Predict: Next gen console tech (9th iteration and 10th iteration edition) [2014 - 2017]

Status
Not open for further replies.
phoenix_chipset said:
People buy new phones every year just because they're new. People care. Not that I disagree that 8k would be a huge waste. I wouldn't expect to see console games at 8k until the ps5 pro if that's a thing.

Wasn't the ps4 CPU already half the size of its GPU? Wii u's CPU must've been like a fourth the size of its GPU lol.
Click to expand...
Its even more if one includes the ROPS, GPU 512KB of L2 cache, gpu interconnects, command processor, and memory controller which rightfully belongs 3/4th to the GPU (if not more as the 8core Jag's bus can only access a max a theoretical peak of 20GB/s of memory bandwidth with the main memory, compared to 176GB/s theoretical peak for the GPU).
 
Last edited:
manux said:
I don't believe raw flops will keep scaling like they used to. Xbox one x is pushing console power envelope at mature 14nm, console pricing and "only" reaching 6TFlop. 7nm is not magically allowing a huge leap over that. Realistic expectation might be something like 10TFlop fp32, 20TFlop fp16. Whatever improvements happen over that would have to be heavily architecture related to allow for more efficiency, accelerating new types of algorithms and utilizing lower precision like fp16/int8. Assuming 399/499 price point it's pretty limited brute force gains what next gen can have over xbox one x.

I'm wondering if next gen would benefit from inference and algorithms using "ai" to do processing? That could be a game changer or complete dud. Exposing tiling explicitly to developers might also be one way to drive efficiency up.
Click to expand...

Doesn't that depend on when the next consoles arrive, the target die size / max TDP, as well as the exact process node (7nm, 7nm+).

i.e. the earliest estimate for PS5's arrival is Holiday 2019. It could easily be Holiday 2020. The latest estimate for a new Xbox is another 4 years (2021), but no sooner than 2020.

I do agree that GPU architecture, efficiency, feature set and the algorithms it will accelerate will all be critical. I'm not sure about int8, but fp16 will probably be more useful in PS5.
 
Megadrive1988 said:
Doesn't that depend on when the next consoles arrive, the target die size / max TDP, as well as the exact process node (7nm, 7nm+).

i.e. the earliest estimate for PS5's arrival is Holiday 2019. It could easily be Holiday 2020. The latest estimate for a new Xbox is another 4 years (2021), but no sooner than 2020.

I do agree that GPU architecture, efficiency, feature set and the algorithms it will accelerate will all be critical. I'm not sure about int8, but fp16 will probably be more useful in PS5.
Click to expand...

2019 is still fairly early for 7nm console. Based on amd q3 report amd is not releasing their first 7nm gpu until 2019. Going from 6TFlop 499$ console to 499$ 20tflop console on 2019/2020 would be completely amazing. Imagine what the pc highend amd solution would look based on that scaling? I would be inclined to believe 20TFlop console in 7nm will not happen unless sony/ms releases 999$ 400W system akin to a highend gaming pc.
 
Last edited:
Megadrive1988 said:
Doesn't that depend on when the next consoles arrive, the target die size / max TDP, as well as the exact process node (7nm, 7nm+).

i.e. the earliest estimate for PS5's arrival is Holiday 2019. It could easily be Holiday 2020. The latest estimate for a new Xbox is another 4 years (2021), but no sooner than 2020.

I do agree that GPU architecture, efficiency, feature set and the algorithms it will accelerate will all be critical. I'm not sure about int8, but fp16 will probably be more useful in PS5.
Click to expand...
These estimates of 2019 are by fanboys and analysts like Patcher who don't know heads from tails about the semiconductor industry. Some of these "analysts" were thinking months after the Switch's reveal (where we knew its form factor and size that it, a $300 tablet sized device) could rival PS4 in performance......
 
MrFox said:
HBM remains a high risk product. There are more and more indications that the reason GDDR6 exists is precisely because HBM failed to deliver on time, price, speed, and yield.

Maybe it will improve in the next few years, but it's not looking good outside of high-end GPUs. If PS5 is +20TF it will need HBM, otherwise GDDR6 looks like a lower cost and less risky proposition.
Click to expand...

HBM is a high-margin product for memory makers with high demand. Sure SK Hynix seems to have dropped ball on HBM2 regarding initial availability and overal performance, but it certainly doesn't look like a years-long mistake.
They've bragged about HBM being an important new source of revenue not long ago.



Lastly, HBM is indeed very expensive to use as system memory, but in relative terms it's also very cheap to use as cache (which HBCC allows for).

That's why a conservative amount of 500-700GB/s HBM3 (say 8GB in 2020) would go really well with large amounts of DDR4/LPDDR4.
Especially given how LPDDR4 is getting to humongous capacities and speeds. Samsung's latest LPDDR4X are 8GB stacks at 64bit 3733MT/s (30GB/s) .
Get two sets of those in 8GB+8GB clamshell for 128bit and you get 32GB LPDDR4 at 60GB/s, in a tiny footprint for the motherboard's PCB.

Granted, 4*8GB LPDDR4 might be a bit expensive now (close to $7-per-GB in LPDDR4X in 48Gbit chips), but 64Gbit chips should bring that down and with production lines for lpddr ramping up the price should go down during the next couple of year. But 2*HBM stacks for 500GB/s + 4*8GB LPDDR4 chips is something that could be implemented right now, in mass production.
 
Pixel said:
These estimates of 2019 are by fanboys and analysts like Patcher who don't know heads from tails about the semiconductor industry. Some of these "analysts" were thinking months after the Switch's reveal (where we knew its form factor and size that it, a $300 tablet sized device) could rival PS4 in performance......
Click to expand...
If the first set of 7nm is 2019, it won't be affordable for mass market consumer prices until at least 1 year later. At the very least.
 
Pixel said:
These estimates of 2019 are by fanboys and analysts like Patcher who don't know heads from tails about the semiconductor industry. Some of these "analysts" were thinking months after the Switch's reveal (where we knew its form factor and size that it, a $300 tablet sized device) could rival PS4 in performance......
Click to expand...
Pachter estimates holidays 2020, and his explanation for the prediction make sense to me, considering his reasoning is from a market perspective.
 
MrFox said:
Pachter estimates holidays 2020, and his explanation for the prediction make sense to me, considering his reasoning is from a market perspective.
Click to expand...

So you are expecting 2020 holiday, 499$ box sold without loss, less than 200W overall system power consumption pumping out traditional style 20TFlops of fp32 precision? En essence being able to more than triple by 2020 what today is known as xbox one x?
 
iroboto said:
If the first set of 7nm is 2019, it won't be affordable for mass market consumer prices until at least 1 year later. At the very least.
Click to expand...
I don't know if it's been delayed but a few months ago TSMC seemed to be on track for (expensive) volume production in Q2'2018. That means 7nm at the end of 2019 would be at the same point as 16nm in 2016 for the PS4 Pro.

Maybe 2020 might see a 7nm+ product if we're lucky?
 
MrFox said:
I don't know if it's been delayed but a few months ago TSMC seemed to be on track for (expensive) volume production in Q2'2018. That means 7nm at the end of 2019 would be at the same point as 16nm in 2016 for the PS4 Pro.

Maybe 2020 might see a 7nm+ product if we're lucky?
Click to expand...

This is what amd had to say during q3 result releasing
AMD execs also touched on the company's roadmap, particularly its plans for smaller fabrication processes. Su told analysts on a conference call today that AMD would partially skip 12nm, rolling it out only for "subset of products" in the CPU and GPU space next year, while most of the company's resources would be going into developing parts using a 7nm process after 2018. AMD has no plans for 10nm, which is giving Intel a hard time lately.
Click to expand...


https://www.theregister.co.uk/2017/10/25/amd_q3_fy2017_great_but_not_great/
 
AlNets said:
That comparison doesn't take into account how Intel's "10nm" is more like everyone else's "7nmFF", I think... yay marketing.
Click to expand...

My point on quote was the after 2018 part. 7nm for amd is not imminent.

The other thing is 10nm is advertized by tsmc to be great yet there is no discreet gpu made with it on market. I would take 7nm with grain of salt and not expect miracles.
 
manux said:
My point on quote was the after 2018 part. 7nm for amd is not imminent.

The other thing is 10nm is advertized by tsmc to be great yet there is no discreet gpu made with it on market. I would take 7nm with grain of salt and not expect miracles.
Click to expand...
I don't understand your argument, we know amd was skipping 10nm it's been known for a long time. Smartphone chips are using 10nm.
Is a prediction of 7nm in 2019/2020... expecting a miracle?
Was 16nm for ps4 pro at the end of 2016 a miracle?
 
Volume might be an issue for 2019, relatively speaking, although I suppose it wouldn't be the end of the world.
 
MrFox said:
I don't understand your argument, we know amd was skipping 10nm it's been known for a long time. Smartphone chips are using 10nm.
Is a prediction of 7nm in 2019/2020... expecting a miracle?
Was 16nm for ps4 pro at the end of 2016 a miracle?
Click to expand...

7nm is not the miracle. Your expectation of performance at console pricing and power consumption at 7nm on the other hand sounds very miraculous.
 
MrFox said:
HBM remains a high risk product. There are more and more indications that the reason GDDR6 exists is precisely because HBM failed to deliver on time, price, speed, and yield.

Maybe it will improve in the next few years, but it's not looking good outside of high-end GPUs. If PS5 is +20TF it will need HBM, otherwise GDDR6 looks like a lower cost and less risky proposition.
Click to expand...

I suppose your expectation is around 20TFlop or possibly even more.
 
Status
Not open for further replies.

Similar threads

A
Full transcript of John Carmack's QuakeCon 2005 Keynote
3 4 5
Replies
98
Views
32K
nemisis
N
Back
Top