If they still have momentum in 2018, they won't lose it by being a year later than MS. PS1->PS2 carried over despite Dreamcast's earlier positive release.
Unless 10nm is a trouble-free node, an earlier 2018-2019 release will provide unsatisfactory performance improvement and low baseline target for devs.
If they still have momentum in 2018, they won't lose it by being a year later than MS. PS1->PS2 carried over despite Dreamcast's earlier positive release.
Unless 10nm is a trouble-free node, an earlier 2018-2019 release will provide unsatisfactory performance improvement and low baseline target for devs.
Last changeover we got an 8 year gap, and even that provided an unsatisfactory level of performance improvement. The kind of RAM and GPU increase we used to get in 5-6 years, paired with a pathetic CPU. 6 years in an industry of slowing development in certain areas (especially CPUs) is going to underwhelm regardless.
I'd like to think that was a victim of timing and circumstances. AMD had major trouble with Bulldozer + GF and both MS/Sony dealt with PS360 hangover.
Every bit helps to push the baseline higher. PS5X2 will underwhelm less if they can get mature yields and a tock off 10nm, as opposed to being forced to 16nm for a rushed 2018 release.
If they still have momentum in 2018, they won't lose it by being a year later than MS. PS1->PS2 carried over despite Dreamcast's earlier positive release.
Unless 10nm is a trouble-free node, an earlier 2018-2019 release will provide unsatisfactory performance improvement and low baseline target for devs.
If 1080p remains the Television standard in 2020, and they left consoles at 1080p as the standard for developers, that'd be an incredible amount of overhead.
Even if the power of the hardware doesn't advance like it did in the 90s/early 2000s, the fact that inf resources aren't being spent going from 480p on 360 to 900p on Xbone means we could get an extremely large relative increase for developers to throw at particle effects and inefficient but neat looking physics and environments.
Also AMD half of the AMD/ATI equation underperformed so much this gen they have infinite room for improvement.
TSMC said it foresees “major product advancements” in three areas during the next two years, including high-end smartphones, high-performance computing and emerging applications such as virtual reality, gaming and automotive electronics.
“All of those advances will be supported by TSMC’s 10nm and 7nm technologies,” Co-CEO Mark Liu said at an event to announce the company’s fourth-quarter 2015 financial results. The company expects to complete process and product qualification for 10nm chips and begin customer tapeouts during the first quarter this year, he said. The company said it is on schedule to start production at the 7nm node in 2017.
7nm in 2017 seems kinda optimistic, unless it's just for testing trial runs for later mass production. But I wonder about the possibility for console-class 7nm APUs releasing in say Fall 2020 ?
Depending on what 7nm from TSMC will actually mean, you/d think it might allow a pretty nice increase in transistors and performance, at reasonable power consumption over the now ancient 28nn node and even the upcoming 14/16nm FinFET node. It's been said that 14/16nm FinFET is based on 20nm design rules. Unlike Intel, whos 14nm CPUs are "true 14nm". And we don't know what TSMC, Samsung, GloFo 10nm is all about.
That said, I don't actually expect next gen console APUs to be produced on anything beyond 10nm.
Back on topic of High Bandwidth Memory. I'd sure hope next gen consoles don't have less than 1 TB/sec bandwidth, regardless of the amount of HBM2 used.
But it's not binary. PS5X2 performance can suck less if they wait for a mature 10nm. PC enthusiasts will benefit greatly from lowest common denominator console spec being better.
They have the choice of giving a headstart to MS or cutting their generation slightly shorter than they want it to be. When you're in the lead there's a temptation to ride it out, but the best way to maintain momentum will be to not allow the competition a one year head start on the market. The PS brand is very strong.
It depends really, Sony didn't rush anything out when the Dreamcast released early, and they were right to do so as the PS1 still had a few great years ahead.
I'm actually of the opinion that the PS3 could've released in 2007 and be in a better position than they were back then.
This won't be as binary as las time IMO. PS5/X2 can come out in 5/6 years time and PS4/X1 can live a long life alongside with cross gen games being even easier than before assuming a common x86 architecture.
Potentially lets Sony/MS milk early adopters while still growing the previous gen to the mainstream.
This won't be as binary as las time IMO. PS5/X2 can come out in 5/6 years time and PS4/X1 can live a long life alongside with cross gen games being even easier than before assuming a common x86 architecture.
Potentially lets Sony/MS milk early adopters while still growing the previous gen to the mainstream.
Well the stacks only come in 1 (8GB max) 2(16GB max) or 4(32GB max) anyway.
Until HBM3 or whatever comes around, middle of the road is what i'm going with.
More than the amount of RAM, i think the bandwidth upgrade will matter more going into the next generation. I think that the days of RAM upgrade sizes from console to console being 5, 6, 7 or even 8 times is unfeasible even 3 years from now.
i-7 level zen CPU, 16GB HBM at 756GB/s and 8TLOP AMD GPU packed into a 14(or maybe even 10?) nm process APU is my final verdict for the PS5. The CPU is gonna be a huge revelation compared to the Jaguar
Why do you think that? In one year (of consumer availability) we are going from HBM to HBM 2 which offers an 8x increase in maximum capacity. Why do you think that it will remain stagnant for 3 more years? HBM 3 will probably be available in consumer products by 2019 and even if it is not I would not be surprised if they further develop HBM 2 to support 8 stacks and further double or quadruple the density making 64GB and maybe 128GB a possibility.
I can see Zen being used if the perf/watt and its absolute power draw are acceptable otherwise the console CPU is likely to be a jaguar successor. I do doubt the core count will go up though meaning IPC improvements are going to be important regardless of which architecture they use.
A 14nm Fury Nano will draw less than 100 watts for the whole card and the die size will come down a good chunk. With further architecture enhancements to reduce power draw and some tweaking of the design I could see a 14nm 8Tflop GPU at around 300mm^2 with a low enough power draw working. It might be viable to be a bit more aggressive but it depends on the CPU. With Zen 8TF seems like a good maximum but with a jaguar successor 10TF might be possible.
Memory density doesn't follow a plan. We thought we'd be stuck at 2 GB Ram for this gen for a reason. And then they managed to double memory density and did it again in rapid succession. Just because it's not possible now doesn't mean it's not possible in three years.
This gens loading times say we need more Ram. And devs will say they want 4x the bandwidth and 8x the amount. There is never too much Ram for a developer.
Sony has especially less incentive with the underperformance of large sections of its businesses that aren't phone camera or game related. iirc there is a third division that's doing well just can't remember it off the top of my head.
I don't expect a huge RAM jump for the next gen until a very cost effective solution comes to the data transfer. It's already a problem, when you have 6GB to fill with a 5400rpm drive on sata2.
For use in upcoming GPU like Polaris/Pascal and Zen-based APUs that could be used in next-gen consoles.
Key specs:
1, 2 or 4 stacks.
1024 bit interface width per stack
up to 8 layers per stack (up from 4) in dual channel mode
up to 1 GB per layer (up from 256 MB)
1 GHz clock (up from 500 MHz)
Damn! With this and affordable SSDs coming down rhe pipeline (2-4TB costing around $300-500) for late 2016 to early 2017. It looks like I will have to build a new gaming tower!
Why do you think that? In one year (of consumer availability) we are going from HBM to HBM 2 which offers an 8x increase in maximum capacity. Why do you think that it will remain stagnant for 3 more years? HBM 3 will probably be available in consumer products by 2019 and even if it is not I would not be surprised if they further develop HBM 2 to support 8 stacks and further double or quadruple the density making 64GB and maybe 128GB a possibility.
I can see Zen being used if the perf/watt and its absolute power draw are acceptable otherwise the console CPU is likely to be a jaguar successor. I do doubt the core count will go up though meaning IPC improvements are going to be important regardless of which architecture they use.
A 14nm Fury Nano will draw less than 100 watts for the whole card and the die size will come down a good chunk. With further architecture enhancements to reduce power draw and some tweaking of the design I could see a 14nm 8Tflop GPU at around 300mm^2 with a low enough power draw working. It might be viable to be a bit more aggressive but it depends on the CPU. With Zen 8TF seems like a good maximum but with a jaguar successor 10TF might be possible.
I am skeptical about seeing hbm3 in the next set of consoles (I may very well eat crow =P), but I do think we will see Zen and HBM2.
I was under the impression that for hardware to be released in 2020, a final build/prototype would be put into testing by 2018 (2year gap) .
And with consoles already having most of their performance being squeezed out of them during this past year, 2years into the Gen, I suspect we will be swing consoles releasing in 2019 making the tech cutoff around 2017 IMO.
These current consoles will not age as well as last gens did as they were severely underpowered from the get-co considering the tech jumps we had over the last 3-4years.
Damn! With this and affordable SSDs coming down rhe pipeline (2-4TB costing around $300-500) for late 2016 to early 2017. It looks like I will have to build a new gaming tower!
Waiting for 10 nm takes too long imo, my guess is it won't be console ready until 2020 or so (2017 for Intel, 2018-2019 for AMD, a year more for consoles). HBM2 with a 14 nm APU will have to do for a 2018 release. The jump won't be as big as previously, but there's little point in waiting years for the next node. It'd feel like a bigger jump for sure but do you really want to stretch the current underpowered hw that far? I feel it's already very limiting to game design on the CPU side, and if Sony or MS want their boxes to run VR properly they'll need a more suitable system for that.
Unless Sony and Microsoft can keep their software suppliers happy with how they handle it... I wouldn't expect new hardware in 2018 unless it's 100% BC and mostly FC with PS4/Xbone. Software engines are costing more and more to make, and take longer and longer to build, and that includes games as well.
With games taking 2-3 years to make on top of preexisting engines, that means companies like Ubisoft at year 4 of a console cycle either have to wait to retool their game, hope Sony has a good tech set devkit for PS5 already available, or have to retool the game part way through development... Unless Sony and Microsoft start adopting an apple-like approach to hardware, I just don't see how the 5 year hardware cycle is going to work anymore.
No one probably knows but has anyone associated with designing and building the new consoles done an interview or even talked on twitter about their opinions or predictions for the next gen consoles?
Maybe a Mark Cerny bathtub AMA with a rubber ducky?
HBM is new so advances will come pretty quickly to begin with and then it will slow down once the easy improvements have been made. I think we will see HBM 3 in time for the next gen consoles but that is a guess and 3 years in tech is an eternity.
2 years seems too long to me due to how fast tech moves. In the gpu market you expect a 6 month cadence from final tape out to consumer availability so I don't see why it would take much longer for a console other than to build more inventory. As long as you know what the target hardware will be as a minimum you can make sure devs have enough data to build the first wave of games.
PS4 specs were announced 9 months prior to launch and I doubt there were more than a handful of prototypes based on final silicon at that point. So to allow for inventory ramping you could get away with a 9 to 12 month lead to fix specs.
And with consoles already having most of their performance being squeezed out of them during this past year, 2years into the Gen, I suspect we will be swing consoles releasing in 2019 making the tech cutoff around 2017 IMO.
These current consoles will not age as well as last gens did as they were severely underpowered from the get-co considering the tech jumps we had over the last 3-4years.
If a nov 2019 launch is planed, which is what I am expecting, then the hardware won't be locked until end of 2018 IMO, maybe even early 2019.
What jumps? We have been stuck at 28nm for 5 years. If anything fabrication tech has stagnated massively which means our CPUs and GPUs have both slowed down their advancements compared to previously.
The biggest reason why the current consoles fell further behind compared to PCs is power consumption. In 2005 a top tier GPU did not have a TDP of 250 watts, and the current consoles themselves were lower power than their predecessors to make them both more affordable and more reliable.
I still doubt that the devs are taking full advantage of async compute and other features.
Next gen I can see Sony and MS being a bit more aggressive on the hardware front. I see $399 as the target launch price but with the success of the PS4 sony may be willing to subsidise the hardware a bit more than this gen and hopefully MS will make a games console that also happens to be a media centre rather than an all in one box.
The current APUs are around 330mm^2. Maybe we will see upto 400mm^2 if power allows and Sony are willing to subsidise a bit more but no bigger. Fiji and GM200 are both around 550mm^2 - 600mm^2.
I think next gen consoles will be somewhat less conservative hardware for the year they're released compared to PS4/XB1 in 2013. Although not bleeding edge like 360/PS3 were in 2005-2006.
10nm FinFET - 160w power budget for the APU.
HBM2 (if consoles release in 2019) or HBM3 (if 2020).
8x Zen cores, 2.5+ Ghz
10 Tflop GPU bare minimum and enough pixel through put to support some types of games at native 4K, even though many AAA games will be native 1080p+ upscaled to 4K.
By the time PS5 come out 4k tv's will be the norm, heck VR might take off. Thus the need for high resolutions and high frame rates, Sony needs 7nm chips for native 4k games at affordable price points. For a significant upgrade, consoles have to be released no sooner than 2020 or 2021 at the latest.
And with consoles already having most of their performance being squeezed out of them during this past year, 2years into the Gen, I suspect we will be swing consoles releasing in 2019 making the tech cutoff around 2017 IMO.
These current consoles will not age as well as last gens did as they were severely underpowered from the get-co considering the tech jumps we had over the last 3-4years.
By the time PS5 come out 4k tv's will be the norm, heck VR might take off. Thus the need for high resolutions and high frame rates, Sony needs 7nm chips for native 4k games at affordable price points. For a significant upgrade, consoles have to be released no sooner than 2020 or 2021 at the latest.
HBM is new so advances will come pretty quickly to begin with and then it will slow down once the easy improvements have been made. I think we will see HBM 3 in time for the next gen consoles but that is a guess and 3 years in tech is an eternity.
2 years seems too long to me due to how fast tech moves. In the gpu market you expect a 6 month cadence from final tape out to consumer availability so I don't see why it would take much longer for a console other than to build more inventory. As long as you know what the target hardware will be as a minimum you can make sure devs have enough data to build the first wave of games.
PS4 specs were announced 9 months prior to launch and I doubt there were more than a handful of prototypes based on final silicon at that point. So to allow for inventory ramping you could get away with a 9 to 12 month lead to fix specs.
If a nov 2019 launch is planed, which is what I am expecting, then the hardware won't be locked until end of 2018 IMO, maybe even early 2019.
What jumps? We have been stuck at 28nm for 5 years. If anything fabrication tech has stagnated massively which means our CPUs and GPUs have both slowed down their advancements compared to previously.
I still doubt that the devs are taking full advantage of async compute and other features.
Next gen I can see Sony and MS being a bit more aggressive on the hardware front. I see $399 as the target launch price but with the success of the PS4 sony may be willing to subsidise the hardware a bit more than this gen and hopefully MS will make a games console that also happens to be a media centre rather than an all in one box.
The current APUs are around 330mm^2. Maybe we will see upto 400mm^2 if power allows and Sony are willing to subsidise a bit more but no bigger. Fiji and GM200 are both around 550mm^2 - 600mm^2.
The newest tech will cost the console makers too much upfront to implement right away, that is why HBM3 will be a no go. Think about the lag time and how long it takes newer/the latest gpu tech to reach mainstream laptops.
And HBM2 will prob have a revision of the process after it's first year release so id expect HBM3 to be announced/released two years after HBM2 releases.. So possibly Q1 or Q2 in 2019.
Current consoles are using GPU/apu tech that was very under powered and a behind what was on the market when released (console gpu tech from 2012..announced and released). For current Gen consoles... the gpu in the ps4 hovers around/near an 7850 while the xb1 hovers around/near an 7790.
Just because the PS4 announced there specs 9 months or so prior to release... Does not mean that they hadn't tested the tech months prior or decided on the gpu months before hand. If you remember, ps4 originally was slated to have 4gigs of GDDR5 memory but later went with 8 as it became very cheap due to the PC market.
I can only assume that testing of prototypes and other research aspects take longer than 9months to work out the kinks. I believe that the spec announcement is to let the devs know what the actual decided power/spec are to optimize release games and coding for the release day. The reason why I say 2years approximately is that that helps provide the devs ample time to start becoming familiar with similar spec'ed hardware so that you can actually have games on release day. Maybe this Gen was easier for devs to work with as it is similar to pc architecture unlike cell or the 360 architecture. And console makers always desire ample stockpile for console releases so that they do not run out of supply during the holiday release.
Sony still cannot afford to absorb costs upfront as there entire company as a whole had some serious draw backs over the last few years..(at least not right now) Last time I heard, there camera division and the PS4 division sections were one of the few sections performing well with the ps4 helping to prop up Sony a bit (phone and tv business is tanking, they sold off there laptop business, they were liquidating assets I believe before as well).
I hope they take there added success and create a console in the future that is a power house and may cost $500-550 to make but then sell it to the consumer for $400 as it will help the consoles age better with the drastically changing pc tech (gpu end, pertaining to resolution, new shading and polygon techniques and etc..as implementing it costs serious resources on the gpu). MS will most likely be going with this route again in the future as they do not like being the underdog. Due to the gpus in the consoles being very similar to pcs, desktop gpus were able to estimate the performance of the consoles months before release, resulting in a PR nightmare for MS. Then it costing them around $100 lost per console earlier on with having to drop the price and give games away for free during the first year of the console (whereas last Gen the architectures were different and it depended on dev coding and how refined there game engines were).
Only time will tell. Next gen will have the same architectures so the turnover process for design to prototype to production/announcement may be shorter! (
Fat chance with that. I doubt steam machines even matter the conversation of consoles 5 years down the line, for the exact same reasons as now. Closed box console ecosystems are just entirely different audiences.
He's mocking Nvidia for trying to pass off a couple of 980M's as Pascal. Given that they weren't displaying real pascal silicon, odds are stacked against Pascal coming out in the next 6 months or so.
Samsung begins mass production of 4GB HBM2 memory chips
Just under a week ago, JEDEC updated the High-Bandwidth Memory standard with provisions for bigger, faster memory packages. Hot on JEDEC's heels, Samsung has taken the wraps off its mass-production 4GB HBM2 chips this evening.
The company says it's fabricating these 4GB dies on its 20-nm process. Each of these packages comprises four 8Gb core dies atop a buffer die at the base of the stack. Consistent with JEDEC's specifications, each of these HBM2 dies will offer 256 GB/s of bandwidth. For comparison, Samsung says that figure is a little over seven times the bandwidth of its through-silicon via (TSV) 4Gb GDDR5 dies.
Samsung also plans to release 8GB HBM2 packages this year, a move it says will allow graphics card designers to enjoy space savings of up to 95 percent versus designing around GDDR5
The 4GB HBM2 package is created by stacking a buffer die at the bottom and four 8-gigabit (Gb) core dies on top. These are then vertically interconnected by TSV holes and microbumps. A single 8Gb HBM2 die contains over 5,000 TSV holes, which is more than 36 times that of a 8Gb TSV DDR4 die, offering a dramatic improvement in data transmission performance compared to typical wire-bonding based packages.
Samsung’s new DRAM package features 256GBps of bandwidth, which is double that of a HBM1 DRAM package. This is equivalent to a more than seven-fold increase over the 36GBps bandwidth of a 4Gb GDDR5 DRAM chip, which has the fastest data speed per pin (9Gbps) among currently manufactured DRAM chips. Samsung’s 4GB HBM2 also enables enhanced power efficiency by doubling the bandwidth per watt over a 4Gb-GDDR5-based solution, and embeds ECC (error-correcting code) functionality to offer high reliability.
In addition, Samsung plans to produce an 8GB HBM2 DRAM package within this year. By specifying 8GB HBM2 DRAM in graphics cards, designers will be able to enjoy a space savings of more than 95 percent, compared to using GDDR5 DRAM, offering more optimal solutions for compact devices that require high-level graphics computing capabilities.
HBM is new so advances will come pretty quickly to begin with and then it will slow down once the easy improvements have been made. I think we will see HBM 3 in time for the next gen consoles but that is a guess and 3 years in tech is an eternity.
AMD isnt discussing how much the latency was reduced as they want to keep the memory subsystem under wraps so competitors have to figure it out themselves.
But that was before the release of HBM1, soooo....
Or these slides that just have the title "Low Latency" with the subtitle "Pseudo channel improves tFAW by 60% compared to DDR4" and a nice picture. But to be honest I don't understand what tFAW is so I don't know if it's talking about latency to CPU/GPU or something else.
Just because the PS4 announced there specs 9 months or so prior to release... Does not mean that they hadn't tested the tech months prior or decided on the gpu months before hand. If you remember, ps4 originally was slated to have 4gigs of GDDR5 memory but later went with 8 as it became very cheap due to the PC market.[/spoiler)]
I believe the switch from 4 to 8 was 'very last minute'. But yeah, adding more of an already locked-down technology is a bit 'easier' than switching to a new generation of a certain technology.
This was a tongue and cheek article showing that Nvidia's Pascal debut was using two 900 series Maxwell mobile GPUs on the PX2 with Nvidia telling everyone that those obviously older Maxwell parts were Pascal. They manufacturing date on the chips were dated back to January of last year.
Never heard of it. How far to market might it be? 2x bandwidth of GDDR5 would be a pretty good increase that may not come with the associated issues of HBM.
Never heard of it. How far to market might it be? 2x bandwidth of GDDR5 would be a pretty good increase that may not come with the associated issues of HBM.
I'd assume fairly quickly (6 months - 1 year), given the Nvidia Pascal rumors that are floating around. (GDDR5X being on the lower end chips while HBM2 is on the higher end).
Never heard of it. How far to market might it be? 2x bandwidth of GDDR5 would be a pretty good increase that may not come with the associated issues of HBM.
GDDR5-X is what will most almost likely be used in the lower to mid tier AMD Polaris and Nvidia Pascal cards. What isn't known is where the line will get drawn between HBM2 cards and GDDR5-X cards.
Never heard of it. How far to market might it be? 2x bandwidth of GDDR5 would be a pretty good increase that may not come with the associated issues of HBM.
It's a minor modification of GDDR5 supposedly so I think that both Pascal and Polaris may end up using it from the start. HBM2 is quite a bit faster though so I expect them all to complement each other - GDDR5 for lower end, GDDR5X for middle and HBM for top end.
