Sony's ReRAM plans - what can and can't ReRAM bring to a console? *spawn

The difference between 12.8 GB/S and 25.6 GB/s wont matter because NextGen wont have more than 16GB of ram available for games to use and the difference between loading your entire memory in 1.25 seconds compared to 0.625 seconds is imperceptible by consumers.

And NextGen games will be streaming in memory and not having to swap out all 16GB at once other than initial load.

 

The 25.6GB/s probably has more to do with "8 chips and PCIe Gen5 x8" (it's right there in the slide) than with ReRAM itself...

 

I couldn't say. It depends on the performance they can get from the parts. The only sure thing is additional complexity is always bad. One big pool of fast storage is better than one large pool of okay-ish storage and one smaller pool of super-fast storage. Every tier of memory access is tolerated because it bring improvements. If it were possible, we'd go straight from CPU to storage without any L1, L2, RAM, IO buffer, storage, but that's not possible, so we break down accesses to increasing sizes of decreasing speed. Consoles can stream effectively from fast enough storage for live content. Everything else is run in RAM. If you want to sit a super-low-latency access storage between RAM and an SSD, what problem would it be there to solve? Why isn't prefetching and caching in RAM a better option?

Tying to force the market through PS5 would be as daft as trying to force Cell's success through PS3. If the market wants ReRAM, it'll want it regardless whether PS5 has it or not and will pay for it.

Arguing this from the other direction, you've pointed out a potentially huge market for ReRAM. What does ReRAM in PS5 bring to Sony's plans to have ReRAM in imaging sensors and computer storage? If this market is huge, PS5 is just a drop in the ocean. If it's not and it's reliant on PS5 to justify the investment, it's an expensive proprietary tech.

 

Why do you want that? I can't see another reason then just because it is ReRam, it won't make much of a difference to you as a end-user other then it being more expensive and possibly more risk and losses to Sony.

Edit:

I want a new Cell or even Emotion Engine to happen, cause it's so cool and exotic, even though it wouldn't really make the games better looking directly, maybe even worse in that case

Btw, if you really want something as fast as ReRam, or even faster, get a Intel Optane, expensive but there you got it.

 

Isn't there scope for ReRAM to be implemented as more of an accessory? It negates the risk associated with having to include it in every console, whilst opening up its benefits for those willing to pay.

For example, let's say the PS5 ships with 16-24GB of GDDR6, and has an NVME SSD running at ~6GB/s. Straightforward components, with predictable prices in the years to come, and guaranteed stability.

Could that same hypothetical PS5 then contain a place to slot a souped-up memory card? 256GB of portable storage with nearly 10x higher performance than the default SSD would likely appeal to a decent percentage of PS5 owners, especially when we consider the ~$200 Xbox Elite controllers.

Sony could sort of have their cake and eat it: they won't be reliant on this new, uncertain tech for penetration of the mass market, but they could utilise a percentage of their tens of millions strong PS5 install base to help better achieve economy of scale.

And if it looks to get cheap enough over the course of the next 3-4 years, they then have a great candidate for main storage or default cache in their mid-gen console.

 

Rather than put in a slot for an as-yet-non-existent upgrade with added cost for all buyers, instead save ReRAM for PS5Pro or a slim model or something, when the ramped up production can produce it reliably enough and cheap enough for a console.

 

Sata SSDs capped at 500MB/s, nvme are 3GB/s right now (and 6GB/s soon) using the same nand chips, for the same price per GB. Nand isn't limiting raw bandwidth, the interface is.

If the games are designed for 32KB or 64KB fetch size, they can reach a similar raw bandwidth as ReRAM using multiple nand chips, and use however many pcie lanes to get that bandwidth. For a much much lower cost.

Compared to HDDs, nand have a lower fixed cost per device, and the cost per GB is dominant. A pair of 512GB is twice the bandwidth as a single 1TB, costing not much more if the consoles use direct IO with a flash-aware file system, instead of expensive controllers doing the caching and wear levelling.

If it's a standard nvme, it will probably be 4 lanes, but if it's custom I wouldn't be surprised to see a 8x pcie.

 

How much cost would such a slot incur though? A couple of dollars per console to include the bus on the PCB?

It seems like the kind of meagre loss that's worth taking when the benefits are potentially substantial.

I take the opposite position to you: if ReRAM really isn't working out when it comes time to create a slim, pro, or just a new revision (i.e. at the same specs, on the same node,) that's when it's worth taking out to further reduce costs.

 

Am I right in thinking that approach would require the NAND to be soldered? So no hot-swappable storage?

And in light of what you're saying, I need to have a look into the interface being proposed for ReRAM that allows its commercial release to reach the touted ~51GB/s.

 

$2 * 20 million first year sales = $40 Million Lost Profit.

Could you front me that meager amount?

 

Across 100 million consoles, that's $200 million additional costs. What does that $200 million get Sony? If they can sell an upgrade to 10% of users at over $20 profit a piece, it's a net win. If not, it's a waste of money.

In a mid-gen refresh or update, it'd be a good USP to differentiate and encourage upgraders.

 

It could be soldered, or a pcie connector, or nvme format. There is a controller in any case, but it can be a very simple direct IO instead of the complex controller we see today. That controller doesn't have to be standard, or 4x, but anything custom means no user expansion unless that storage is only used for caching recently played games, while allowing any storage speed for expansion.

Non volatile ram formats already have a jedec standard for direct addressing dimms. The whole point of NVRAM is that the data is being accesses as granularly as ram, unlike flash which needs up to 256KB full page writes (only usable for storage).
https://en.m.wikipedia.org/wiki/NVDIMM

 

This is something that people that look at news and PR announcements about new technology often miss.

It doesn't matter how good the technology is, how much faster it is than existing technology, or how applicable it can be for certain market segments.

Everything needs to fit into the scope of profitability in order for it to come to market AND succeed. Most new technology is announced with the idea that a lot of work still needs to be done to make it cost effective to produce. In many cases, the researchers can never find a way to manufacture it cost effectively.

Some examples.​

• OLED took a long LONG time to get to the point where it could be cost effective to produce for the mass consumer market.​
  • Even longer for manufacture in sizes large enough for Televisions.​
• MicroLED TVs are still a huge unknown whether they'll ever reach the point where they can be manufactured cheaply enough for the mass consumer market.​
• HAMR for HDDs has been delayed multiple times for multiple years as researchers struggle to make them cost effective to produce (this includes making them reliable enough that they aren't constantly failing).​
• Optical displays that user lasers to beam the image directly into a person's eyes has been researched since the 90's, announced multiple times, and still isn't something that is on the roadmap for consumer adoption and quite possibly never will be.​

Memory and Storage technologies litter the virtual landscape of announcements.

Some of them have managed to find limited exposure in markets that can support their high cost of manufacture (Optane, for instance). Some have fallen by the wayside (Rambus for the most part). Other's have never made it the market except in very limited and very specialized cases (Hybrid Memory Cube, for instance).

Heck, back in the 80's while at University, exciting developments in computing and storage were announced including things such as...

• Organic CPUs
• Holographic memory
• Crystalline storage
• And so many other things which you never hear about anymore.

ReRAM is something that will at the very least see some form of commercial release. Whether it remains forever limited to specialized markets which can bear the costs or whether/when it'll become cheap enough to be used in the mass consumer market is a key question.

Will it get to that point in time for the PS5? Extremely unlikely, IMO. Hell, I still wouldn't place bets on it ever becoming a mass market technology, even though Sony's talk implies that it'll get there. But that means almost nothing when it comes to new and relatively unproven technology.

For instance, Seagate has for many years made announcements that HDDs featuring HAMR platters will be coming to market in X year.

Announcements and talks regarding unreleased products should always be viewed as the Corporation announcing them being hopeful it'll be ready within the timeline they've publicly announced, but that there is little to no guarantees that it will actually be available within that timeline, as oftentimes the announcements are made while R&D is still ongoing WRT to making them cost effectively manufactured (cost, reliability, etc.).

In that light, I would be especially skeptical of any claims that it would be included in the PS5. The PS5 is one of the key products for Sony to remain solvent as a corporation. How much are they willing to bet that ReRAM will be mature and cheap enough to manufacture for the PS5 by the time the PS5 needs to be launched?

Regards,
SB

 

Indeed. These are PR remarks to show how great the company's R&D is, pushing boundaries, and to set the ground work if it ever pans out. No company is going to go on record saying, "we've got this great tech in the lab, and we want to develop it, but it'll probably crash and burn like 80% of tech ideas do." They either don't talk about it, or talk about it in the most positive light.

Those willing to take such announcements at face value probably just haven't been around long enough to know the real pattern these things follow.

 

I want them to include ReRam so I can cannibalize new consoles for profit. My soldering iron is around here somewhere.

 

If they do a custom ssd they’ll have to offer multiple skus to offer large sizes for people with a lot of games. Better option is standard pcie4/5 nvme drives so users can upgrade on their own to the size of their choosing. Nvme will offer a massive improvement over current console drive performance. No reason to go with something exotic at the expense of a single sku that can be upgraded

 

Cold boot. Linux. 3 seconds. NVME.

when you program to take advantage of the hardware you will get results.

a lot of work here though to make Linux in rust. Make your own drivers etc. But this enables the creator to maximize the hardware.

 

Not when you have 16 GB or less of memory for games in the consoles.