the costs for HBM is more directly correlated with the assembly - as in the cost of stacking the chips on properly, vs the actual chip themselves?
Predict: Next gen console tech (9th iteration and 10th iteration edition) [2014 - 2017]
- Thread starter Shifty Geezer
- Start date
- Status
Rumors are saying Vega 20 is supposed to be 150W chip with more than 12TF on 7nm with half rate dp. I assume Navi on 7nm will be without the DP and be more power efficient and probably mean we can hit 15TF by 2020 on power budget of 150W for the GPU. Pricing on the other hand might me more limiting. 8 'low clocked' zen+ cores will probably use less power than the current jaguar cores. 32GB of GDDR6 should be available and probably cheaper than 8GB of GDDR5 was when the ps4 launched.
SSDs are cheaper now but spinning rust is cheaper. I don't expect ssds to make it to consoles except for maybe a premium version.
SSDs are cheaper now but spinning rust is cheaper. I don't expect ssds to make it to consoles except for maybe a premium version.
Blazkowicz
Legend
HDD have a minimum price, like $50 or $40.
We're pretty much hitting 240GB/256GB for really cheap SSDs, so cheap hardware and SSD will more and more often go in hand. But if we're comparing 2TB HDD with 2TB SSD, sure that breaks down atrociously.
I tihnk we want both, but we'd get one less teraflops because of the $ budget.
Games are only getting bigger. I would find it hard for console makers to justify a SSD large enough for games. Minimum size of drives for next gen console would need to be 1TB, even if drives goes from $0.30 down to $0.12 per gb in 3 years, you are looking at a $120 for an ssd vs <$40 for an hard drive, if you competitor was using hard drives, they would undercut you by almost $100. And that is just the minimal drive, a 2TB hdd is like $60 but a 2TB SSD would be $240, good luck launching a console that is $200 more than the competitor's version with an HDD.
The memory bandwidth bottleneck is an issue that may plague PS5/Xbox4. HBM manufacturing complexities may keep it out of the pricerange of $400-500 consoles. GDDR5X/GDDR6 may not be sufficient. If HBM is still too expensive the console manufacturers may need to include a gpu sram cache to suppliment the GDDR6.
Transponster
Newcomer
Wishful Prediction:
PS5:
7nm
8 (post-)Zen+ cores(desktop/laptop performance lvl) at ~2.5GHz
10TF Navi GPU
32GB of low-cost HBM with bandwith "slightly" under a TB.
Targeting $499, ~190W without losing too much efficiency to the standart $399, ~150W(AMD's upcoming "Scaleability" should get us there)
November 2019
PS5 Pro:
5nm
Same CPU at 3+GHz
20TF GPU
64GB of HBM with higher bandwith
Same targets, possible of running all PS5 games in native 4K(doing that on original PS5 is rather wasteful even if you focus on the GPU setup), 8K marketing(?
; reconstruction("upscale") from 1/4 of the picture).
November 2022
/A year shift is likely.
As long as prices are going down, inital prices and production costs shouldn't matter all that much. MS showed this gen, selling millions of X1's for 499, when there was(not often in stock, though) a better $399 system on the market, that selling a new and hot product for $399(/little to no profit margin) doesn't make much sense, even Wii U sold rather well in its first quarter, so even with production costs below $400 they might and should put up a $499 price tag for at least half a year.
$399 Price and production cost target can only hurt, not just PS4 and PS4 Pro(that can do great beyond this decade), especially if PS5 comes out in 2019, but PS5 itself in future, being less future-proofed due to lower budget.
PS5:
7nm
8 (post-)Zen+ cores(desktop/laptop performance lvl) at ~2.5GHz
10TF Navi GPU
32GB of low-cost HBM with bandwith "slightly" under a TB.
Targeting $499, ~190W without losing too much efficiency to the standart $399, ~150W(AMD's upcoming "Scaleability" should get us there)
November 2019
PS5 Pro:
5nm
Same CPU at 3+GHz
20TF GPU
64GB of HBM with higher bandwith
Same targets, possible of running all PS5 games in native 4K(doing that on original PS5 is rather wasteful even if you focus on the GPU setup), 8K marketing(?
; reconstruction("upscale") from 1/4 of the picture).November 2022
/A year shift is likely.
As long as prices are going down, inital prices and production costs shouldn't matter all that much. MS showed this gen, selling millions of X1's for 499, when there was(not often in stock, though) a better $399 system on the market, that selling a new and hot product for $399(/little to no profit margin) doesn't make much sense, even Wii U sold rather well in its first quarter, so even with production costs below $400 they might and should put up a $499 price tag for at least half a year.
$399 Price and production cost target can only hurt, not just PS4 and PS4 Pro(that can do great beyond this decade), especially if PS5 comes out in 2019, but PS5 itself in future, being less future-proofed due to lower budget.
Yeah, I was not thinking very technically - just that Sony will want a big number and try to achive it. I suppose looking at PS4 figures then a ~5x TF jump would be good so 9/10TF and then (of course) if everythng else get's a double Scorpio boost we'll be seeing native 4k60 across the board (or at least should be) with much better textures/LOD/pop-in etc
From this Perspective yes, Flops could become the new "bits", "MHz", the easiest way to market the strength of the product despite it not being representative of how the actual speed of the system is occurring.
I guess we can never get away from the horsepower race.
Sent from my iPhone using Tapatalk
Ah yes, the old processor speed limits - good point. The other 'issue' I could see is why would we want to push past 4K? It's already pointed out that you need to be sitting(or more likely standing) pretty close to a big screen to start to notice any pixels/issues. I wonder what the next marketing feature will be in the regard...maybe units of reality where graphical effects that acurately simulate real life are considered features (like HDR)!?
Michellstar
Regular
OT: debatable given the scene today with PC. I was an OG Oculus backer and since then to now, no significant gaming title has been made. Experiences are a different matter though. VR experiences seem to hold up well.
Sent from my iPhone using Tapatalk
Curiously, but is anyone else reading the Vega 10, 11, 12 thread? Some of the posts show concern about GCN unable to scale beyond 64 CUs. Leaving clock speed as the only method to increasing output.
If true, that would severely limit consoles ability to scale. From my understanding the goal for consoles is to go wide and reduce clock speed to reach a desirable TDP.
Anyway, just putting that out there. Could be an important piece of information for determining when a new generation hits. Perhaps a new generation is when we leave GCN.
Sent from my iPhone using Tapatalk
If true, that would severely limit consoles ability to scale. From my understanding the goal for consoles is to go wide and reduce clock speed to reach a desirable TDP.
Anyway, just putting that out there. Could be an important piece of information for determining when a new generation hits. Perhaps a new generation is when we leave GCN.
Sent from my iPhone using Tapatalk
Well this is a stretch and of course the software tools would have to be created from scratch but it would be interesting to see a combination of Intel/MIcron's Crosspoint tech plus some procedurally generated stuff like this Euclideon.jazz as a big change for a new generation of console.
http://www.hardocp.com/article/2016/09/12/bruce_dell_euclideon_holoverse_interview/1#.V-qAMfArJhF
The tech in the feature streams from hard disks so if you were using something like Crosspoint to use as your base instead something fundamentally new could be created.
http://www.hardocp.com/article/2016/09/12/bruce_dell_euclideon_holoverse_interview/1#.V-qAMfArJhF
The tech in the feature streams from hard disks so if you were using something like Crosspoint to use as your base instead something fundamentally new could be created.
Anarchist4000
Veteran
It can scale past 64 and AMD has already demonstrated the capability with Tonga. That's not likely how they will scale it though. Fatter CUs with more efficient and faster processors is the more likely solution. With current fab tech limiting scaling, that's more than ample for the die sizes produced. With the increasing focus on compute, getting more ROPs and geometry capabilities with async now present isn't really worthwhile.
Michellstar
Regular
The little we know about Navi:
-Nextgen memory (after HBM2) whatever it is
-Scalability, whatever that mean
Michellstar
Regular
I believe a more focused gaming market will follow mainstream prices, the best shot right now It´s PSVR
Anyone saw this?
It would open up the possibilities if the lower cost HBM is truly low cost.
But wow, HBM3 is really feeding my wishful thinking
http://www.extremetech.com/gaming/2...tions-to-blow-the-doors-off-the-memory-market
It would open up the possibilities if the lower cost HBM is truly low cost.
But wow, HBM3 is really feeding my wishful thinking
http://www.extremetech.com/gaming/2...tions-to-blow-the-doors-off-the-memory-market
- Status