Well okay then, with the help of co-processors just for the SSD we will sure get a very speedy solution yes. One other thing though, it doesn't have to be exactly as described in the patent though, but along those lines.
Next-gen console versus PC comparison *spawn
- Thread starter PSman1700
- Start date
Well okay then, with the help of co-processors just for the SSD we will sure get a very speedy solution yes. One other thing though, it doesn't have to be exactly as described in the patent though, but along those lines.
DmitryKo
Veteran
If you are talking about SSD patent application in another thread, that is US20170097897A1 https://patents.google.com/patent/US20170097897A1/en
- that's not actually "file system internal to the SSD". It's just a custom flash-aware file system on a separate disk or partition. SSDs know nothing about file systems, they work with blocks/sectors.
That figure describes their software stack for disk access, where the SSD partiton uses a custom file system to access read-only, protected game level data. This filesystem makes use of large 64 KB blocks (clusters), and the NVMe SSD controller supports hardware decompression of these 64 KB blocks, while contiguous allocation and data compression is handled in the software stack. The latter also implements a custom lightweight API (i.e. without HDD-based 'abstractions' ) to read data from this SSD partition.
Of course the software stack also supports standard file systems typical for USB media and optical/hard disk drives and SSDs, with full read-write access.
If loading times are not improved, which problem this additional RAM actually solves?
First, I doubt the patent will be granted. They basically describe an integrated 4-channel flash memory controller with a NVMe (PCIe) interface, and an address translation layer implemented with multi-core ARM processors and dedicated look-up caches. There are already dozens of such integrated chips from several vendors like Phison, Silicon Motion, Samsung, Intel, SandForce, Marvel, and others.
Also, hardware compression has already been implemented by SandForce and probably a few others, to limited succes; major vendors chose to completely ignore it. Compression can increase read throughput for lower-end flash memory parts, but only as long as your data is compressable with a decent rate. It cannot substitute for flash-aware filesystems and device IO patterns based on large blocks.
And custom API layer for accessing flash memory storage is not such a novelty either.
Last edited:
Cross platform games between Xbox and PC is a thing, and games will use this streaming of the SSD for large open world games etc, they must work something out then.
Are you talking about the Sony patent? If so, why are they regestering it, if there are already integrated chips in the hardware that do just that, compression to speed things up.
DmitryKo
Veteran
Existing SSDs are based on 512B sectors and 4 KB clusters, which fits hard disk-derived data access patterns of existing operating system. This kills their otherwise impressive sequential read speeds in raw disk IO tests by two orders of magnitude, when you run real-world random read tests on these SSD disks and operating systems.
Excerpts from the above Sony patent, which describe this problem at length:
Last edited:
I think it will take some time before developers release games which truly push the bandwidth afforded by these next gen consoles anyway. For a period after launch, we're going to be getting cross-generation multiplat games which will still be designed around the constraints of older hardware. The only games which MAY push boundaries are ones that are likely to be exclusive to PS5 to begin with.. and thus it wont really be comparable to anything.
I think developers will begin raising RAM requirements and initial load times in PC versions of next gen games will be slightly longer than their console counterparts. The games should be designed to fill up as much RAM as possible to reduce pressure on the PCIe bus. I think that might be the solution until faster drives, an even faster bus, and other possible improvements come down the pipe.
Regardless... it's all very interesting to see what happens and how it changes game design.
I think developers will begin raising RAM requirements and initial load times in PC versions of next gen games will be slightly longer than their console counterparts. The games should be designed to fill up as much RAM as possible to reduce pressure on the PCIe bus. I think that might be the solution until faster drives, an even faster bus, and other possible improvements come down the pipe.
Regardless... it's all very interesting to see what happens and how it changes game design.
This is not new. It’s not in the consumer PC space but it’s a field of research and vendors are already supplying products in the data center space.
DmitryKo
Veteran
The question was how that additional RAM would improve game loading speeds for a typical Windows SSD with heavy fragmentation from small cluster sizes.
Why not - there's a 50% chance for an US patent application to be granted. To me, it strongly looks like a collection of prior art, but I did not examine each and every claimn, like patent professionals do.
Presumably they could load more of the game into a larger memory pool and that would allow the PC to stream in assets fast enough to maintain parity with the consoles. Whereas without that additional RAM, games might just freeze for seconds at a time while levels/assets load it.
So, the in-game loading/streaming would potentially be faster, but the initial game load would take longer... which was probably the loading you're referring to.. so I guess the answer is that it wont. =/ lol
Right but this is 2019 not 2005. Games going forward are going to be made for phones / tablets / consoles / streaming services and so on.
Lets look at the main game platforms. You have phone platforms with their built in hardware but you also have xcloud and stadia which are both powered by x86 and amd graphics. Then you have the consoles which the ps4/xbox one and forward are powered by … you guessed it x86 and amd graphics cards. So a developer is still going to target pc hardware and is still going to allow you to addd more bell and whistles for those with the hardware to run it. Also going forward developers will get an extra pop of sales if they add in advance features that might not run well on the future consoels. Why ? because they will get pc sales and in the future it will allow them to easily push out enhanced editions on the streaming platforms.
Thanks for the explanation about the firmware API file system.
This is exactly what I read, they try to go around the bottleneck of PC SSD on software is optimized for HDD usage, using hardware for having less pressure on the CPU, be able to decompress fast enough the data during game streaming and customize the controller I/O for reading usage. All other things are inside every SSD.
I know some SSD have data decompressors and they use ARM CPU too but it seems the decompressor can be useful for games. This is not a PC for doing tons of things out of game. On Microsoft side, it seems the games will be decompressed by the GPU at loading time.
After I don't think this will be very fast a probably looking like benchmark SSD speed but in real-world application, 4 to 5 GB/s max or 2/3 GB/s and this is huge.
Last edited:
Sure you will get Minecraft, fortnight, rocket league that can scale to very low spec hardware.
Not only can games scale they can be successful, good and enjoyable. No one is denying or saying that's not the case for many games.
But if you think every game can scale like that then you are simply mistaken.
The reason current minimum spec for many big budget cross platform games aren't higher is because the games was made with the current gen consoles in mind.
There are things that aren't done or even considered simply because of the consoles performance profiles.
I'm not talking just scaling graphics which is the easiest part to do.
Same with ssd, currently on pc ssd's are used as quality of life improvement because consoles have hdd. When cross platform games make use of ssd, you will not be able to use a hdd based pc.
Sure you could run it, but it won't be what anybody would consider playable.
Something to consider regarding cross platform/minimum spec support.
- Longer because tools, engines and api's are the same
- Shorter because already have knowledge of api's, tools and engines
Are patents ever denied these days? The Patent Office makes money from granting patents. It's in the best interests to grant all of them, and let the law courts argue out the details of violations and precedent.
D
Deleted member 11852
Guest
The main cost is in filing the patent, or related "valued-added services", not in the granting of the patent or maintaining it.
The patent system is in dire need or reform but nobody can agree how. You have to protect those sinking a ton into expensive R&D for genuinely new useful methods and technologies because if you don't there is no incentive to spend to innovate if competitors can use your work for free, but equally some things patented are just ridiculous. The biggest stumbling block is the need change the patent system globally - you can't just do it in one country or region.
Not from what I've seen. It's £445 to file if proceeding all the way to grant, and then increasingly expensive to maintain up to £610 in the 20th year, totalling £4640 to maintain it. Every patent not granted is a long term loss of four grand. The PO makes no money from rejecting patents, hence my doubt (in the face of clear evidence to the contrary with patents granted for tech that clearly isn't patentable) that they do any work to ensure patents are actually valid and instead just grant all of them.
D
Deleted member 11852
Guest
Ah, I wasn't aware it changed last year in the UK. As you can see from the changes, it is unashamedly about revenue whereas previously it was not.
DmitryKo
Veteran
I just wanted to clarify this in the context of the above patent being applicable to the PC.
Hardware compression would save CPU cycles, but that's less of a concern for a 8-core CPU, and the effect on the read throughput would be minimal - block level data compression is excellent for text (ASCII) files, but far less effective for binary data like textures and geometry, and absolutely ineffective for compressed files.
Fast load times and high read throughput are allowed by the custom read-only file system that uses large blocks and contiguous allocations. If you only have to write your data once, and cannot change it afterwards, there is much less fragmentation - only the free space can become fragmented when you delete some data, but with contiguous allocation, defragmenting the entire volume to consolidate the free space could only take seconds.
Emulating this on the PC with a standard NTFS file system will be ineffective even with large 2MB clusters, unless you implement similar file acces / allocation / defragmentation APIs on the OS level.
Yes, it does look so. But even if that Sony patent is granted, its claims include both SSD controller hardware and OS software stack - and this claim won't apply to the PC, since there can be no single product that implements both claims. There are SSD vendors (and they are not interested in data compression) and there are OS vendors - who would they sue in court?
Yes, we are talking about ways to achieve both <1s start time and fast level loading times. And it doesn't look like a PC with 32 GB of RAM - but a SSD formatted with a standard file system, which gives you typical random read thoughput of 20 MB/s on your typical heavily fragmented volume - will be comparable to a game console with a custom file system and proprietary APIs.
Last edited:
I just wanted to clarify this in the context of the above patent being applicable to the PC.
Hardware compression would save CPU cycles, but that's less of a concern for a 8-core CPU, and effects on the read throughput would be minimal - block level compression is excellent for text (ASCII) files, but far less effective for binary data like textures and geometry, and absolutely ineffective for compressed files.
Fast load times and high read throughput are allowed by the custom read-only file system that uses large blocks and contiguous allocations. If you only have to write your data once, and cannot change it afterwards, there is much less fragmentation - only the free space can become fragmented when you delete some data, but with contiguous allocation, defragmenting the entire volume to consolidate the free space could only take seconds.
Emulating this on the PC with a standard NTFS will be ineffective even with large 2MB clusters, unless you implement similar file acces / allocation / defragmentation APIs on the OS level.
Yes, it does look so. But even if that Sony patent is granted, its claims include both SSD controller hardware and an OS software stack - and this claim won't apply to the PC, since there will be no single product that implements both claims. There are SSD vendors (and they are not interested in data compression) and there are OS vendors. Who would they sue in court?
Yes, we are talking about ways to achieve both <1s start time and fast level loading times. And it doesn't look like a PC with 32 GB of RAM - but a SSD formatted with a standard file system, which gives you typical random read thoughput of 20 MB/s on your typical heavily fragmented volume - will be comparable to a game console with a custom file system and proprietary APIs.
Textures use GPU format compression and geometry does not need it and have its own form of compression like displacement mapping usable on GPU. I think they want to compress some of the files for keep at low as possible game size on SSD.
D
Deleted member 11852
Guest
Remember that most SSDs (a single drive/controller package) employ a native data structure that bears little relation to the OS filesystem, managed by the controller to encrypt, map bad blocks, manage garbage collection and wear levelling and crypto-shredding but custom file systems are supported all most desktop OSs so a patent predicated drive, controller and filesystem (software stack) wouldn't necessarily preclude it applying on PC.
But I doubt Sony care about anybody using their implementation on PC, they presumably want to make it difficult for competitors in the console space to use their approach. By PC I exclude macOS as Apple have been switching to custom controllers for years and APFS is the most solid-state friendly desktop filesystem in widespread use.