phoenix_chipset
Regular
*Zen lite
I will eat my gamecube and post it on here if the Ps5 doesn't have at least 24GB of memory. Assuming there is something called a Ps5!
Predict: Next gen console tech (9th iteration and 10th iteration edition) [2014 - 2017]
- Thread starter Shifty Geezer
- Start date
- Status
I will eat my gamecube and post it on here if the Ps5 doesn't have at least 24GB of memory. Assuming there is something called a Ps5!
Well, you are certainly wrong about the SSD. When we look at current SSDs, the top performing SSD (Samsung 960 PRO M.2 PCIe NVMe) only reaches 3500MB/s (read) and 2100MB/s (write) and still costs about 1250€. Maybe you meant 500MB/s read/write. This you could achieve with a SAMSUNG 850 EVO SATA. But this baby still costs about 650€. Cheapest 2TB SSD (Crucial MX300) is still 550€. I really, really doubt prices will come down that much in two years.
Regarding Zen, Navi & 24GB HBM3. I think we can pretty sure that a 2019 console SOC will use TSMC 10nm process. Maybe you could build a beast like this, but what would the power consumption be? I doubt it will still fit in a console chassis. And it will be pretty expensive to build. 10nm is way more expensive than 16nm. But the real question here is what do you do with 16 Zen cores? It's not easy at all to keep 16 cores fed. Ahmdal's Law will kick you in the nuts here, full swing. Navi may or may not be a possibility, but I would not count on it. For a 2019 it might be too risky and currently we know too little about it. I also doubt 24 GB HBM3. HBM3 is scheduled for 2019/2020 by Samsung. Won't make it in time if PS5 is released 2019. I think 8 GB HBM(2)/GDDR6 + 16 GB DDR4 is far more realistic (and probably cheaper). Remember that Vega is supposed to have support for paging.
<rant>I think one of the challenges for the next generation is to figure out what your target audience will look like. A lot of people, maybe the majority, will still use 1080p displays. 8K will still be a niche. Digital distribution may be viable for parts of the world, but not for others (especially given the expected sizes of future games). BC will also be a hot topic. I'm curious to see how it will play out.</rant>
I think the next consoles will be 7nm in 2019/20, maybe 21 for MS, likely driven by when reasonable yields and costs are achievable for a ~300mm SOC.
I don't think there's a need to go for more than a 8 core setup with SMT that will fit into a ~25-30 watt budget.
On the GPU side, I think around a 12 Tflop target is reasonable, double Scorpio's shader count and at a slightly higher clock.
The big question is the RAM amount and the type. The bandwidth will likely have to be around 1TB/sec so I think HBM might be the only possibility and that will certainly limit size. Let's see how Vega turns out...
I don't think there's a need to go for more than a 8 core setup with SMT that will fit into a ~25-30 watt budget.
On the GPU side, I think around a 12 Tflop target is reasonable, double Scorpio's shader count and at a slightly higher clock.
The big question is the RAM amount and the type. The bandwidth will likely have to be around 1TB/sec so I think HBM might be the only possibility and that will certainly limit size. Let's see how Vega turns out...
Sorry, you are mistaken here. The Macbook is using an Samsung OEM drive. The Macbook SSD and it big brother the 960 PRO can reach write speeds greater than 2.1 GB/s just very briefly, but they cannot sustain it.
The Samsung SSDs already use 3D NAND, so which 3D memory technology are you talking about? Intel Optane? The first Intel Optane SSD will ship end of the year. It's an enterprise product with an enterprise price point. And the whole technology is still in it's infancy.
A write speed of 10.000 MB/s would make little sense as long as you are just using x4 PCIe lanes, since even with PCIe 4.0 that would be over the theoretical bandwidth limit of a PCIe 4.0 x4 connection. So I'm calling Bullshit on this. Besides what would you do with a SSD that can write 10.000 MB/s in a game console? From where would you read data so fast that you require a 10.000 MB/s write speed?
Even with read speeds above the current maximum of 3500 MB/s we are talking about diminishing returns here. Look at any current real life benchmark, this here is a good example. The loading times of the fastest and slowest SSD for Battlefield 3 differ by just 2 seconds, even though the fastest SSD (Samsung 960 PRO) can read data (sequentially) 5 times faster than the slowest (Samsung 850 EVO). How much would a 5000 MB/s SSD be faster? Another .5 seconds?
When designing a console it's all about the biggest bang for the buck and not bragging rights.
D
Deleted member 13524
Guest
I don't believe this..
Even if solid-state chip makers can eventually make them cheaper to produce than HDDs on a price-per-GB ratio, they'll try to maintain a price premium over HDDs for as long as possible.
Right now the price difference per-GB between HDD and SSD is still around 8-to-1. By 2019 it may be reduced to 3-to-1 or 2-to-1 but parity is unlikely IMO.
I agree with ULP Zen but 8 cores / 16 threads max at ~3GHz, also most probably Navi and low-cost HBM. With game engine developers now working on trimming VRAM demands for games because has seemingly gotten out of hand, 16GB may be enough and a lot more cost-effective.
As for mass storage I think all home console makers will want to keep a large HDD with something like 2TB, but they may complement the APU with a PCIe 64~256GB fast storage as cache.
I think a silicon budget of ~300mm^2 on 7nm will be pretty good for next gen.
Probably fit a 8 core zen3 CPU clocked at 3ghz in 60mm^2. Stacked ram and high bandwidth cache should be cheap then. 8GB of HBM3 + 32GB of DDR4/5 would probably be a cheap enough solution, I am not sure how the SoC will talk to both ram but a solution can probably be made then, memory access should still be unified and the physical implementation abstracted. Reason to not go with a physically unified memory will probably come down to price, by the looks of what AMD has shown in Vega demos, that tech should be very powerful for lowering the requirements of extremely high speed memory. I believe that this setup will offer better performance than 32GB of GDDR6 at a lower price and lower power. The solution should be a lot cheaper than a complete stacked memory approach while performance is to be seen, judging by AMD's demos, there is a lot of promise. The GPU will get the majority of die area and will probably be 64 CUs with 1.5ghz clocks minimum, which will be >12TF.
I think flash storage is still not cheap enough for the whole console to run on so best case with be an SSHD.
Probably fit a 8 core zen3 CPU clocked at 3ghz in 60mm^2. Stacked ram and high bandwidth cache should be cheap then. 8GB of HBM3 + 32GB of DDR4/5 would probably be a cheap enough solution, I am not sure how the SoC will talk to both ram but a solution can probably be made then, memory access should still be unified and the physical implementation abstracted. Reason to not go with a physically unified memory will probably come down to price, by the looks of what AMD has shown in Vega demos, that tech should be very powerful for lowering the requirements of extremely high speed memory. I believe that this setup will offer better performance than 32GB of GDDR6 at a lower price and lower power. The solution should be a lot cheaper than a complete stacked memory approach while performance is to be seen, judging by AMD's demos, there is a lot of promise. The GPU will get the majority of die area and will probably be 64 CUs with 1.5ghz clocks minimum, which will be >12TF.
I think flash storage is still not cheap enough for the whole console to run on so best case with be an SSHD.
During PS4 Pro promotion tour, Cerny very clearly said that Sony still fully believes in "generations", but they are now open in providing midgen updates. There will be PS5.
I fucking hope there will be, I kinda even don't want devs creating exclusive games for Pro beause it's hardware is not that epic. It's just a beefier PS4. True PS5 needs to blow us away.
I like this, but I'd rather see them get there sooner.
Inuhanyou
Veteran
My guess for PS5 for a while has been a 10tflop GPU, 16GB HBM in the 500 to 600 GB/s range with an Zen 8 core CPU designed for low power console space.
I don't know how feasible these things are, but my point of reference is holiday 2019 at 399 US dollars.
I know some people who would be looking some sort of crazy 16tflop, 32GB 1TB/s beast would disappointed about its relative increase on Scorpio, but its an extreme leap from base PS4, which i assume Sony bases any hypothetical PS5 design on rather than how far they can get from Pro, which is designed as the PS4 for in between buyers. The main point of differentiation between Pro and PS5 IMO will be exclusive software to take full advantage of the hardware as opposed to what we have with the iterative machines now, which i think makes all the difference honestly.
People said diminishing returns going from PS3 to PS4 before PS4 came out..but going from RDR to Horizon for example, its quite something.
What i do think Sony can take away from the iterative concept of Pro is how exactly to treat legacy software, more specifically how to not do it and how to ape their competitors . Just based on what i've heard over the past few days, MS have done it right in how they allow system wide super sampling, optimizations to allow legacy games to run and look as good as possible without having to wait for a boost patch that may or may not show differences.
I have no interest in Scorpio(or Pro for that matter) but with x86 seemingly being the way forward for these two companies, there is no excuse to not support legacy software at this point in time. I could understand dropping PS3 support, hell i can even understand no PS2 and PS1 support outside of emulation. But PS5 absolutely has to hit it out of the park specifically in regards to PS4 legacy software support, if we are to take anything away from Sony's claims of supporting a consistent ecosystem through their Playstation Network service.
I don't know how feasible these things are, but my point of reference is holiday 2019 at 399 US dollars.
I know some people who would be looking some sort of crazy 16tflop, 32GB 1TB/s beast would disappointed about its relative increase on Scorpio, but its an extreme leap from base PS4, which i assume Sony bases any hypothetical PS5 design on rather than how far they can get from Pro, which is designed as the PS4 for in between buyers. The main point of differentiation between Pro and PS5 IMO will be exclusive software to take full advantage of the hardware as opposed to what we have with the iterative machines now, which i think makes all the difference honestly.
People said diminishing returns going from PS3 to PS4 before PS4 came out..but going from RDR to Horizon for example, its quite something.
What i do think Sony can take away from the iterative concept of Pro is how exactly to treat legacy software, more specifically how to not do it and how to ape their competitors . Just based on what i've heard over the past few days, MS have done it right in how they allow system wide super sampling, optimizations to allow legacy games to run and look as good as possible without having to wait for a boost patch that may or may not show differences.
I have no interest in Scorpio(or Pro for that matter) but with x86 seemingly being the way forward for these two companies, there is no excuse to not support legacy software at this point in time. I could understand dropping PS3 support, hell i can even understand no PS2 and PS1 support outside of emulation. But PS5 absolutely has to hit it out of the park specifically in regards to PS4 legacy software support, if we are to take anything away from Sony's claims of supporting a consistent ecosystem through their Playstation Network service.
Silencerx98
Newcomer
My personal predictions
Release Holiday 2020
7nm TSMC fabrication process
8 cores Zen+ @ 2.8GHz
13.5 TFlops Navi GPU
32GB HBM+, 800GB/s bandwidth
2TB 7200RPM HDD ($399, Standard versions)
1TB SSD ($449, Premium versions)
* + basically refers to low power, mobile variants which may be unnamed for now. I'll probably revise when we get new info on these.
This isn't necessarily strong evidence for predictions, but looking at top end GPU's of a specific year, we can guess what a mid range GPU will look like in 3 years, something which will be equivalent to the next gen console GPU at the time. For instance, the PS4's GPU is equivalent to the GTX 580, which was high end in 2010. The Pro's GPU is equivalent to the GTX 780 Ti, which was high end in 2013. Hence, assuming a Fall 2020 release, the PS5 GPU should easily be at least equivalent to the Vega 10. If we're talking a Fall 2019 release, it could be around Titan X Pascal level.
Release Holiday 2020
7nm TSMC fabrication process
8 cores Zen+ @ 2.8GHz
13.5 TFlops Navi GPU
32GB HBM+, 800GB/s bandwidth
2TB 7200RPM HDD ($399, Standard versions)
1TB SSD ($449, Premium versions)
* + basically refers to low power, mobile variants which may be unnamed for now. I'll probably revise when we get new info on these.
This isn't necessarily strong evidence for predictions, but looking at top end GPU's of a specific year, we can guess what a mid range GPU will look like in 3 years, something which will be equivalent to the next gen console GPU at the time. For instance, the PS4's GPU is equivalent to the GTX 580, which was high end in 2010. The Pro's GPU is equivalent to the GTX 780 Ti, which was high end in 2013. Hence, assuming a Fall 2020 release, the PS5 GPU should easily be at least equivalent to the Vega 10. If we're talking a Fall 2019 release, it could be around Titan X Pascal level.
Last edited:
Transponster
Newcomer
Huh? Vega is gonna be less powerful than GP102.
Silencerx98
Newcomer
Real world performance? Probably, but Vega isn't out yet, so we can't be too sure. At best, I think the Vega 10 will be somewhere between the GTX 1080 Ti and the new Titan Xp (yes, the one launching in a while, full GP102 chip). In Flops performance, though, Vega 10 is supposedly 12.5 TFlops, GTX Titan Xp is 12 TFlops, GTX Titan X Pascal is 11 TFlops. So, I was referring purely to Flop count, though it may not be the best measurement of GPU horsepower.
I know those articles are quite old, but can we trust their timetable?
Risk production of 7nm Zen CPU and Navi GPU in 2018:
http://wccftech.com/amd-vega-gpu-navi-gpu-hbm2-2017-2018/
Low power "netbooks, embedded devices" 10W Zen CPU (4 cores) at 7nm planned for a 2019 release:
http://wccftech.com/amd-gray-hawk-zen-7nm-apu/
I thought Sony would release their PS5 in 2020 but could they aim for a holiday 2019 release using freshly released 7nm tech ?
Risk production of 7nm Zen CPU and Navi GPU in 2018:
http://wccftech.com/amd-vega-gpu-navi-gpu-hbm2-2017-2018/
Low power "netbooks, embedded devices" 10W Zen CPU (4 cores) at 7nm planned for a 2019 release:
http://wccftech.com/amd-gray-hawk-zen-7nm-apu/
I thought Sony would release their PS5 in 2020 but could they aim for a holiday 2019 release using freshly released 7nm tech ?
phoenix_chipset
Regular
Considering the last time one of the big 3 released a truly state of the art console (with chips from the same year) was the xbox 360, i'd say no. It's probably a combination of not having the necessary yields, and cost of course.
If the chips exist in 2019, expect a 2020 release, if not later.
Run! Don't look back! Save yourself!
Oh boy did I just unintentionally opened up a nasty can of worms
? I really gotta tread carefully these days.
I'm sure technically it is possible if everything lands on schedule. But there are other factors for these decisions that we are probably very well skipping.
Like what the feature set will be besides gaming? Is there OS ready for these features? What changes do they need for these features. What media capabilities? Will drivers be ready? BC?
Software development doesn't move as fast as hardware development. other questions to ask might be whether Sony wants to trigger a new generation so soon when PS4 is very profitable. Will they stand to gain more revenue as a result of triggering a new generation? Or will they wait until they see a decline.
That being said, I suspect the earliest we would see next gen is holiday 2021. i know it sounds late, but it would also help to get the bulk of PC onto DX12 and have 4K as a mainstream TV in everyone's home, and not just the massive houses in NA.
Perhaps the cusp between 4K - 8K would be a good launch frame.
Sony uses TSMC, not Global Foundries. Apple is funding a large portion of TSMCs 7nm process so i expect them to be on schedule for 2018 ready for the next iPhone. So 2020 is 2 years later wich should be long enough for 7nm to mature as a process
D
Deleted member 11852
Guest
That's crazily late. Eight years after this gen debuted and 5 years after the Pro was introduced as a "mid generation" update. While plenty is outside of Sony's direct control, they are hardware people and understand hardware issues and risks. If they have to pull the trigger in 2019 at a node/process they didn't intend I believe they will do that rather than wait two more years.
- Status