Next Generation Hardware Speculation with a Technical Spin [2018]

Status
Not open for further replies.
Even on eMMC 5.1, the throughput base is not as high as one would think. Naturally if you ramp up to the largest size currently available, 256GB, you might get faster performance. I'm not certain if the interface scale up to 4x.

"Achieving maximum sequential Read/Write speeds of 330MB/s and 200MB/s (based on 64 GB),"

256GB eMMC 5.1 part: https://www.samsung.com/semiconductor/estorage/emmc/KLMEG8UERM-C041/

This is true, this part is for power saving (mobile) I believe some are more netbook focused which may offer more performance at a higher wattage which would be ok in a console, however the random access to it should be quicker than the current gen laptop drives, and I would think and more able to fill that theoretical pipe. Also if seperase from the HDD it would help reduce is bourdon, and game DVR etc which all have allocated bandwidth atm further reducing the performance we see.

I do not expect crazy fast storage will be selected, something cheap and readily available that offers a meaningful boost to performance.

Edit found one for the automotive industry, others must be available.

https://news.samsung.com/us/samsung...versal-flash-storage-automotive-applications/

Sequential reads for the 256GB eUFS can reach 850 megabytes per second (MB/s), which is at the high end of the current JEDEC UFS 2.1 standard, and random read operations come in at 45,000 IOPS. In addition, a data refresh feature speeds up processing and enables greater system reliability by relocating older data to other less-used cells.
 
What are you talking about? :runaway: Xbox One X already have 12GB of RAM, we speake about next gen consoles what will be released in 2020-2021.
He's talking about realistic costs and what compromises can be made in an intelligent post. Your drama is a bit silly. RAM prices are stupidly high. For comparison, PS4's BOM had $188 for SOC and RAM. RAM in PS3 was $24. DRAM exchange currently has DDR4 at over $7 per GB, making it $112 for 16 GB and $224 for 32 GBs. So realistically, how can these consoles afford 24-32 GB/s RAM in 2020 short of an unpredictable worldwide RAM price crash of epic proportions?

It's silly to use past technical progressions as the only roadmap for future hardware designs, as if every design should be past design x8. Tech changes and the options for a well balanced system change. Which is what Gubbbi pointed out...
 
Don't you know these ram companies were fined 8 billion for price fixing?

I don't think posts are intelligent and rational when they're so utterly sure of the future and saying other speculation is dumb. I think a wager would be fun like we get to pick forum tags if we're right :)
 
The early speculation was marred by the lack of info regarding higher density GDDR5, but we had a pretty good idea of what options were possible on a particular memory bus circa 2011/early 2012. The 4Gbit variety wasn't quite public info until towards the latter half of 2012.

There were also cost considerations to temper expectations (16 chips in clamshell for 256-bit bus for how many years, devkits, was 8Gbit even happening etc.).

Folks probably shouldn't have said "impossible" or "never". Literally.

It's been interesting that GDDR5 has lasted for about a decade now.

----

Anyways, you're not offering much of a discussion point by appealing to a past prediction of an unrelated HW component ( that not everyone even necessarily, definitively asserted) as a means for justifying your expectations for non-volatile storage. :oops:
I'm basically saying peeps are too sure of themselves and acting like their speculation is the only sensible one.
 
So realistically, how can these consoles afford 24-32 GB/s RAM in 2020 short of an unpredictable worldwide RAM price crash of epic proportions?

This willbe in 2-3 years, we don't know what prices will be.

I really think what next gen consoles should launch in 2022, because if all you guys talking about is true, then new consoles will not be next gen. Next gen consoles should be at least 24 Tf and with 64GB of RAM. Only then difference in graphics will be big enough.
 
^ Exactly why even release new consoles with 12gb? They wouldn't. It's like people look at current price and that's as far as they can look.

I'm willing to bet we'll get at least 24gb ram.
 
I'm basically saying peeps are too sure of themselves and acting like their speculation is the only sensible one.
They're presenting logical, reasoned arguments, with data, and refuting the logic of the counter-arguments.

posit - Next gen will have lots more RAM
theory - "Every generation has lots more RAM"
counter-argument - RAM prices have sky-rocketed, options change based on new techs, alternative solutions may present better economies
theory - RAM price references, flash prices references, transfer speed considerations, looking at development like tiled resources
counter-argument -" ya-boo sucks to you, I know what I like. Stop dissing my theories just because you disagree."

o_O

Discuss properly or not at all. If there are good logical arguments in favour of 24+ GB's RAM, present them. So far the only theory as to why more than 16 GBs will be present is that the amount of RAM just seems small, ignoring that it's what's on screen that matters and if 12-16 GB/s doesn't limit that thanks to technology advances elsewhere, it's not a problem. The second 'theory' isn't logical but hopeful - maybe RAM prices will come down to <1/5th their current level. It shouldn't surprise you that people are arguing against that position, and you should either step up your argument with some better logic, or agree to disagree.
 
This willbe in 2-3 years, we don't know what prices will be.
Engineers need to design their console now for release in 2-3 years! If you were an engineer at Sony or MS, would you seriously sit in front of the other engineers and say, "let's put in another $100 of RAM in there because it might be cheaper when we start manufacturing tens of millions of these things."??

Engineers have to look at the current state of the market and technology (including things like HBM) and come up with a competitive design now based on what is available and what developments are plausible for launch and the years beyond. Die shrinks aren't too plausible. Significant flash price drops are. RAM suddenly halving its price is a long-shot. Stacked memory might turn out cost effective.

Get this wrong and you have a PS3 on your hands. Get it right and you have a PS2 or PS4.
 
Engineers need to design their console now for release in 2-3 years! If you were an engineer at Sony or MS, would you seriously sit in front of the other engineers and say, "let's put in another $100 of RAM in there because it might be cheaper when we start manufacturing tens of millions of these things."??

They don't say maybe will be. They know trends and know how prices can change.
 
They're presenting logical, reasoned arguments, with data, and refuting the logic of the counter-arguments.

posit - Next gen will have lots more RAM
theory - "Every generation has lots more RAM"
counter-argument - RAM prices have sky-rocketed, options change based on new techs, alternative solutions may present better economies
theory - RAM price references, flash prices references, transfer speed considerations, looking at development like tiled resources
counter-argument -" ya-boo sucks to you, I know what I like. Stop dissing my theories just because you disagree."

o_O

Discuss properly or not at all. If there are good logical arguments in favour of 24+ GB's RAM, present them. So far the only theory as to why more than 16 GBs will be present is that the amount of RAM just seems small, ignoring that it's what's on screen that matters and if 12-16 GB/s doesn't limit that thanks to technology advances elsewhere, it's not a problem. The second 'theory' isn't logical but hopeful - maybe RAM prices will come down to <1/5th their current level. It shouldn't surprise you that people are arguing against that position, and you should either step up your argument with some better logic, or agree to disagree.

If you're not able to speculate while incorporating future events, you're the one that shouldn't be discussing. Consoles are started on the second the current one is out, if Sony or whoever did what you do they'd be looking at making the ps2 in 1996 and say well damn we'll never be able to afford more than 8mb ram.

And no, faster storage doesn't make up for a lack of memory completely. Certainly not 12gb since you also have to factor in OS allocation so in reality unless Sony and MS suddenly stop wasting 3gb+ on the OS, that'd be a 1 gb jump. 1 gb jump is enough for a generational leap? Intelligent discussion my ass.
 
They don't say maybe will be. They know trends and know how prices can change.
It's more like, "let's plan for this amount of memory but if prices change a ram increase will be simple. And I expect a change." Like exactly what happened with ps4.

I imagine with AMD creating Navi for sony that the gpu is set in stone, but ram can easily change between here and now even for a 2019 release.

So while they do look at current prices and not blindly plan, they have to make intelligent speculation for late changes.
 
He's talking about realistic costs and what compromises can be made in an intelligent post. Your drama is a bit silly. RAM prices are stupidly high. For comparison, PS4's BOM had $188 for SOC and RAM. RAM in PS3 was $24. DRAM exchange currently has DDR4 at over $7 per GB, making it $112 for 16 GB and $224 for 32 GBs. So realistically, how can these consoles afford 24-32 GB/s RAM in 2020 short of an unpredictable worldwide RAM price crash of epic proportions?

It's silly to use past technical progressions as the only roadmap for future hardware designs, as if every design should be past design x8. Tech changes and the options for a well balanced system change. Which is what Gubbbi pointed out...

I won't excuse @Liandry 's unnecessary drama, but most news so far are pointing to RAM prices falling within the next few years. There's the lawsuit in the USA and the probe in China attacking the big three DRAM manufacturers (Samsung, Micron, SK Hynix) for price fixing, and now there's a bunch of chinese fabs making DDR4.

At the very least for the increase in supply through the chinese factories, we should be looking at a downward trend in overall price-per-GB.
Certainly, the engineers developing a console that will release in 2-3 years are taking that into account. Sony didn't design the PS4 with GDDR5's cost-per-GB from 2010 in mind.



OTOH, an equally valid question is "will developers actually need much more than ~16GB in a ~10-15 TFLOPs console"?
Maybe with a 32/64GB NVMe storage acting as cache (unlike DRAM, solid state storage's cost-per-GB has been falling sharply) they won't really need lots of DRAM.
I imagine 64GB of solid state going through a 2x PCIe 4.0 bus (theoretical 4GB/s) and a capable controller could make it possible to forego a bunch of data that nowadays needs to be transferred to the DRAM for the CPU/GPU to work on it. For example, direct streaming of large sequential files like texture data should be possible, and that makes up quite a big chunk of today's VRAM requirements.
 
The thing about 16gb + ssd for cache is... while that might be enough for asset streaming and all the new textures and geometry for next gen - the combination of the likely event that 4gb will be the new amount for OS allocation, and the increase in new game logic that will result from ryzen cpu capabilities - how is 16gb going to be enough? We'd have to be stuck with the same old game design limitations.

I think this kind of speculation only factors in a graphical leap and not one in gameplay design.
 
...Certainly not 12gb since you also have to factor in OS allocation so in reality unless Sony and MS suddenly stop wasting 3gb+ on the OS, that'd be a 1 gb jump. 1 gb jump is enough for a generational leap? Intelligent discussion my ass.

You may want to revisit your math.

I think this kind of speculation only factors in a graphical leap and not one in gameplay design.

This line of reasoning seems to place too much emphasis on RAM while overlooking the components likely to see significant improvement. A proper CPU (Zen), for example, can offer a proper "generational leap" in world simulation.
 
You may want to revisit your math.



This line of reasoning seems to place too much emphasis on RAM while overlooking the components likely to see significant improvement. A proper CPU (Zen), for example, can offer a proper "generational leap" in world simulation.
I noticed what I did but didn't care to change it. Point still stands, Console OS by Sony and MS aren't going to get any less bloated. With 12gb you get jack shit improvement over Xb1 X. Here's our brand new state of the art console with the exact same memory count we had before! Yeah.

What are you going to do with ryzen if you've got the same amount of ram for world simulation as last gen?
 
The thing about 16gb + ssd for cache is... while that might be enough for asset streaming and all the new textures and geometry for next gen - the combination of the likely event that 4gb will be the new amount for OS allocation, and the increase in new game logic that will result from ryzen cpu capabilities - how is 16gb going to be enough? We'd have to be stuck with the same old game design limitations.

I think this kind of speculation only factors in a graphical leap and not one in gameplay design.

You seem to forego the fact that streaming textures and geometry directly from a SSD cache would free a sizeable chunk of the DRAM currently being used by developers. Whereas in the past devs would have e.g. 1GB available for framebuffer + graphics/compute shaders + shadowmaps, and other non-texture/geometry data, they would now have 8-10GB for the same kind of data, and the texture+geometry storage would be upgraded to 30+ GB.


And why would the OSes' DRAM demands rise to 4GB?
Honest question, what new OS-driven capabilities are in the pipeline to be implemented and require a lot more DRAM?
The only upgrade I can see is framebuffer recording on video going up to 4K, but that would most probably be encoded into H265, so the actual memory requirements would hardly go up compared to the mid-gens' current 1080p.
 
You seem to forego the fact that streaming textures and geometry directly from a SSD cache would free a sizeable chunk of the DRAM currently being used by developers. Whereas in the past devs would have e.g. 1GB available for framebuffer + graphics/compute shaders + shadowmaps, and other non-texture/geometry data, they would now have 8-10GB for the same kind of data, and the texture+geometry storage would be upgraded to 30+ GB.


And why would the OSes' DRAM demands rise to 4GB?
Honest question, what new OS-driven capabilities are in the pipeline to be implemented and require a lot more DRAM?
The only upgrade I can see is framebuffer recording on video going up to 4K, but that would most probably be encoded into H265, so the actual memory requirements would hardly go up compared to the mid-gens' current 1080p.
I don't think faster storage would do anything except allow for graphical improvements. Even NVME is a fraction of a fraction the speed of DDR4, and unless we get a hybrid storage solution with a tiny nvme we won't be seeing that anyways. You still have to store the entire scene in memory, the storage would just allow faster switching of assets. You can't process data directly from storage. You can get clever with more constrained games like Uncharted, but if you want much bigger scenes with more things going on it's not going to help much. It's not a replacement for ram in that respect.

Xb1X was going to use 4gb for its os initially, but they decided to keep the dashboard and everything at a native 1080p. I can't see Next gen not having everything display at Native res, + they'll surely have faster web browsers and and even heavier focus on streaming. It's a good bet they'll be using 4gb, if not more! I'm still shocked Sony and MS couldn't use anything less than 3gb this gen it's just terrible.

Basically i'll eat my boots if Ps5 or Scarlett has 12gb ram. 16 for Ps5 in 2019... maybe. I wouldn't bet on it, I still think we'll see 24. For 16 they'd have to get really clever with storage and get OS allocation down.
 
Last edited:
Interesting discussion on footprint.

If we are Cost restricted on sizing, then the only other sensible option is too look further into compression/or other ways to remove assets out of the installation.

Where are we on those fronts? Any expected changes with regards to compression or just entire removals of redundant asset creation ?

ie: could features like tesselation reduce the need for high resolution textures?
Or
If we move you GPU side dispatch we can remove the need to batch and instance reducing redundant assets?
 
Last edited:
Status
Not open for further replies.
Back
Top