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

Status
Not open for further replies.
Cyan said:
Wasn't HBM2 supposed to be less power hungry than GDDR5 while providing more performance? Time ago it sounded like the next big thing to me.
Click to expand...

I found some tidbits of info. for HBM1 earlier in the thread:

Prophecy2k said:
Couldn't find anything on HBM2 power consumption, but I did find an Anandtech article for estimated power consumption for HBM1 (on the AMD Fury X) and GDDR5x at around 14W and 20W respectively.

Seems in general HBM is lower power, but I think may carry a significant price premium at the moment.
Click to expand...

So you may be right in that HBM is lower power than GDDR5/5x. Clearly, however, there's a reason the two GPU IHVs are limiting its use to only their absolute top end cards. Whether that be down to price and/or capacity I'm not sure, but I'd wager the former being the biggest consideration.
 
Prophecy2k said:
I found some tidbits of info. for HBM1 earlier in the thread:



So you may be right in that HBM is lower power than GDDR5/5x. Clearly, however, there's a reason the two GPU IHVs are limiting its use to only their absolute top end cards. Whether that be down to price and/or capacity I'm not sure, but I'd wager the former being the biggest consideration.
Click to expand...

Cost. Using HBM requires an additional interposer layer in addition to the memory chips. This is currently significantly more expensive than the board space you save by putting the chips on the GPU package. The benefits for HBM are lower power and higher bandwidth per chip.

For lower end cards the power savings are beneficial but the additional bandwidth isn't necessarily required, especially as you move further down the performance chain of the GPU itself.

The next area where HBM would make the most sense would be high performance SOCs that aren't going to be used in price limited designs. At least currently. That dynamic may change in the future dependent on cost reductions to HBM and whether GDDR can continue to scale up in speed to meet the demands of higher performing GPUs.

Regards,
SB
 
The low cost revision of HBM2 is planned to be usable on organic interposers, use half as much TSVs, and ditch the interface layer. The compromise is 200GB/s instead of 256GB/s per stack.
 
Crazy idea: what about taking the PS4-pro APU to a downclock from 911 MHz to 400 mhz (just to get 1,84 tflops) and then also the jaguars back to 1,6 Ghz... then add 4 x 2 gigs GDDR5.... and a screen and a lithium battery. Would never be possible a portable PS4 in such a way ?
 
Dont think so... at 400 mhz it will drain much much less than the ps4 slim, also a new production node can help, and also new wider GDDR5 chips can help... then -with just a wi-fi connection (no disk reader of course)- it can go down to 20 watts... games on USB pen or downloaded. Of course all this is just an idea for PS4 to reach new markets. Also XB1 can do this and probably in an easier way. Seeing the success of Switch I'm sure both MS and Sony are thinking about that... Actually instead of Scorpio that would have been a smarter move...
 
Last edited:
HBRU said:
Dont think so... at 400 mhz it will drain much much less than the ps4 slim, also a new production node can help, and also new wider GDDR5 chips can help... then -with just a wi-fi connection (no disk reader of course)- it can go down to 20 watts... games on USB pen or downloaded. Of course all this is just an idea for PS4 to reach new markets. Also XB1 can do this and probably in an easier way. Seeing the success of Switch I'm sure both MS and Sony are thinking about that... Actually instead of Scorpio that would have been a smarter move...
Click to expand...

It's far too early to be declaring the success of the switch after only two months on the market. Did we forget how fast PS4 and XB1 sold at the start of this gen, and look how XB1 sales tailed

Plus, I'm sure the success of the console was also much more down to Zelda being a 95+ metacritic scored launch game, rather than the portability of the console. Had the switch launched without Zelda I'm 100% confident the launch wouldn't have been as strong out the gate. The switch will live or die off the back of it's software library and it clearly needs more than Nintendo support the thing for it to excel in that area.
 
With the PS4PRO APU (GPU downclocked to 400 MHZ, CPUs to 1,6 Ghz... maybe with 4 GDDR5 chips of 2 Gbytes, 256 Giga SSD) should be quite easy to build something like this with a quite durable battery...
Only developers could patch the PS4 games in quite an easy way ? The TFLOPS are the same but geometry of the GPU would be different....

 
HBRU said:
With the PS4PRO APU (GPU downclocked to 400 MHZ, CPUs to 1,6 Ghz... maybe with 4 GDDR5 chips of 2 Gbytes, 256 Giga SSD) should be quite easy to build something like this with a quite durable battery...
Only developers could patch the PS4 games in quite an easy way ? The TFLOPS are the same but geometry of the GPU would be different....
Click to expand...

The Xbox One's SOC was stated to be able to drive power down in its idle state to 2.5% of max power. For chips that burn 100W and up, that's a sizeable chunk of a portable's power budget that's eaten up by a chip that is just slightly above being fully off.
The PS4's power consumption was notably higher, and had a separate chip for its standby usage.

The Pro's power consumption is higher by some amount, although it's not the same process or design. That may make it hard to compare, although the chip's GPU is bigger.
If the SOC is not designed for mobile use, however, there's a good chance just being on is going to leave very little for any scenario where it needs to be used.
 
Not the same chip, but I saw at Tom´s the 480 (8Gb variant) consuming 15w just idling.
The Switch consumes less than 10W in portable mode with the joycons at full steam.
 
Exactly Michellstar... at 400 mhz (or less, maybe even 250 mhz are enough if 720p is choosed for the screen output) the PS4PRO-APU + GDDR5 is going to absorb something like 25 watts@16FF+. Beyond that... the APU can be also shrinked at 10nm quite immediatly (as Apple is alredy producing at 10 nm with TMSC)... additionaly unfortunate chips are going to the mobile version, best chips to the PS4PRO (slim). Regular PS4 dismissed. So I think if Sony wants it can be done.... and it has also industrial meaning. Even MS can go a similar rouote with Scorpio's APU.
 
From what I understand, the 10nmFF node (Samsung/GF) is more-or-less a stop-gap node because the gains are not going to be as significant (still better than 28 -> 20nm planar) for the cost. That Apple is making large orders for that node speaks more to necessity and being able to afford the premiums at this nascent stage. Recall that Apple did have the A8/A8X produced even though it was a terrible node & price 3 years ago.

A mobile PS4 is not going to be sold for up to a grand.

The wider memory bus relative to most mobile offerings will also have implications to the bottom line for power consumption & may present issues for minimum die size anyway.
 
HBRU said:
Exactly Michellstar... at 400 mhz (or less, maybe even 250 mhz are enough if 720p is choosed for the screen output) the PS4PRO-APU + GDDR5 is going to absorb something like 25 watts@16FF+. Beyond that... the APU can be also shrinked at 10nm quite immediatly (as Apple is alredy producing at 10 nm with TMSC)... additionaly unfortunate chips are going to the mobile version, best chips to the PS4PRO (slim). Regular PS4 dismissed. So I think if Sony wants it can be done.... and it has also industrial meaning. Even MS can go a similar rouote with Scorpio's APU.
Click to expand...
The number for the Xbox One SOC is where widespread clock and power gating have shut off much of the chip, not including other components. On its own, a mostly shut-off Durango would eat up a quarter of the Switch's power budget. The PS4, if it were as effectively gated, would be worse.
That's the bare minimum that doesn't go away. Once the circuits stop being power-gated, the leakage component that is inherent to just being powered can only go up since most of the chip was cut off from voltage.
The Polaris idle numbers include a significant fraction of the chip being power-gated off, although as a full product it has other components.

FinFET has helped with leakage, but once things are active it's not enough to overcome designs that are not intended for high utilization at 1-2 orders of magnitude less power.
 
SK Hynix plans to launch denser capacities later on with up to 16 Gb or 2 GB chips in the second half of 2018.

What about a new ps4-K with 16 giga GDDR6 ?
Mem controller should not be much different of the GDDR5 one...
What about Scorpio going be released with an old mem type.... this is actually disappointing.
SK-Hynix-GDDR6-3.jpg


http://wccftech.com/sk-hynix-gddr6-nvidia-volta-gpu-gtc-2017/
 
256-bit PS4pro bus, with just a memory upgrade, will double its memory bandwidth -using GDDR6- going from today's 218 GB/s to 436 GB/s .... and this without much silicon changes... also the rest can quite easily overclocked to reach and go beyond Scorpio TFLOPS....
 
Status
Not open for further replies.

Similar threads

A
Full transcript of John Carmack's QuakeCon 2005 Keynote
3 4 5
Replies
98
Views
34K
nemisis
N
Back
Top