PlayStation 4 (codename Orbis) technical hardware investigation (news and rumours)

[Pixel]

"In reality, it's actually slightly slower since the electrical interface is no longer as clean with another chip hanging off of some of the shared lines" - 3dilettante

No need to tell me that clamshell mode doesn't directly impact memory bandwidth. I just searched up 3dilettante's posts. The challenges posed by signals integrity may have forced Sorny to drop gddr5 memory speeds from 6Ghz to 5.5Ghz

The APU seems like it would have been up to the task, AMD was able to handle speeds in that range elsewhere. The speed grades for the highest density GDDR5 chips in clamshell mode may have been where it came down to capacity vs bus speed.

Clamshell mode works by having two GDDR5 devices share the same command lines and split the normally 32-bit bus between them.
In the PS4, one of the GDDR5 chips in each pair mirrors its counterpart on the other side of the PCB, and each one claims 16 bits of data bus for itself.


Basically, two different chips act like they are a single double capacity chip. The actual bandwidth total is the same as if there were just one chip with all 32 bits to itself.
In reality, it's actually slightly slower since the electrical interface is no longer as clean with another chip hanging off of some of the shared lines. The capacity upgrade has been pointed to as a reason why memory speeds went down from 6.0 to 5.5 Gbps.

Another point that was raised was that clamshell mode impinges on the top clock a given bin can reach. Clamshell makes GDDR5 devices share various signal lines, which is not as clean electrically.

DRAM capacity doesn't really have much effect on power consumption, and clamshell mode means the power-hungry interface isn't growing.

 

[Deleted member 11852]

"In reality, it's actually slightly slower since the electrical interface is no longer as clean with another chip hanging off of some of the shared lines" - 3dilettante

I don't disagree with this is principle.

No need to tell me that clamshell mode doesn't directly impact memory bandwidth. I just searched up 3dilettante's posts. The challenges posed by signals integrity may have forced Sorny to drop gddr5 memory speeds from 6Ghz to 5.5Ghz

However in practise there is no evidence to support this. x16 (clamshell) mode is an alternative to using higher density chips. Rather than accessing a single 2 gigabit module you're accessing two 1 gigabit modules. Our server farm memory setup is similar and we run at much higher frequencies and memory densities and had no issues in using our equivalent of clamshell setup for boosting capacity without impacting bandwidth.

It's not like PS4 is running at the bleeding edge - we're talking a mere 176 gigabytes a second and you're not at the type of electrical tolerances where splitting the modules is going to measurably impact performance.

 

[Deleted member 11852]

Probably minor everywhere. It's not even like unlocking 1/14 of the GPU or anything (assuming it's only half a core or something they're giving back).

Why are you "assuming" it's half a core when there is no suggestion of that?

 

[Clukos]

Yeah wording makes it seem like it's a full core rather than just accessing a part of it.

 

[Recop]

Probably minor everywhere. It's not even like unlocking 1/14 of the GPU or anything (assuming it's only half a core or something they're giving back).

Keeping up with the Joneses though. Now Xbox will only have a slight clock advantage (the part that will never go away) and it will be harder to detect in practice then when they had clock+7th core advantage, which was pretty minor to start. You might see a FPS or two difference somewhere when CPU limited...

Actually, we didn't really see this advantage even when developers claimed to use it : http://gamingbolt.com/project-cars-...cpu-to-offload-tasks-game-utilizing-amds-eqaa

 

[function]

Sony were conservative with reservations, probably to give them room to counter SNAP TVTVTVSPORTS if that proved to be a killer use case for consoles. It hasn't, so now I wouldn't be surprised to see them releasing some of their substantial memory reserve too. Anything to help with caching has to be a good thing, given the slow laptop drives that console makers favour.

 

[Clukos]

Theoretically 3.5 inch drives wouldn't be a good fit for consoles due to heat and size(maybe X1 could fit one because it's a huge console). I just hope we get Sata 3 controllers in the slim models of each console so i can just replace the slow mechanical drives with an SSD. I am guessing by the time the slim models come out 1TB entry-mid tier SSDs will cost around $200. Oh and trim support of course.

 

[BRiT]

Theoretically 3.5 inch drives wouldn't be a good fit for consoles due to heat and size(maybe X1 could fit one because it's a huge console). I just hope we get Sata 3 controllers in the slim models of each console so i can just replace the slow mechanical drives with an SSD. I am guessing by the time the slim models come out 1TB entry-mid tier SSDs will cost around $200. Oh and trim support of course.

Never going to happen.

The drive speed will always be limited/gated due to backwards compatibility and removing a different configuration from software testing. You will never get a remarably faster loading console after the first model had been out. That's not how consoles work and for good reason. Significantly faster loading would/could result in new bugs, so to prevent that they never make those changes. Even changes in memory systems are such that they perform exactly the same.

 

[Clukos]

Yeah, i know. Just wishful thinking. I'd imagine a fast SSD + Sata 3 would be a huge upgrade for open world games. Just imagine Bloodborne with 3 second load times

 

[Shifty Geezer]

Never going to happen.

The drive speed will always be limited/gated due to backwards compatibility and removing a different configuration from software testing. You will never get a remarably faster loading console after the first model had been out. That's not how consoles work and for good reason. Significantly faster loading would/could result in new bugs, so to prevent that they never make those changes. Even changes in memory systems are such that they perform exactly the same.

You can replace the internal HDD in PS3 & 4 with alternatives including SSDs and there are no bugs - only improved performance. The change in controller might never happen though as there's a need for two IO interfaces in FW then.

 

[DJ12]

Why are you "assuming" it's half a core when there is no suggestion of that?

Because thats what Microsoft did, Sony must do it, right. It's fanboy logic.

 

[function]

My assumption would be it's the full core. They don't have MS's legacy support for Kinect to factor in. Hopefully MS will follow suit. As market leader for consoles, more cores used on PS4 might be good for the PC too...

 

[MrFox]

Sony was planning a VR headset for PS4 since 2010, maybe they reserved the cores until they are sure of what they'll need. The OS interface will certainly have to be in VR I guess.

... or it's just a coincidence and this is an optimisation of the OS which allowed them to do release it. I still have questions about the stupid amount of memory they reserved.

 

[Shifty Geezer]

I think it's the reservations being re-evaluated, just as they were on PS3. As we discussed at launch, better to reserve too much (although there really was too much reserved!) and have the option to cut back later, than reserve too little and find you have a wall when later features are required.

Being cynical, we could also suggest Sony aren't planning that much by way of improvements given track record.

 

[bgroovy]

You can replace the internal HDD in PS3 & 4 with alternatives including SSDs and there are no bugs - only improved performance. The change in controller might never happen though as there's a need for two IO interfaces in FW then.

Doesn't the PS4 use a USB to SATA bridge chip? I would assume the OS never sees anything other than a USB drive and the actual interface for the drive itself could be changed at will. They could switch to IDE or SCSI of Fiber Channel if they want, as long as it can be bridged to USB.

 

[Shifty Geezer]

Doesn't the USB controller limit performance? I'm thinking PS4 is stuck with that USB controller and that bottlenecks things, but I don't recall if the USB impact was negligible or not.

 

[Allandor]

Theoretically 3.5 inch drives wouldn't be a good fit for consoles due to heat and size(maybe X1 could fit one because it's a huge console). I just hope we get Sata 3 controllers in the slim models of each console so i can just replace the slow mechanical drives with an SSD. I am guessing by the time the slim models come out 1TB entry-mid tier SSDs will cost around $200. Oh and trim support of course.

It is not the sata port that the bottleneck. Even sata 1 should be fast enough even for ssds. You won't see much improvement because of the sata standard used. Maybe in some cases, but those are really rare.

 

[Grall]

Doesn't the PS4 use a USB to SATA bridge chip?

The most current PS4 revision dropped the bridge IIRC. Performance is throttled to the same limit as older PS4s though, for compatibility reasons...

 

[BRiT]

The most current PS4 revision dropped the bridge IIRC. Performance is throttled to the same limit as older PS4s though, for compatibility reasons...

Yup, exactly as I said it would be for compatability reasons and to cut down on testing required.

 

[MrFox]

The same fujitsu bridge they were using was benchmarked at 270MB/s. That would mean the sata2 connection is the bandidth limiting factor, not the bridge. And the software throttling would be lower, as the stock HDD is nowhere near this number.