO driver or some complex interconnect, not easily made resistant, but overall yield should be good.

If MS' SOC is 400mm² they get around 130 dies out of wafer. Assuming each wafer costs $5000 and yield after packaging is 60% (probably low) you get a SOC price around $65.

Cheers

That's all true, I was merely commenting on the overall volume vs Apple. Apple would still be getting 4-5X the number of chips per wafer. And would probably have more pull at securing an overall wafer allotment on a new process.
 
For all the folks trying to customize an existing CPU core design, take note of the release date for the standard core, and then add 12-18 months to it.

I believe I'm the only one that customized the CPU core. And he said manufactured in 2013. I believe they could tape out a Jag with VXA2 by Dec 31st :cool:
 
Box size: 17"W x 3"H x 8"D standard av style
TDP: 200watts
Die budget: 400mm2
BOM: $400

8core jaguar 2GHz with 4cu igp on-board ~160mm2
pitcairn 220mm2 1ghz
8gb GDDR5
 
PS4 with the standard hdd removed - ~60$ savings.
this leaves some cash for an embedded framebuffer/L3 cache with a capacity of 16-32mb, this doesnt need to be ridiculous high speed but is aimed at keeping the bus free from the most demanding operations.

and regarding the lack of storage media - add some 16GB cheap flash for basic functionality and leave the real gaming systems as higher priced sku or just the choice of the buyer.
first this opens up buyers to pick the hdd of their choice without getting a dead weight HDD and then the mass market media will just compare the cheapest price with competing consoles - ignoring the features they provide out-of-the-box. you end up with a console that costs more but delivers more, and the casual buyer will only realize this if he is already invested in the hardware.
 
I think the following console would last Sony a long time.

CPU -- Duel Enhanced CELL at 20 or 28nm. More cache and faster bus between spes. All eight spes working.

GPU -- 32 cu chip or gtx 680
 
I am perfectly happy with the current specs. So my changes are minimal. This is my take.

Constraints:
Power consumption: < 120W
Price: 399$ or less

CPU:
8-16 1GHz Cores -following sebbbi's suggestion-

GPU: 18 CU
32MB eSRAM

Memory:
8GB GDDR5

SHAPE audio chip

Storage:
500GB HDD

1 wireless controller

Kinect
 
ARM A15x32 cores(4x4x2 AMBA-4 interconnect). 4GB DDR3(2.5ghz per core)

ARM Mali-T678(8 cores). 2GB DDR3

14nm FinFet(something arm is already doing..for real)

Not more powerful but more higher performance than the Xbox 360 and ALOT cheaper than PS4 or Xbox one.

Sell the console for $99.(costs only $60-$70)
 
GPU: APU 14 CU 6 Jags 64MB of edram
CPU: APU 2 CU 4 jags
256 bit interface 6 GBs of gDDR5

GPU jags can be totally devoted to gpgpu functionality.

CPU can operate as standalone for low power performance-xbl/psn arcade titles/streaming/security.

Allowances for concurrent processing.
 
Last edited by a moderator:
Samsung Console:

APU1:
8 core jag
24 CU gpu
64mb dram on interposer
8GB GDDR5

APU2:
quadcore A15 + quadcore powervr gpu
2-4GB LPDDR3

Korea #1
 
If you're not going to run into bandwidth limitations I think investing in a huge pool of on die ESRAM is probably a big waste. I would think you could probably get most of the benefit at a fraction of the cost and without adding a software managed memory pool by simply upgrading the GPU's L2 cache from 512-768KB to something like 2-4MB.
 
North Korea is best Korea!
specs :
some creeky microcontroller with no graphics hardware to speak off - let it manually generate a quarter res PAL signal.
a Z80 clone for sound
128KB or 256KB of RAM (make that SRAM!)
two NES or Genesis style controllers (unlike the capitalist pig practice of only providing one controller)
Games distributed on what's used for BIOS chips on our motherboards
If you hack it to run unauthorized games then you're sent to Gulag (along with all your descendents) (not that there are any actual technical measures to begin with)
 
I have kinda wondered what would happen if Sony lost it and built something crazy like a 64 SPU machine, maybe with a larger LS, and possibly some improved coherency. I imagine it'd be a nightmare to develop for, but I bet it would certainly be interesting to see what results. I'm sure it wouldn't be an efficient design though.

Trying to be on topic, I'm not sure what I would do. Maybe take Sony's base design and try and squeeze in a few more Jaguar cores or more bandwidth to feed the APU?
 
Well in front of the lack of definition for "better" I think that such a system could have been a contender given an one year head start:
SoC
8 expresso cores with 4MB of shared L2 @1.2 GHz
8 CUs 8ROps GPU (GCN class)
128 bit bus to 2GB of GDDR5
Integrate audio processor.
All @ 28nm TSMC, Burn in the 60 Watts while gaming.

Cheap chinese made ARM based SoC with 1 GB of (cheap) stacked memory (running OS and services)

BRD player
16GB of Flash (with at least half of it reserved for caching, partial instal).

Ship with Wiimote+ and a new sensor bar with a proper microphone array.
Cost 150-200$ sell for 299$$ with a revamped 3ds XL which allows remote play and has a second analog stick. Sells in the grey at first but should be slightly profitable fast.
Great offer for kids and the family.

A "base" sells separately and includes an 500GB HDD for the system (sold at a profit).
 
Last edited by a moderator:
Back
Top