General external expansion discussion? *spawn*

[BRiT]

If rumours are true, the O/S is only 2GB. Not a huge chunk.

That's memory use, not how much drive space it uses or reserves for purposes like streaming/recording.

I could see it using vastly more drive space than memory space if it's modular and paged in and out of memory for various sections like game storage management or the store, etc and pre-packaged apps like music playing or media playing or bluray etc.

 

[function]

Wouldn't expect early indications via direct storage.
Doesn't sound like it's due on pc until next year?

Sorry, I wasn't clear. What I meant was that I expect drives that certify for PS5 compatibility to likely be good at handling the demands of Direct Storage.

Sony have surely built a pretty strong SSD setup, and any third party SSD that can stand up to that (and go beyond) is likely to be a good choice of device for future PC work - particularly once PC games are taking advantage of Direct Storage!

But again you’re assuming a lot. You’re assuming the 5.5 figure from Sony is sustainable, not sure they (or even MS) have stated as much?

I think it’s fair to say whatever measure Sony are using they are aware of what products are coming out that will be compatible, IIRC Cerny even states that drive slow will be out around the same time as PS5.

All this assuming the worst when it comes to Sony and the PS5 is frustrating.

No-one is assuming the worst for Sony. There will be a baseline of PS5 SSD performance that can be sustained regardless of power/temperature. Whether that's full throttle or something lower doesn't change the fact that there is a baseline for acceptable performance.

A third party drive will have to be able to meet a minimum level of capability across all areas of performance. That's why they're testing drives before recommending them.

This is simply how things are, and not an attempt to victimise Sony.

MS have stated that their SSD won't throttle btw.

 

[function]

If rumours are true, the O/S is only 2GB. Not a huge chunk. Not sure if it will have virtual memory set aside but that's a difference conversation entirely.

Not a chance that the OS partition of the SSD is only 2GB I'm afraid. Think in the tens of GB. Maybe more.

That's memory use, not how much drive space it uses or reserves for purposes like streaming/recording.

I could see it using vastly more drive space than memory space if it's modular and paged in and out of memory for various sections like game storage management or the store, etc and pre-packaged apps like music playing or media playing or bluray etc.

Yup! And I have real doubts about the "2GB system reserved ram". It came from sources with no credibility. MS have never needed more memory for OS features, infact they have always needed less (yes, even on PS4/Pro which had an additional memory chip).

It would be odd for Sony to have an advantage here, and also for them to have said nothing about it.

 

[iroboto]

But again you’re assuming a lot. You’re assuming the 5.5 figure from Sony is sustainable, not sure they (or even MS) have stated as much?

I do assume it is. MS has stated already that theirs is guaranteed 2.4GB/s all scenarios. So thermal throttling has been removed from the equation.

I just assume Sony has done the same thing. It’s hard to optimize a console with a moving target.

 

[arandomguy]

Read workloads relatively speaking are very easy on SSDs. It's really only long sustained write work loads on NVMe drives on the PC that can cause throttling eventually. Writes in general is going to be a negligible work load for gaming, I don't think either Sony or MS has even stated anything with respect to write performance?

 

[Silent_Buddha]

Read workloads relatively speaking are very easy on SSDs. It's really only long sustained write work loads on NVMe drives on the PC that can cause throttling eventually. Writes in general is going to be a negligible work load for gaming, I don't think either Sony or MS has even stated anything with respect to write performance?

Writes cause heat to build up faster in the NAND chips but reads will also generate enough heat to cause the NAND chips to heat up enough for the drive to throttle.

Reads and writes can both easily cause the controller chip to overheat as well. That's a major source of throttling on cheaper high speed NVME drives.

Regards,
SB

 

[arandomguy]

Writes cause heat to build up faster in the NAND chips but reads will also generate enough heat to cause the NAND chips to heat up enough for the drive to throttle.

Reads and writes can both easily cause the controller chip to overheat as well. That's a major source of throttling on cheaper high speed NVME drives.

Regards,
SB

Can you show some examples of throttles in reads?

If you look at TPU's last 10 NVMe drive reviews as an example 9 out of the 10 did not throttle reads at all even after 10 minutes of sustained peak reads. Only 1 drive, Crucial P5, throttled after over 3 minutes and even the reviewer noted that he can't remember another time it's happened in the test. Controller temperature difference is pretty dramatic between the sustained read and write scenarios, 10C or more (some I think almost 30C?).

It seems like it would essentially be extremely unlikely for the SSDs in the either console to throttle due to heat in most operating conditions. I'm not even seeing the possibility of 1st party games for Sony actually sustaining peak reads for prolonged duration (as in minutes at a time) in practice (not sure what it would be used for?).

 

[Deleted member 11852]

Everybody is talking like it will not be possible to get an expansion drive for the ps5 before the end of times. But Sony are delivering a 825GB drive with the console. And the price for that console is 499/399 including the drive.

Stop it with your crazy logic!

This article touches on it somewhat.

https://www.eetimes.com/the-future-of-charge-trapping-flash-memory/

https://cdn.selinc.com/assets/Literature/Publications/White Papers/0015_NANDflash_IO_20141211.pdf?v=20170217-161047

Thanks for coming back, these are from a while back (2013 and 2014) and things have moved on - in that time we've seen cells nodes decrease without issue. The reason the larger nodes are liked is a) size generally isn't an issue, and b) older large node processes have really good yields. Heat off the controller is something easily managed and it's certainly easier to eat that cost than more to a more expensive process. Engineers tend to do that the there are no other options.

 

[iroboto]

Read workloads relatively speaking are very easy on SSDs. It's really only long sustained write work loads on NVMe drives on the PC that can cause throttling eventually. Writes in general is going to be a negligible work load for gaming, I don't think either Sony or MS has even stated anything with respect to write performance?

#133

Samsung 980 Pro.
We've worked out that Xbox uses 2.4GB/s writes. It writes and compresses out all contents of memory and reads compressed into memory on quick game swap ~6 seconds consistently.

 

[Silent_Buddha]

Can you show some examples of throttles in reads?

If you look at TPU's last 10 NVMe drive reviews as an example 9 out of the 10 did not throttle reads at all even after 10 minutes of sustained peak reads. Only 1 drive, Crucial P5, throttled after over 3 minutes and even the reviewer noted that he can't remember another time it's happened in the test. Controller temperature difference is pretty dramatic between the sustained read and write scenarios, 10C or more (some I think almost 30C?).

It seems like it would essentially be extremely unlikely for the SSDs in the either console to throttle due to heat in most operating conditions. I'm not even seeing the possibility of 1st party games for Sony actually sustaining peak reads for prolonged duration (as in minutes at a time) in practice (not sure what it would be used for?).

Iroboto posted a link to it, twice now.

But to add to that, the specific 7 GB/s drive (Samsung 980 PRO) is relatively affordable due to moving to TLC, but that also means it's endurance has been halved compared to the 970 PRO. It's more of an upgrade to the 970 EVO than the 980 PRO.

Anyway, as to heat. Samsung did 2 things to attempt to control heat and mitigate throttling.

• There's a heat spreader over the controller chip.
• There's a copper backplate that helps to absorb heat.

The problem is that this was designed with the average PC use case in mind. Both of those are only able to absorb heat from brief bursts of high read speeds. After that neither are sufficient to cool the chips enough to keep the drive from throttling.

With the copper backplate it can absorb a limited amount of heat. Usually enough for a quick burst of activity that PC users currently see due to the file system and games being unable to maximize the potential of the SSD. Since it's on the back of the drive, there is pretty much no way to actively cool the copper backplate to a significant degree. It's similar for the controller chip. That heat spreader can absorb momentary increases in heat, but has limited ability to dissipate that heat rapidly.

So, it can handle, for example, a common PC use case of loading a game, then minutes going by before it needs to load another level. Or the relatively light duty of streaming in textures and data which is limited by the file system (small file i/o is SLOW on NVME drives MB/s and not GB/s, but still faster than HDD) and game design. This is going to change when DirectStorage and games designed for it arrive on PC. Without active cooling the drive is going to be in a throttled state likely the entire time a game is being played unless the PC user attaches heatsinks to the 980 PRO and combines that with active cooling.

Cheaper NVME drives from budget makers like ADATA, Patriot, etc. may not even have those. Meaning they'll throttle quicker and harder than the Samsung drive. Although ADATA's engineering sample drives did feature a metal heat spreader on the controller chip. So they might at least have that.

Regards,
SB

 

[Deleted member 13524]

Below is the 980 performance over time with temperature.
It didn't even last 110 seconds before it started throttling. And it hit a massive drop further down to below 2000 GB/s.

You think the SSD in the PS5 will need to be sustaining 7GB/s for over 110 seconds?
That's 770GB, or around 10x a typical game's install size, in less than 2 minutes.

How is "not even lasting 110 seconds at sustained 7GB/s" even an actual concern to anyone?

 

[BRiT]

You think the SSD in the PS5 will need to be sustaining 7GB/s for over 110 seconds?
That's 770GB, or around 10x a typical game's install size, in less than 2 minutes.

If the games are designed around streaming in the entire scene from frame to frame like many have been talking about for so long, then yes.

 

[iroboto]

You think the SSD in the PS5 will need to be sustaining 7GB/s for over 110 seconds?
That's 770GB, or around 10x a typical game's install size, in less than 2 minutes.

How is "not even lasting 110 seconds at sustained 7GB/s" even an actual concern to anyone?

Is it relevant whether it does or does not? If Sony guarantees 5.5GB/s all times no throttling and developers develop around that number, you would have to ensure that the accepted drives can operate in those same conditions and still meet those requirements.

 

[manux]

If sony was smart at all and I believe they are they would have designed ps5 so that there is good airflow around ssd. i.e. similar solution that pc would have if you add a fan for ssd. Samsung 980 pro is 6.2W active read. Doesn't take a lot of airflow to cool that down. If that is not enough design a generic heatsink the additional ssd attaches to. This could come to the point cerny said that not all nvme ssd's would fit inside ps5.

I can't wait for ps5 teardown to see what sony really did there

 

[Deleted member 13524]

If the games are designed around streaming in the entire scene from frame to frame like many have been talking about for so long, then yes.

Frame-to-frame, even at 60FPS (which I very much doubt), means it's taking data once every 16ms. You can't assume it's continuously taking data throughout the whole 16ms. At 8GB/s effective I/O on the PS5 that means taking 128MB per frame.
How would any game need 128MB of new data every frame, continuously, for almost 2 minutes?

Is it relevant whether it does or does not?If Sony guarantees 5.5GB/s all times no throttling and developers develop around that number, you would have to ensure that the accepted drives can operate in those same conditions and still meet those requirements.

Yes, it is important. I'll bet what you want that no console from 2013 onwards could run furmark + prime95 for over a minute without shutting down or thermally thottling.
No matter how much safety there is in place, gaming consoles are engineered for realistic gaming scenarios, not power viruses. Imposing a continuous 7GB/s throughput on a SSD is a storage power virus, not a realistic scenario.

 

[BRiT]

How would any game need 128MB of new data every frame, continuously, for almost 2 minutes?

That's been my question since the beginning.

 

[manux]

One datapoint could be miles morales. PS5 install being 50GB it would be difficult to imagine that game would require constant streaming speed of 10GB+,... Even if the whole world still consists of blocks and those blocks are streamed in realtime same as ps4 version. We might have hit the point where cost of asset creation and game install size is bigger issue than io-speed. if that turns out to be true developers would probably be pretty happy, one less problem to worry about.

 

[iroboto]

Frame-to-frame, even at 60FPS (which I very much doubt), means it's taking data once every 16ms. You can't assume it's continuously taking data throughout the whole 16ms. At 8GB/s effective I/O on the PS5 that means taking 128MB per frame.
How would any game need 128MB of new data every frame, continuously, for almost 2 minutes?

Your assumption here is that all data is the same, thus all data is equally compressible. There are very specific texture encodings like BC7 that Kraken cannot compress, even with oodle is only getting 10% more if you want to stay lossless. You can be anywhere between 5.5GB/s raw to up to 9 or 11. But depending on what you are retrieving from the SSD is going to be dependent on how much you will transfer. So depending on what's happen you may need to call more information.

Yes, it is important. I'll bet what you want that no console from 2013 onwards could run furmark + prime95 for over a minute without shutting down or thermally thottling.
No matter how much safety there is in place, gaming consoles are engineered for realistic gaming scenarios, not power viruses. Imposing a continuous 7GB/s throughput on a SSD is a storage power virus, not a realistic scenario.

This is sound logic, but once again, we have so little infromatoin about Sony's console on a hardware level that I cannot say what type of typical usage scenarios they have decided to build the platform around.

I know for MS they rallied behind a guaranteed 2.4GB/s raw bandwidth in all scenarios with thermal throttling. That is all we know, and that the drive itself sold by Seagate will replicate the results in the internal drive.

I can't say for sure if there is or isn't thermal throttling on their drives, or Sony's drives. I can't say for sure if Sony provided guaranteed numbers or just the highest drive speed number. And that's why I've been generally cautious instead of optimistic on what type of drives we can purchase for ps5.

 

[Deleted member 13524]

Your assumption here is that all data is the same, thus all data is equally compressible. There are very specific texture encodings like BC7 that Kraken cannot compress, even with oodle is only getting 10% more if you want to stay lossless. You can be anywhere between 5.5GB/s raw to up to 9 or 11. But depending on what you are retrieving from the SSD is going to be dependent on how much you will transfer. So depending on what's happen you may need to call more information.

How does this make a continuous request of 7GB/s raw data more feasible? All you're saying is the 128MB per frame is an average and not the same value. Well yes it is, but does it matter for discussing whether or not the Samsung 980 Pro is able to go into the PS5 based on the fact that it throttles after 110 seconds of continuous 7GB/s reading?

Besides, the add-on NVMe probably won't even read at 7GB/s. Cerny simply said he'd probably need 7GB/s capable controllers for their switching speed to compensate for the different setup of priority lanes. I.e. the NAND won't be pushing 7GB/s.

 

[iroboto]

How does this make a continuous request of 7GB/s raw data more feasible? All you're saying is the 128MB per frame is an average and not the same value. Well yes it is, but does it matter for discussing whether or not the Samsung 980 Pro is able to go into the PS5 based on the fact that it throttles after 110 seconds of continuous 7GB/s reading?

Besides, the add-on NVMe probably won't even read at 7GB/s. Cerny simply said he'd probably need 7GB/s capable controllers for their switching speed to compensate for the different setup of priority lanes. I.e. the NAND won't be pushing 7GB/s.

Well gaming sessions can last hours on end, the throttling isn't a result of loading for 110 seconds, it's a result of heat. Once the controller has gone beyond the heat boundary it throttles. Going hard for 7GB/s for 2 minutes generates heat that eventually leads to the throttle. We don't even know how to insert a nvme drive into the PS5, I have no clue what it looks like in there. You have these gaming sessions for hours on end, the ambient temperature of that case is bound to be higher than what you find in a PC.