Next Generation Hardware Speculation with a Technical Spin [post E3 2019, pre GDC 2020] [XBSX, PS5]

Status
Not open for further replies.
I'm only guessing that 580 was referenced because 6TF (make of that what you will in terms of die size vs yields vs clocks etc).

edit: oh, I guess 480 could get to 5.8TF. ¯\_(ツ)_/¯

nevermind!

Yup, and there were many non-reference RX480 clocked over 1.3GHz, effectively surpassing the 6 TFLOPs mark.

NVME has over a magnitude more bandwidth than a standard mechanical HDD, but level loads rarely exceed 2x speed.
And so will my RAM drive that scores over 8GB/s in crystalmark, because it's highly bottlenecked by decompression / decryption (implemented on single threads in PC games) and not memory transfer speeds.
Remove that bottleneck and only then can you take advantage of NVMe speeds.
 
Last edited by a moderator:
Unless neither company is using standard NVME. It's highly likely that at least Sony are not, as NVME (at least on PC) can't come close to matching the numbers that Sony have demonstrated. Not even remotely close.
NVME doesn't come close to approaching what Sony has demonstrated.

This performance is quite achievable with large block sizes - the reason why it's rarely sustainable on PCs are legacy file systems and IO APIs that were designed for hard disk drive technology of the early 1990s.

If you move away from legacy sector sizes like 512B and even 4Kn, efficiency increases at a steep curve. Large block IO operations are flying in all NVMe SSD tests, and it does not require any changes to firmware or new NVMe extensions.

Considering that they are not likely to be using standard NVME hardware and drives, it would be counterproductive to allow the user to change it
Even if the SSD uses a non-NVMe compliant controller (which I doubt), using a standard M.2 form factor makes more sense for the logistics.

Cerny also talked about their customisations needed to pull it off. I think it fairly likely we're looking at a custom solution and potentially mobo soldered,

Everything Cerny said so far also makes sense for a fast PCIe 4.0 x4 NVMe SSD with a custom flash-aware file system and some features from NVMe 1.3 and NVMe 1.4, like Persistent Memory Region, IO Determinism and Host Memory Buffer Enhancements.

There is a Sony patent for an enhanced SSD subsystem, which does include a flash-memory aware file system, along with some unlikely hardware enhancements like SRAM buffers and custom IO processors.

http://www.freepatentsonline.com/20170097897.pdf
https://www.resetera.com/threads/ps...nd-sonys-ssd-customisations-technical.118587/
 
Last edited:
If Flute is PS5, and it all points to that being the case, bandwidth from benchmark (~530GB/s) and number of chips tell us its using 256 bit bus.

With that bus and BW I dont see how we can expect more then 9TF +RT hardware in next gen consoles. This also concerns Scarlett as snippet showed 320 bit bus and 14Gbps chips which would point at most 560GBs.
Or HBM 435 GB/s + DDR4 102,4 GB/s as the past rumor.
 
Everything Cerny said so far also makes sense for a fast PCIe 4.0 x4 NVMe SSD with a custom flash-aware file system and some features from NVMe 1.3 and NVMe 1.4, like Persistent Memory Region, IO Determinism and Host Memory Buffer Enhancements.
You may be able to achieve the same results with more recent standards, but for me, this quote heavily implies proprietary...
“The raw read speed is important,“ Cerny says, “but so are the details of the I/O [input-output] mechanisms and the software stack that we put on top of them.
The patents go back to 2015, so Sony's been working on this for years rather than waiting for later NVMe standards to enable it with a generic solution.
 
Unless neither company is using standard NVME. It's highly likely that at least Sony are not, as NVME (at least on PC) can't come close to matching the numbers that Sony have demonstrated. Not even remotely close.

NVMe is using the PCIe phy layer so it's limited by the available PCIe mode's bandwidth on the implemented PCIe interface. NVMe 4.0(PCIe 4.0) can provide about 2,4,8,16,32GB/s depending on the PCIe lanes. 2x over PCIe 3.0.

Ryzen 3700 NVMe is just below 8GB/s.
 
A marginally faster solution dependent on HBM memory, previous arch, and too power hungry for consideration.

<shrug> Minus "previous arch" you just described Vega, as well. I just don't see how you define the 580 as not being mid-range. AMD's high-end being crap doesn't really change the positioning of the 580.
 
<shrug> Minus "previous arch" you just described Vega, as well. I just don't see how you define the 580 as not being mid-range. AMD's high-end being crap doesn't really change the positioning of the 580.
The point was that applying desktop product stratification to console GPUs is improper. They have different market dynamics that inform their decisions.
 
Surface Pro X has a standard M.2 2230 slot - though this form factor is typically used for B-key Wi-Fi cards, there are also SATA SSDs from Samsung, Toshiba, Kingston, Transcend.

But I'd guess PCIe 4.0 NVMe x4 SSDs would require a M-key 2280 module to push the bandwidth to the maximum possible 8 GByte/s.

Microsoft has its own rssd with hardware bitlocker software on it. You can't change out the rssd without using special software to swap them. But what do I know , I just work there
 
<shrug> Minus "previous arch" you just described Vega, as well. I just don't see how you define the 580 as not being mid-range. AMD's high-end being crap doesn't really change the positioning of the 580.

Don't feed him. Vega64 wasn't the best high end but it was much and much more powerfull then a one x/RX480 or 580.

Bandwidth-wise it can and has.

PC market will have to adopt, speeds are there, or brute-force with massive amounts of ram as a requirement for games, like the 32GB main ram for that star wars game :)
 
Sony's been working on this for years rather than waiting for later NVMe standards to enable it with a generic solution
NVMe 1.4 has some useful features, but all they really need is add support for LBA sector size that matches the physical write page size (8 to 32KB), which should improve read speeds.

This should be possible on every NVMe 1.x device through the "Identify - LBA Format" command, which produces a list of preferred LBA sector sizes, ranked according to their perfomance. The sector size field, 'LBA Data Size (LBADS)', is a 7-bit value that encodes the exponent of a power of two (i.e. 2^N) - this allows a practical range of 512B (2^9) to 16MB (2^24), and up to 2^128 bytes, which they don't yet have a binary prefix for.


A custom flash-aware file system, which uses write-once logic to reduce write amplification, seemed like a great idea back in early 2000s - unfortunately, all research papers that I've read so far conclude that flash-aware filesystems perform worse in real-world workloads than legacy hard-disk based filesystems, because current SSDs implement very efficient background garbage collection.
This could probably work if Sony made a breakthrough in performing garbage collection on the OS level, and devised an allocation strategy that would align all writes to the size of the erase block (typically 1-2MB).

I'd rather say they simply need to fine-tune their APIs and applications to the more effective usage patterns of NVMe SSDs - that is, schedule deep IO batches with large block sizes, and avoid unnecessary buffering and data copying by taking full advantage of their reduced access times and enormous read bandwidth.

Microsoft has its own rssd with hardware bitlocker software on it. You can't change out the rssd without using special software to swap them.
What you refer to as 'hardware bitlocker' is the 'Encrypted Hard Drive' mode that uses the AES256 hardware encryption and TCG OPAL authentification in the hard drive. This capability was introduced in 2011 for Windows 8/Server 2012 and is currently available in many SATA and NVMe SSDs, and these drives can be managed by either BItLocker GUI or console commands.
 
Last edited:
https://www.resetera.com/threads/ne...ise-and-hot-air.129193/page-340#post-25379009

I retrieve the Dualshock 5 patent and gofreak find more details about it

We know, or at least I know, the retail cost of controllers is approximately a 800-1000% markup on production cost but I do worry a little about what manufacturers will try to charge for additional controllers for nextgen consoles because of the perceived nextgen-technology. I've always bought two controllers at launch because me and my fiancé want a controller each for two-player games and we often spring for a third because one controller always conks out before the other.

In the UK we had a variety of PS4 launch bundles that effectively included a second DualShock 4 and the PS Camera for zero when buying one of the premier launch games (WATCH_DOGS - which slipped from launch, Knack (luz) or Killzone Shadow Fall - which I actually enjoyed). I do wonder how the RRP is going to be for these things and whether nextgen consoles will support existing controllers.
 
We know, or at least I know, the retail cost of controllers is approximately a 800-1000% markup on production cost but I do worry a little about what manufacturers will try to charge for additional controllers for nextgen consoles because of the perceived nextgen-technology. I've always bought two controllers at launch because me and my fiancé want a controller each for two-player games and we often spring for a third because one controller always conks out before the other.

In the UK we had a variety of PS4 launch bundles that effectively included a second DualShock 4 and the PS Camera for zero when buying one of the premier launch games (WATCH_DOGS - which slipped from launch, Knack (luz) or Killzone Shadow Fall - which I actually enjoyed). I do wonder how the RRP is going to be for these things and whether nextgen consoles will support existing controllers.
Anyone who follows sales people on twitter or other mediums know that gamepads have been selling extremely well this gen. I am very frugal but find myself being tempted by the electric purple DS4 :) And I've also seriously considered the elite controller even though XB1X isn't my main platform. I wouldn't be surprised to see a MSRP increase to 70 or 80. As long as game prices stay $60, I don't think there will be much furor.
 
HOW DOES GAMSPOT KNOW THAT PROJECT SCARLETT USES AN ARCTURUS GPU?

https://www.gamespot.com/articles/ps5-vs-xbox-scarlett-comparing-next-gen-console-sp/1100-6470431/

They claim that previous information about the cpu and gpu was incorrect and ammended the information. Unless microsoft told them, its all rumor, so it should be impossible to know if the new information is correct.
Those 'specs' are laughable. It's just a click-bait article. I expect the corrections are just from the public pointing out MS's GPU is 'Arcturus' based on a rumoured devkit spec or somesuch. Acturus is a 128 CU chip by accounts. Is that really going into a console?
No.
 
Status
Not open for further replies.
Back
Top