Next Generation Hardware Speculation with a Technical Spin [2018]

[eastmen]

Crucial just launched new 500MB/s value SSDs. $29 for 120GB. $89 for 480GB. Seems silly to think these won’t be feasible for consoles with bulk pricing.


https://www.guru3d.com/news-story/crucial-silently-launches-bx500-ssd.html

shhhh your going to awaken Mr Fox and he will be mad that someone suggests that nand will replace mechanical drives !

But yes I think something like this will be feasible for next gen consoles. THey need to move on from mechanical drives . I would assume they will be pushing 16-24 gigs of ram (hopefully 24gigs) and it will take a really long time to fill that ram up with a mechanical drive.

 

[Lalaland]

shhhh your going to awaken Mr Fox and he will be mad that someone suggests that nand will replace mechanical drives !

But yes I think something like this will be feasible for next gen consoles. THey need to move on from mechanical drives . I would assume they will be pushing 16-24 gigs of ram (hopefully 24gigs) and it will take a really long time to fill that ram up with a mechanical drive.

I share Mr Fox' skepticism, recall that the 500 Million PS4 has a 2TB drive and that the Pro ships with 1TB, marketing will never allow them to ship a box with a smaller number than the last one. Games are also starting to blob much larger than 60GB plus, BF1 is already 110GB, Destiny 2 recently got a 35GB "patch" (forced DLC install) so a 500GB drive just isn't going to cut it in a new mainstream console. I am encouraged by talks of a 3D NAND glut but I still remain very doubtful we'll see a standard SSD, more likely to see a 7200rpm mechanical drive first which should be the minimum disk IO boost for next gen.

 

[BRiT]

It keeps coming back to budget, and where to spend it. I dont see them able to get a traditional SSD mass storage drive for a reasonable tiny fraction of the BOM similiar to what a mechanical HD mass storage drive uses. So I'm left to hope for a fast scratch-pad/launch-pad accelerator drive around 120 GB in addition to the mechanical mass storage drive.

 

[Deleted member 13524]

I share Mr Fox' skepticism, recall that the 500 Million PS4 has a 2TB drive and that the Pro ships with 1TB, marketing will never allow them to ship a box with a smaller number than the last one. Games are also starting to blob much larger than 60GB plus, BF1 is already 110GB, Destiny 2 recently got a 35GB "patch" (forced DLC install) so a 500GB drive just isn't going to cut it in a new mainstream console. I am encouraged by talks of a 3D NAND glut but I still remain very doubtful we'll see a standard SSD, more likely to see a 7200rpm mechanical drive first which should be the minimum disk IO boost for next gen.

I think the point is solid state storage seems to be well under way to match (and eventually surpass) hard drive storage in price-per-TByte.
A 500GB SSD costs as much as a 2TB 2.5" HDD. Next year could be 1TB SSD = 2 TB HDD and in 2020 they could be close to parity.

 

[Shifty Geezer]

I think more importantly, $30 additional BOM for an SSD alongside an HDD is a no-brainer. Launching at $430 instead of $400 or whatever won't be a problem and the benefits will be well worth the added cost. It'd also mean later consoles could go SSD only and benefit from reduced prices assuming SSD starts pricing below HDD cost per GB, while using the speed advantage properly because the first consoles had the same SSD for devs to target.

 

[MrFox]

Going 100% flash, there's a problem with expandability if they want guaranteed performance, that will seriously limit the sort of external SSDs allowed. Consumer SSD have unpredictable behaviour, with sporadic performance dips.

Two tiers would allow low cost mass storage, low cost expandability, while still providing fast and predictable local storage.

The question is, what can they get for about $50 and will that be enough. Whenever they launch. 1TB for 50 is quite some time away, and might not even be enough depending on next gen game sizes.

 

[DieH@rd]

Crucial's latest SSD lineup features a $29 120GB drive.
https://www.guru3d.com/news-story/crucial-silently-launches-bx500-ssd.html

 

[BRiT]

Going 100% flash, there's a problem with expandability if they want guaranteed performance, that will seriously limit the sort of external SSDs allowed. Consumer SSD have unpredictable behaviour, with sporadic performance dips.

Two tiers would allow low cost mass storage, low cost expandability, while still providing fast and predictable local storage.

I think we were all thinking about that, but haven't vocalized it, so thanks for clearly stating it. If the internal drive is only SSD then that limits expansion of external drives to be SSD only. Certainly that would be bad from a consumer perspective. That seems to be more of a push to a hybrid storage pool approach.

 

[eastmen]

I share Mr Fox' skepticism, recall that the 500 Million PS4 has a 2TB drive and that the Pro ships with 1TB, marketing will never allow them to ship a box with a smaller number than the last one. Games are also starting to blob much larger than 60GB plus, BF1 is already 110GB, Destiny 2 recently got a 35GB "patch" (forced DLC install) so a 500GB drive just isn't going to cut it in a new mainstream console. I am encouraged by talks of a 3D NAND glut but I still remain very doubtful we'll see a standard SSD, more likely to see a 7200rpm mechanical drive first which should be the minimum disk IO boost for next gen.

I think your focusing on the wrong thing. I do agree they would have to go 1TB for a next gen console , but at least on MS's side they allow you to easily add a hardrive. So there is no reason why they can't release with a lower capacity drive. They will just focus on the speed instead of the size.

Look at it this way Western Digital is starting to close hardrive plants. They will close one in 2019 and open another ssd plant. The writing is on the wall. SSD's continue to get faster and cheaper while hardrives have largely stayed the same. These companies will continue to transition away from mechanical drives as users flock to SSDs

A WD Black 7200rpm 5 1/4 drive will get you 130/MBs sustained reads a decent ssd will do in the 400MB/s range, if you go NVME you can go into the thousands of MB/s . I don't think we will see a new console till 2020 at the earliest , with that said I think in that time you will see 1TB sata ssds come down in pricing under the $100 mark . All the companies are announcing 96layer 3D nand for this year , QLC drives and other options all coming down .
QLC seems ideal for a console. Its something you will write too sparingly but it increases storage capacity 33% over TLC.

 

[Entropy]

There is no reason for a new console design to use a legacy interface like SATA.

 

[Lalaland]

There is no reason for a new console design to use a legacy interface like SATA.

Cost.

NVMe is SSD only which would make expansion all but impossible as the latencies and b/w differences vs an SSD on USB3.1 type 2 (let alone type 1 or 3.0) would be dramatically worse. Developers could choose not to use all that lovely low latency access (and due to encryption will be dealing with some discount to raw perf anyway) but then why saddle yourself with the cost?

I think a hybrid set up is my best case scenario for next gen I/O 120 GB or so of high write tolerance flash (so very not QLC, perhaps not TLC) streaming games from a larger mechanical disk. The biggest concern I would have is whether it is possible to do this seamlessly or would the API need to hint to the caching mechanism what is up next? In an open world game if I am walking around could they design an O/S level caching algorithm that could ensure prompt asset loading, after all if you overload vRAM (on PC) it results in obvious stuttering and that's a RAM to vRAM transfer that happens at far higher peak data rates than mechanical I/O to RAM. Or would the normal asset management provided by the game itself be enough. I guess I'm thinking of a theoretical open world game like Just Cause, right now jets in JC are daft slow to allow asset streaming to keep up (partially anyway) in a hybrid I/O setup do I set the jet speed for when the game is all in the scratchpad or a lower speed to cope with those moments when the game is halfway between the two to prevent asset load pauses?

Edit: I take your point Eastmen regarding reduced mechanical disk manufacturing capacity but I would argue that has little to do with the technology becoming obsolescent and more to do with it's largest consumer market dying. Mechanical disk are like tape, unsexy and slow but like tape they have a USP, tape is it's robustness and longevity, mechanical disks is a combination of capacity and good enough random access performance for Near Line storage. With the death of the general consumer PC (and increasingly notebooks) there just isn't much demand for bog standard mechanical drives but they will still sell millions upon millions for data centres for decades to come. So for the foreseeable future there will be plenty of 2.5" SATA mechanical drive manufacturing capacity to leverage.

 

[anexanhume]

Cost.

NVMe is SSD only which would make expansion all but impossible as the latencies and b/w differences vs an SSD on USB3.1 type 2 (let alone type 1 or 3.0) would be dramatically worse. Developers could choose not to use all that lovely low latency access (and due to encryption will be dealing with some discount to raw perf anyway) but then why saddle yourself with the cost?

I think a hybrid set up is my best case scenario for next gen I/O 120 GB or so of high write tolerance flash (so very not QLC, perhaps not TLC) streaming games from a larger mechanical disk. The biggest concern I would have is whether it is possible to do this seamlessly or would the API need to hint to the caching mechanism what is up next? In an open world game if I am walking around could they design an O/S level caching algorithm that could ensure prompt asset loading, after all if you overload vRAM (on PC) it results in obvious stuttering and that's a RAM to vRAM transfer that happens at far higher peak data rates than mechanical I/O to RAM. Or would the normal asset management provided by the game itself be enough. I guess I'm thinking of a theoretical open world game like Just Cause, right now jets in JC are daft slow to allow asset streaming to keep up (partially anyway) in a hybrid I/O setup do I set the jet speed for when the game is all in the scratchpad or a lower speed to cope with those moments when the game is halfway between the two to prevent asset load pauses?

Edit: I take your point Eastmen regarding reduced mechanical disk manufacturing capacity but I would argue that has little to do with the technology becoming obsolescent and more to do with it's largest consumer market dying. Mechanical disk are like tape, unsexy and slow but like tape they have a USP, tape is it's robustness and longevity, mechanical disks is a combination of capacity and good enough random access performance for Near Line storage. With the death of the general consumer PC (and increasingly notebooks) there just isn't much demand for bog standard mechanical drives but they will still sell millions upon millions for data centres for decades to come. So for the foreseeable future there will be plenty of 2.5" SATA mechanical drive manufacturing capacity to leverage.

Indeed, they may need to deploy it as an oversized cache, with some graceful degradation accounted for, and I hope it’s not soldered to the MB.

https://www.techpowerup.com/220432/high-end-slc-ssds-no-more-reliable-than-mlc-ssds-google-study

 

[Lalaland]

Fascinating study, impossible to replicate with TLC given the relative newness of that tech but I look forward to the follow up that gets there. I'll bet that study made a lot of my former storage colleagues very sad, the margin on SLC paid for a lot of very nice cars

 

[Entropy]

Cost.

Depends on the scenario.
If the SSD is just a buffer for a harddrive, using SATA makes no sense at all. Users wanting greater capacity will exchange/complement the hard drive, not the fast buffer. In fact, it might just as well be soldered.
If you actually store your purchases on the SSD, then you might eventually want to replace or add a complementary drive. The easiest way is to hang your original SSD on PCI-e and simply unplug it, and plug in your new and bigger/faster.
It is only in the scenario you painted, where you store all your games, and run them, from an internal SSD, and want to complement that with an external SSD connected via USB-C where games are also stored and run from, and you assume that having a faster internal SSD could create a speed mismatch that could somehow create problems.
To me, that seems like an unlikely scenario. I don’t think it is the most probable hardware configuration (one large SSD to store everything in the base configuration), and even if it is, I wouldn’t assume that having a slow internal SSD is a good idea anyway. Saddling all systems with a needlessly slow SSD on the assumption that the developers or OS cannot handle the user adding slower external drives seems like dubious design decision when it is actually easily solvable (if indeed the design follows this approach in the first place), by just designating part of the fast internal storage as scratch space for the active game.
(Personally, I think having 128GB of really fast SSD acting as a buffer for games stored on a large, cheap harddrive is a reasonable approach. Cost matters. Speed matters. This approach adresses both in a reasonably cost effective way.)

 

[Lalaland]

My bad, I misunderstood your original post as implying that the consoles should forgo SATA entirely and go for a pure NVMe SSD based solution, if they do hybrid I completely agree the SSD portion should be on NVMe.

 

[Entropy]

My bad, I misunderstood your original post as implying that the consoles should forgo SATA entirely and go for a pure NVMe SSD based solution, if they do hybrid I completely agree the SSD portion should be on NVMe.

Expandability is nice. It allows the big buyers to, well, buy more! What platform holder would want to inhibit that? And it also allows them to ship a base console that is cost cut to the bone, yielding an attractive price of entry. I don’t think think we need to worry too much.

 

[DieH@rd]

PS5 should have a replaceable mass storage HDD drive and replaceable small M.2 drive, with users having the ability to split the cache[for slow HDD]/data SSD partitions how they like. That way, slow HDD will always get fast caching, and users who want larger SSD can easily add it.

 

[MrFox]

A user replaceable flash (m.2 or sata) would mean devs have to account for the worst case ssd performance on the market. The console have to spec a guaranteed transfer rate and latency for the devs.

Putting the nand cache on the motherboard allows both cost cutting and performance guarantees.

 

[Shifty Geezer]

Nah. Target the default performance. If someone chooses to use slower flash (assuming the default isn't the slowest!), that's their call. Anyone savvy enough to replace the flash drive should be expected to be able to follow some performance guidelines in the manual, and The Internet will provide lots of advice to look up.

 

[Arkham night 2]

Grazy bandwith rumours 500 -700 GB/s Ps5