Next Generation Hardware Speculation with a Technical Spin [pre E3 2019]

How many GBs of game assets, minus the video and audio, do you expect next-gen games would be? I assume the video and audio parts of a game can remain on the slower storage.

With the idea you have about how Sony will approach SSD solution, do you think it's better than a standard 1TB 4GB/s nvme 4.0 soldered on the board with an option of adding slower storage later on?

And which do you think is cheaper?

Where will a double duty cell be good at? Raytracing?

Like this "leak"?
PS5 - $399
PS5+ - $549
https://pastebin.com/yCa2XzVu

What do think will be the differentiating factor between the two variants? I hope the CPU and GPU are the same and the differences are somewhere else.

 

What will be the difference in games between an optane-like SSD cache vs a standard nvme SSD?

Does having a 3d xpoint like SSD cache opens up gameplay not possible with a standard ultra fast nvme SSD? Let's say both are at the raw speed of 4 GB/s, does the I/O operation matter?

(I'm not tech savvy like you guys are but I'm reading about 3D xpoint and I like what I'm reading. lol)

Can Sony pull off an optane-like SSD cache on a $399-499 machine?

 

Optane is almost as expensive as DRAM. Modern TLC drives guarantees 1500 write cycles. That's 1500 TB written for a 1 TB drive or 15000 100GB game installs, or one 100 GB game install every day for 41 years !!

Wear is a non-issue.

Cheers

 

There's just no market acceptance because the market doesn't want/need it yet.

Why XQD versus other solutions like a user accessible NVMe port? What's the advantage? I'm guessing not cost because it's a niche format no-one's using yet.

 

Ah. No 3d xpoint for us then. I was already getting excited.

If modern TLC drives support up to 1500 TB of writes then I would go for 1TB of NVMe and my original solution of 20 GB cache for each games.

I was reading about NVMe 4.0 and the theoretical expectation is at least 4 GB/s. Can that be translated into an actual real world bandwidth of 4 GB/s in a close box like consoles? 4 GB/s to a 20 GB of RAM is 5 freaking seconds! There will be practically no loading times if that is the case.

 

I wasn't serious with this post

 

Not that I believe we will see such a solution - even though the NVMe based CFexpress 2.0 with it's theoretical 4 GB/s could be fast enough, and 8x PCIe was also promised - I can only think of the following reasons:

• spreading the standard to increase the chance for adoption like Sony did with DVD and BluRay, which in turn drives down the cost. Of course it's different, since those were required to ship games and allowing console owners to watch movies was an added bonus.
• ease of upgrading the storage for the average user. With CFexpress 2.0 you just take the current one out and put the new card into the slot, whereas with M.2 SSDs you have to open a replacement slot on the console, unscrew the previous M.2 and take it out, then screw the new one in and close the replacement slot. A non technical user which just wants to play games probably prefers the first approach.

 

Thinking back to what Cerny said about the importance of the details in their I/O and software stack implementation. It seems that they have solved the problem about NVMe mentioned in this article. https://searchstorage.techtarget.com/opinion/Understanding-the-NVMe-performance-problem

How much does this cost?
https://www.notebookcheck.net/Sony-...s-for-highest-read-write-speeds.411009.0.html

 

DVD and BluRay were open industry standards everyone was already using or going to use. Compare XQD to MemoryStick - Sony used MemoryStick (MS) exclusive in its devices to try and force adoption, and fail, and just pissed users off. The CF organisation going with CFexpress means it'll happen as an industry standard, but its not happening now and it's not something Sony could ever force. Sticking it in their console without industry backing would just mean making another over-priced proprietary format.

My point isn't that XQD won't happen, but Sony have no plans with it and aren't going to use PS5 to drive adoption. If CFexpress if coming and looks set to become affordable, it might see inclusion in PS5, because it's an industry standard and not because Sony are trying to force it to become one.

It's really not hard, a rare experience, something easily walked through with online videos these days, and not at all worth the additional cost of a proprietary format over the industry standard. If Sony went with XQD with a 30% markup, say (no idea what prices are), the people most interested in upgrading, the tech-savvy core users, and just going to complain why do they have to pay extra over NVMe, same way they complained about having to pay Sony MS premiums versus SD cards.

It's all about price and performance. If there isn't a cheaper option, or there isn't as performant an option, XQD is considered. As long as cheaper alternatives exist, it makes zero sense, for Sony and for users.

 

If it is indeed custom, do you think it's user-upgradeable or will they solder it on the board? Can they use a cfexpress 2.0 4GB/s and have a more or less similar performance to intel optane?

 

I seriously don't know, it's really out of my comfort zone.

Key points of 3d xpoint have been discussed long ago, better data granularity(random read and write speeds, lower latency on smaller data which is the most important for caching), no degradation of performance when the drive is full, much longer life span because of the higher write cycles, and for the guys saying the write cycles are meaningless, every instalation of a game on the drive and the constant migration of data would wear the drive enough in my opinion.

Someone said that 3d xpoint is priced like RAM which obviously is not true otherwise the purpose of the drive would never get out of the paper, which is to be an intermediary solution that can store data(non-volatile)

Well, there was some interesting articles on extremetech about PSN servers using PS3 SOC clustered on a MB at smaller nodes, if the PS3 soc could be used to do other things on the system like dedicated decompressor or something to free resources from the "main" system it would be interesting, RT i don't think it would be ideal, but again i'm don't think i have knowledge to talk about it.

I said out of the blue as a speculative possibility, haven't seen any leaks about that, i assume that PS4PRO will receive a price reduction and possibly a slim version as well, 399 could be a good suggestion for the PS5 barebones, without UHDBD, without full retrocompat, etc... But let's just ignore it for the sake of productivity.

 

Not impossible though...

 

I think it will be custom solution as mark said, soldered on the board. The upgradability as i suggested would be via some form of Memory card, i brought the XQD because it is the smallest one and if i'm not wrong the fastest, the ideia is that it could extend the cache size of system, it would need to be as fast or faster than the 3d xpoint solution.
Not sure about the metrics, memory cards as hole has very specific characteristics in terms of performance, i would need to take a good read to bring that info "updated" to you, but they are generally not great "all around performers".

 

Well, a dedicaced chip or dsp, why not, but Cell ? This thing is pretty much dead since 2012 and the ibm qs22 I believe...

 

https://www.extremetech.com/gaming/...dware-with-eight-ps3s-on-a-single-motherboard
My ideia is that cell could integrate the system have a double duty(ps3 games), not the ideal dedicated chip, but at 12nm or lower node it could be very efficient and significantly powerfull for this type of task.

 

Based on what maths? Going by Gubbi's data, 1500 TBs over five years would be 300 terabytes of data per year. Nearly 1000 GBs per day.

Price isn't governed by purpose. If demand for Xpoint is low and costs to make is high, it'll be priced high. The only way to know whether xpoint is priced like RAM is to actually compare prices.

There's a lot Cell could be used for. However, no-one's going to want to develop code for it. Best chance of Cell in PS5 is for BC with Sony providing system libraries that use it for something-or-other so devs don't have to get their hands dirty. So yeah, audio processing and I/O management. It's a long-shot (long, long, loooonnngg shot!), but not completely impossible.

 

I read some anandtech tests on this, but i think if samsung, for example does not explicitly says on the papersheet is possibly has some dramatic variability based on factors that they can't control or simply don't want to show. 1TB 970 EVO has 5-Year Limited Warranty or: 600 TBW...
Basically half of the 1000-1500 P/E cycles.

That's a complex arguement, i rather not discuss without a deep investigation, but one think is clear, Intel has a big chunk of the market in a wide variety of technologies, if some industry leader would know better what the type of product at what price i would bet on Intel, and just remebering that Stadia uses lots of 3d Xpoint memory and for sure it will impact in the price even more and some fabs going to China seems a decent move to get the price for the consumer market affordable.

Yeah, as i said PS3 would be a primary reason for it, using it just for back compat purposes would be very costly overall for a not decisive feature, so a double duty function could be a bonus, obviously on much lower and efficient node.

 

this does 1600 cycles.

Early TLC drives were made from planar flash, where the small cell size resulted in low number of write cycles. Modern TLC 3D NAND has larger sized cells and thus a higher number of write cycles. It is also cheaper so you can provision more NAND for error correction purposes which together push the write cycles to what planer MLC NAND enjoyed 3-4 years ago.

XPoint's primary advantage is its ability to provide byte-sized write accesses. NAND need to be erased and written in blocks (typically 256KB). Write granularity is completely irrelevant because games are large.

Cheers

 

Even if only half, that's 500 GB a day. Is it at all realistic for games to require even 100 GB a day write? You'd have to be constantly writing data, tens of megabytes a minute, to threaten the durability of these drives.