• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Official HBM2 memory standard posted.

Renekton

Member
2019 seems optimistic?

2017 - Intel 10nm Cannonlake
2018 - TSMC 10nm for small die SoCs
2019 - TSMC/GF 10nm for large die GPU
2020 - 10nm mature enough for console SoC?

Head start vs horsepower is up for debate if you consider Dreamcast and Wii U.
 
That's why i intentionally aimed down to midrange hardware. I'm going on the assumption that Sony want to repeat the PS4's success and go for functionality and efficiency both power/wattage and cost wise above just putting powerful components in the box.

8tflops right now is far beyond what is necessary to run something like Infiltrator in an actual game in real time(which ran on a 680 mind you in the tech demo version)

I think that is still quite a jump and largely satisfactory for GPU power.

Also, not every game is gonna be 4k compatible, i don't even think many will be outside of smaller less demanding titles or PS4 remasters. We have to remember that we are piling on next gen rendering in addition to 4 times the pixels unnecessarily, which is probably going to eat up a lot of the GPU time alone just rendering at that res. I'd take 1080p with great IQ again and far better visuals myself.

But the ability to scale everything to 4k in the menus is a must.

What you say about 4-5 power increase also is puzzling to me. Are you not taking into account the HBM? The bandwidth and RAM alone will give a huge boost, and the CPU, if they do get Zen working how we'd hope, will be a beast monster compared to Jaguar.

There's far more to the power of the overall console than just the GPU, as we are abundantly made clear this gen

Yup, its a good reasonable prediction, because its like the PS4. But that's also the problem.

Yes, Infiltrator looks nice and ran on a 680. Same with Agni. 8TF is a lot of power... for 1080p. I'm thinking about Witch Chapter 0 stuff, and that was rendered at 4k. The Titans were there just for 4k and the ram reqs. Power wise, that could probably run on a 680 as well... but 4k will be "the push"...

I'm operating under the assumption that "the push" next gen will be 4k. 48" - 55" inch tvs were already $700 - $1000 this past BF. Name brands like Samsung, Sony and LG. 4k G-sync monitors are getting a huge push this year, gaming + HDR media focused ones will be 4k60fps at minimum. In 3 years, 4k will be the reason to upgrade your tv and your console. Like how 1080p is now, and what "HD" in general was for 360/PS3. Waiting until 2023/2024 for console games at 4k would be absolutely terrible.

4x is the GPU. 5x is with the CPU and RAM on top. They're the bow/the cherry: the cou de gras for the whole package that brings the whole thing together. Meanwhile, the GPU is doing the heavy lifting. The PS4 GPU is 1.7+TF, the CPU is .1+. Different tools for different tasks and all, but there's a reason for this.

All that bandwidth is nice, but will remain underutilized without the power to back it up. Kinda like how the Fury X chip itself is bottle necking HBM tech right now. All that bandwidth, not enough juice.

Also, Zen is supposed to be like the i7 2xxx/ i7 3xxx, right? We've had that forever in the PC space. Better CPU will mean capability for bigger, faster, all at the same time... potentially...or it'll mean that we'll actually be able to run the stuff we already have without being bottlenecked by various crap. A computationally complex game like Star Citizen could run on PS5/XB2 thanks to a not bad CPU and more RAM. Star Citizen already runs on i5/i7s, in rigs that have as low as GTX 660s in them, the graphics just run at lower settings. CPUs are certainly important, but also the most stagnant and passe part of PC game complexity. I agree, RAM capacity/badwidth is worth the hype because we can push higher res, higher quality assets, and faster and more detailed graphics rendering in general with HBM.
 
Also, regarding claims of power savings and performance increases, everyone exaggerates or selectively picks data. If it was 40% more powerful than Fury X while using 40% of the power (60% reduction as you say) that would be the largest increase in perf / watt since probably the 1990s. We will see what it winds up actually being.

As you could imagine, the person you quoted severely butchered what AMD is saying. It's 60% power reduction at the same performance. And there's little to no chance of AMD seeing this big of a leap in performance again by the time the next gen consoles come out.
 
2016 tech in a next-gen console, I hope not for the sake of progress.

If Inuhanyou is onto something, which I think he is, sub Fury X is the graphics power level were looking at. 2015 PC Enthusiast end power in 2018/2019. I'd be more concerned with that, rather than architectural stuff.
 
As you could imagine, the person you quoted severely butchered what AMD is saying. It's 60% power reduction at the same performance. And there's little to no chance of AMD seeing this big of a leap in performance again by the time the next gen consoles come out.

Considering the cards compared on the slides say 86w vs 140w, they're probably comparing to the 960-970 in the Nvidia range, meaning we're talking essentially mid range GPU (in terms of power draw) from previous gen vs low-mid range AMD upcoming gen. Probably an example selected to be most favorable to them, as high end cards are by no means guaranteed to scale that efficiency fully up to the 200-250w realm.

Graphics card hype is always pretty funny tho. Like a guessing game where we try to work out how they're trying to bamboozle us this time.

P.S. 84w is 60% of 140w, but you reduce 140 by 40% to get 86. So it's a 40% reduction in power draw.
 

Reallink

Member
The 980ti is like 7TFlops (the specs ITT called for 10) and sips 250w alone. Maxwell is a pretty power efficient architecture. The timeframe being discussed is 2 years away, the timing is wrong for a GPU that powerful at an affordable price.

To be fair the 980Ti is on the horribly outdated 28nm process. Most are expecting 2x efficiency with Pascal, so you will all but certainly see a sub-150W 980Ti equivalent part.
 
If Inuhanyou is onto something, which I think he is, sub Fury X is the graphics power level were looking at. 2015 PC Enthusiast end power in 2018/2019. I'd be more concerned with that, rather than architectural stuff.

The PS4, the more powerful of the two similar APUs, had a GPU equivalent to roughly a GTX 660-670. This was the mid range chip that came out like 18 months before the PS4. The 680 was more powerful, the 780 was way more powerful, the Titan was way more powerful and also had 6GB of GDDR5 memory.

The PS5 matching 2015 top of the line hardware (in terms of power, not necessarily architecture) in 2018 would be broadly similar to the PS4 matching top of the line hardware from the 500 generation - which would not be tooo far from the same ballpark, excepting, again, architectural improvements and exra VRAM. There are other wrinkles though, there is now a super high cost "prosumer" tier of video cards whereas in the 500-600 generations there was not.
 
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

This is pretty close to what I would guess, especially since the generation should be closer to a normal generation length. The more profitable console launches mean Sony and MS don't need to extend the generation as long for the sake of profits. Also, the standardized hardware should mean that BC should be simple which should make the software transition easier. Because of these factors, I agree on a 2018-2019 launch range. This seems reasonable for 2018, or a bit higher if it's 2019. 32GB ram and a 10TF GPU is the highest I'd consider possible by 2019 in a reasonable console.
 
The PS4, the more powerful of the two similar APUs, had a GPU equivalent to roughly a GTX 660-670. This was the mid range chip that came out like 18 months before the PS4. The 680 was more powerful, the 780 was way more powerful, the Titan was way more powerful and also had 6GB of GDDR5 memory.

The PS5 matching 2015 top of the line hardware (in terms of power, not necessarily architecture) in 2018 would be broadly similar to the PS4 matching top of the line hardware from the 500 generation - which would not be tooo far from the same ballpark, excepting, again, architectural improvements and exra VRAM. There are other wrinkles though, there is now a super high cost "prosumer" tier of video cards whereas in the 500-600 generations there was not.

Mmmm, a GTX 580 or so. So architectural bumps are that important, I guess. I'd would've used AMD cards for a more apt comparison, PS4 is basically a hypothetical 7850+. Sounds like were on a similar wavelink, anywho. Prosumer stuff does throw a wrench in that though. I still think that 4k will be "the thing", and Fury X (or lower) + architectural improvements does not bode well for that power ballpark. That screams 4k20-30fps, bad AF, and post AA/no AA to me. Like a repeat of this gen at a higher resolution, unless devs start designing console games again and not PC downports (read: more Tearaway, less Ubioft/EA AAA).
 
Mmmm, a GTX 580/+. So architectural bumps are that important, I guess. Sounds like were on a similar wavelink, anywho. I still think that 4k will be "the thing", and Fury X + architectural improvements does not bode well for that power ballpark. That screams 4k30fps with frame drops, bad AF, and post AA/no AA to me. Like a repeat of this gen at a higher resolution, unless devs start designing console games again and not PC doownports (read: more Tearaway, less Ubioft/EA AAA).

4k 30fps is very doable on Fury X. 4k 60hz is problematic for high end games.

1080p blows up very well to 4k (pixel doubling) so I could see a dichotomy between 4k 30hz and 1080p 60hz with some better AA depending on the developer. I don't think Sony or anyone else will be mandating 4k although it may indeed be an attraction. But yeah it would be an issue if the draw was "It's graphically moderately better than PS4, but with great image quality!" and you would want more than that if you were pushing 4k in a big way.
 
Considering the cards compared on the slides say 86w vs 140w, they're probably comparing to the 960-970 in the Nvidia range, meaning we're talking essentially mid range GPU (in terms of power draw) from previous gen vs low-mid range AMD upcoming gen. Probably an example selected to be most favorable to them, as high end cards are by no means guaranteed to scale that efficiency fully up to the 200-250w realm.

It says at the bottom of the image that it's a 950.
 

Pachinko

Member
I like the cut of ThoseDeafMutes jib.

New memory architecture that's faster and more power efficient is all well and good but to whoever mentioned the 64 GB number - what on earth would you need that much memory for exactly in a game console. These aren't meant for rendering the next Pixar film , they just have to play games that look prettier than the current ones and at best - 4K might show up in gaming by 2019 running at 30 fps with a special effect feature set on part with direct x 12 or whatever the update for DX is after 12 (maybe they go to 13, maybe they just add a decimal , who knows?)

Whether you're sony/MS or Nintendo - the end game here is to make the best console you can for whatever the launch year is within the confines of a target price point - a price that should be profitable or very near profitable right out of the gate to avoid relying on software sales as much early on.

The 20-30$ it might cost to go from 24/32GB of HBM2 ram up to something truly absurd like 64GB (and honestly, that sounds more a couple hundred dollars to me) would be much better spent putting in a large Solid State hard drive. Right now the biggest bottle neck on the current crop of systems is both a lack of storage space and the lack of high speed access within existing storage.

It's already been mentioned but after a switch up in terms of storage space, the CPU needs to be massively upgraded, it's a real sore point in the PS4/XB1 but it's also an efficient CPU , so toss in the 2018/19 version of that same chip (I don't know what that would entail, perhaps 50% higher clockspeed with 12 cores and each core is also 25-50% better at what it does so it's like 2-3 times better overall ? ) couple that CPU with 4 GB of DDR4 system memory that would be reserved for the OS and everything outside of videogames and then throw in a 16GB HBM2 memory based GPU.

Basically, buy a PC for about 1000$ next September and that's roughly what you'll be paying 400$ for in 2019 in a PS5. Any downward fluctuations in price should be used to expand that SSD drive , right now I'm thinking 512GB is about the max and that's assuming prices fall 75% on such a drive by the end of 2018. Of course a regular old HDD is still an option - throw in a 2TB one for 30$ maybe give it an 8GB SSD cache ? Who knows.

As for what this all means for PC use - well, I don't think it's that big a deal to want to upgrade a PC now , it does mean that the next group of videocards will probably be substantially better than the current crop , they will also cost a lot more.

If anything, hopefully this means I can buy another 280 or a 380 for pretty cheap. Course, for financial reasons and "canadian dollar is ass" reasons I'm not in a big rush to upgrade , I'd love to replace my mother board and CPU because I'm still rocking an i7 860 but it plays everything I throw at it quite well soo....
 

Grief.exe

Member
The 20-30$ it might cost to go from 24/32GB of HBM2 ram up to something truly absurd like 64GB (and honestly, that sounds more a couple hundred dollars to me) would be much better spent putting in a large Solid State hard drive. Right now the biggest bottle neck on the current crop of systems is both a lack of storage space and the lack of high speed access within existing storage.

It's a huge quality of life issue, especially when coupled with slow CPUs, but an SSD will only add to the cost but not increase the actual performance overhead.
 
4k 30fps is very doable on Fury X. 4k 60hz is problematic for high end games.

I'm kinda referring to the closed boxed curse. Like how i3 + 750ti performs better in multiplats, even without console optimization. The only game that doesn't apply to is SW:BF and only because of per map optimizations on consoles. Sure, a Fury X + and i5/i7 does great at 30fps, but what about a hypothetical Zen + PS5 gpu in that ballpark? 'PS4 version of TW3' kinda shenanigans, basically.

1080p blows up very well to 4k (pixel doubling) so I could see a dichotomy between 4k 30hz and 1080p 60hz with some better AA depending on the developer. I don't think Sony or anyone else will be mandating 4k although it may indeed be an attraction. But yeah it would be an issue if the draw was "It's graphically moderately better than PS4, but with great image quality!" and you would want more than that if you were pushing 4k in a big way.

Being stuck at 1080p until 2023/2024 for consoles...ew. 1080p: the zombie resolution :p I'd take the graphics hit in everything except texture res, AF and post/mulstisampling AA for native 4k resolution.
And 60fps.


I'd have to agree on an improvement in storage for next gen. Loading times are really long now, and with HBM and the overall increase in asset quality/size coming, along with the prevalence of DD, we need a a better/faster storage solution. This gen, we should've had a hybrid solid state/HDD solution at the minimum tbh. SATA II was also a super cheap-ass move.
 

Finaika

Member
I'd have to agree on an improvement in storage for next gen. Loading times are really long now, and with HBM and the overall increase in asset quality/size coming, along with the prevalence of DD, we need a a better/faster storage solution. This gen, we should've had a hybrid solid state/HDD solution at the minimum tbh. SATA II was also super cheap-ass move.

500GB SSD should be a minimum for next gen.
 
500GB SSD should be a minimum for next gen.

500gb is not enough storage space for games as large as they are this gen. You'd need 750-1000gb if they get bigger again (more and more games closer to the 50gb mark than the 25) to have the same relative space you got from the PS4/XBO having 500gb HDDs. And in an ideal world you'd want like 2TB of space.

SSHDs seem like a fine compromise - you could have a 1TB SSHD standard and 2TB in premium models.
 
LOL, no way are we going to see that happen unless you want the consoles to possibly be more expensive than $400.

16 GB's is what we can expect at the very least.

I've seen this before.... Hmmmmm Where have I seen this? Oh, that's right, in every "next gen specs speculations"
I remember "PFFF 8gb? That's impossible, unless you want the PS4 to be another flop at 599USD. We'll get a 2gb PS4, that's reasonable" Look where we are now.
 
I've seen this before.... Hmmmmm Where have I seen this? Oh, that's right, in every "next gen specs speculations"
I remember "PFFF 8gb? That's impossible, unless you want the PS4 to be another flop at 599USD. We'll get a 2gb PS4, that's reasonable" Look where we are now.

As ThoseDeafMutes said earlier, we barely got 8 GB's of GDDR5 RAM inside of the PS4. If the prices of 8 GB's of that RAM weren't lowered at the last minute, we would've been stuck with only 4 GB's inside of the system.
 
As ThoseDeafMutes said earlier, we barely got 8 GB's of GDDR5 RAM inside of the PS4. If the prices of 8 GB's of that RAM weren't lowered at that last minute, we would've been stuck with only 4 GB's inside of the system.
Give me that magic ball of yours that predicts the price of memory in a few years, please.
 

McHuj

Member
LOL, no way are we going to see that happen unless you want the consoles to possibly be more expensive than $400.

16 GB's is what we can expect at the very least.

I dunno. I think a split memory pool maybe on the table for next gen. 8-16GBHBM plus 32+ GB RAM of DDR3/4. Effectively consoles end up looking even more PC-like.
 

TheFatOne

Member
Forgive me if someone already answered this, but roughly when are the new gpus for NVIDIA and AMD supposed to come out?
 
Next gen RAM was mostly at least 8x current gen RAM.

Xbox 64 MB
Xbox 360 512 MB
Xbox One 8 GB

PS1 2 MB + 1 MB
PS2 32 MB + 4 MB
PS3 512 MB
PS4 8 GB

Nintendo 64 4 MB
Gamecube 40 MB
Wii 88 MB
WiiU 2 GB

Going by that we can easily expect 64 GB Ram in next gen consoles.

PC with 32gbs is over kill today and be over kill in 5 years, next gen system will have no more 16gbs-24gbs
 
Is HBM a clear improvement for discrete CPUs? Is there any Intel or AMD roadmap indicating any use of HBM for discrete (=not APUs) desktop CPUs?

HMB is a clear improvement for almost everythings in terms of performance, performance per watt and reducing size. What it is not is expandable, so as long as people want to be able to upgrade their RAM capacity HBM will have trouble gaining a foothold in discreet applications.

That could change with the advent of stuff like Intel's 3D XPoint tech which bridges the gap between RAM and storage. You don't need huge amounts of RAM if your hard drive is suddenly almost as fast as DDR3, opening the door to a smaller pool of much faster HBM for main memory.
 

Tagyhag

Member
I dunno. I think a split memory pool maybe on the table for next gen. 8-16GBHBM plus 32+ GB RAM of DDR3/4. Effectively consoles end up looking even more PC-like.

I think people are kind of going overboard.

Remember, aside from the 4GB, people thought the PS4 was going to be as strong as a high-end 2014 PC.

People are going to be overshooting it again.
 

mrklaw

MrArseFace
MS have most to gain from short shifting this generation and hoping Sony holds fire because the PS4 is doing well (not sure they would, PS4 would be sellit to a different audience than PS5 consumers by then, and cross gen will be fully in effect to mitigate costs for publishers, so PS4 could still do a 10 year cycle)

But the critical thing will be the cost vs performance. HBM sounds great, but for a console I think you'd prefer to wait for processes to mature, yields to go up, and costs to come down. If we aren't even expecting big HBM chips from nvidia/AMD until 2H16 as they'll want to sort out process yields on smaller dies, I think 2018 could be challenging for a large die APU at a reasonable cost. Maybe MS will be willing to take the hit to get that headstart?

One big benefit is that edram goes away, allowing MS to use the entire die for CPU/GPU. You can still go split ram if you want more space (and if HBM is too expensive to slap 16-32GB on there). I could see a 16GB HBM APU and maybe another 16GB DDR3 for OS/cache/other. I'd like a total of 32GB ideally - that'd be 8x what the expected ram of PS4 was going to be.
 

mrklaw

MrArseFace
The PS4, the more powerful of the two similar APUs, had a GPU equivalent to roughly a GTX 660-670. This was the mid range chip that came out like 18 months before the PS4. The 680 was more powerful, the 780 was way more powerful, the Titan was way more powerful and also had 6GB of GDDR5 memory.

The PS5 matching 2015 top of the line hardware (in terms of power, not necessarily architecture) in 2018 would be broadly similar to the PS4 matching top of the line hardware from the 500 generation - which would not be tooo far from the same ballpark, excepting, again, architectural improvements and exra VRAM. There are other wrinkles though, there is now a super high cost "prosumer" tier of video cards whereas in the 500-600 generations there was not.

I think that'd be pretty good - 980/980ti would give you games at 4K/30 (30fps never going away on consoles), which would fit the likely 4K focus for Sony at least. Although 4K adoption will be still very low throughout the generation, that might mean 1080p/60 with AA for the rest of us plebs which would be good.

And a 980ti would probably be enough to drive a PSVR2 with a 4K screen and foveated rendering?
 

hesido

Member
...
But if PS5 is coming out in 2018 as part of a 5 year cycle, we're not going to wind up with the same kind of jump in power.
....

The PS4 has too much wind its sails for Sony to release a PS5 in 2018, and they are releasing PSVR for the platform, they'd want not want take away the spotlight from their PS4 as their flagship product.
 

Finaika

Member
I think that'd be pretty good - 980/980ti would give you games at 4K/30 (30fps never going away on consoles), which would fit the likely 4K focus for Sony at least. Although 4K adoption will be still very low throughout the generation, that might mean 1080p/60 with AA for the rest of us plebs which would be good.

And a 980ti would probably be enough to drive a PSVR2 with a 4K screen and foveated rendering?

The 980Ti alone costs twice as much as the PS4.
 
The PS4 has too much wind its sails for Sony to release a PS5 in 2018, and they are releasing PSVR for the platform, they'd want not want take away the spotlight from their PS4 as their flagship product.

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.
 

Urthor

Member
Sony will have nothing to do with dictating the next cycle, idk why anyone would think that.


Sony will want the cycle to continue for as long as it can with the dominance it has. It'll want to roll with the market share it has and progressively build as large an install base as possible until it is forced to move, same MO as the PS2, which was forced to shift by the 360/Wii and only grudgingly. It'll have its hardware division start working on a PS5 on schedule no doubt, but they'll match their development cycle to Microsoft, and they'll know exactly when Microsoft starts working on it because Microsoft has to talk to hardware partners etc.

And we don't even have to mention Sony and 4k content, do we.

2018 is almost crazily optimistic, hell we have this gen console releases like Final Fantasy 7 remake realistically sketched out for a Christmas 2018 release date, and people are expecting a Christmas 2018 reboot?

Both console manufacturers will want to go until 2019 with harvest years now that they've got some sort of installation base up and going with the new consoles. The 5 year cycle talk is insane, that means that they'd have run at a loss for 18 months to two years, and only had 3 profitable years. Even Microsoft, with the biggest incentive to reset, is going to want more harvest years than that. The Xbox division will have to break even on its losses at a minimum.

Christmas 2020 is what I'd regard as a hopeful date for the reset, 2021 is what I'd regard as on schedule between profitability and Microsoft's need to change the market share paradigm. 2022 plus is entirely possible though depending on Microsoft's internal revenue demands.
 
Both console manufacturers will want to go until 2019 with harvest years now that they've got some sort of installation base up and going with the new consoles. The 5 year cycle talk is insane, that means that they'd have run at a loss for 18 months to two years, and only had 3 profitable years.

What loss for 18 months or two years? This generation they did not ship subsidised monstrosities, they were somewhere close to breaking even on hardware sales from the start. Do you have fiscal statements that the divisions were unprofitable for 2 years after console launch?
 

Urthor

Member
What loss for 18 months or two years? This generation they did not ship subsidised monstrosities, they were somewhere close to breaking even on hardware sales from the start. Do you have fiscal statements that the divisions were unprofitable for 2 years after console launch?

Impossible to predict I'm fairly sure because we can't split the 360/PS3 carry over revenue from the PS4/Xbone revenue that well. Or at least I can't.

But okay, it's fair to say they weren't as subsidised this time around, but they still spent those two years building a market from an install base of zero up to one that is still not incredibly large in the scheme of things, they won't want to go through that pain again that quickly when they'd be forgoing a lot of potential revenue by holding on with a mature console base. Even if Microsoft is significant fraction of what Sony's is, that's still a good amount of revenue they're bringing in, so it's a question of whether their goals are maximising earnings for the bottom line, or progressively competing Sony out of the business by squeezing them, which is unlikely to happen.

When the consoles were still growing with single digit install bases, they were effectively earning not much, even if it wasn't a loss. 2014 calendar vs 2018 calendar is going to be a biblical difference, and Microsoft even with half the install base is still going to have fairly profitable years as they get greater numbers, because their raw numbers are still perfectly good as standalone figures, just not as good as Sony's.

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.
 
Top Bottom