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

edit: Nevermind, outdated article.

 

I find it a bit bizarre that they went with an SATA -> USB 3.0 bridge rather than just have the SATA controller hosted on the APU. Perhaps they originally planned that but something went wrong with SATA controller on the APU? I can't imagine it would be cheaper to add a bridge chip versus a small amount of APU silicon.

Which is interestingly enough one area where an SSD could have helped level loads a bit more if read speeds weren't capped by the SATA -> USB 3.0 bridge.

BTW - even level loads will generally show some benefit from lower random read speeds unless either [1] all levels have unique data or [2] shared data is replicated multiple times (wasteful) in order to keep data contiguous for sequential reading during level loads.

Regards,
SB

 

Perhaps Sony wanted a consistent experience with every console - basically they don't want massive speed and latency jumps when people move to SSD's, to provide a uniform experience for everyone?

 

Final PS4 CPU clock speed is 1.8ghz?

http://pc.watch.impress.co.jp/docs/column/kaigai/20131029_621310.html

 

Where ? I don't see it.

 

2.0GHz...

 

Nothing confirmed, that's just their guess/estimation.

 

Nope 1.8

Must be a .. mistake?

http://pc.watch.impress.co.jp/img/pcw/docs/621/310/html/09.png.html

 

We don't know what the clock speed is, and we may never know.

 

That'd be rather stupid. Everyone gets the same base experience when they buy the machine. If others want to invest in improving it, that's there choice. Capping the IO speed would be similar to capping the resolution to 720p regardless if some people have 1080p TVs, to give everyone the same experience.

The choice for a USB bridge will likely come down to cost one way or another.

 

I would think so, too.
There are other mistakes. They group the CUs in 5 groups of 4. This is clearly wrong. From the die shot (Chipsworks) I would say Orbis has two arrays/engines (however AMD call these today, cannot remember) containing 4 + 3 + 3 CUs per group. Pitcairn had four times 3+2 CUs. That's part of the reason for the changed layout of the shader/TMU part of Orbis compared to Pitcairn.

 

Me either.

 

Yup, on the slide where it clearly says "estimate".

 

The block diagram may be wrong but are all the numbers in the diagram accurate ? (Besides the CPU clock speed and derived info).

 

The Fujitsu bridge chip is marketed for its "high-speed encryption hardware support", which is a likely reason for Sony using it.

 

With my own tests, I got about the same 10s SSD versus 20s HDD for Resogun.
Knack however doesn't show much loading time advantage (when manually loading a level out of order).

Resogun is 466MB on disk, loading it all in memory should take less than 5 seconds on a normal HDD, or maybe 2 seconds on an SSD. The load time difference cannot be attributed to bandwidth. I can hear the drive trashing when loading it, so the advantage is probably caused by the SSD seek time.

Knack doesn't really seek much it sounds like it read a big linear blob.

 

I was watching the archive of Giant Bomb's PS4 stream and it seemed that the UI would get really laggy when you were trying to install, play and download stuff all at once. It would be interesting to see if an SSD would improve the experience in those situations where the contention for the HDD is getting bad. That's harder to test, of course.

 

I've seen reports at various places of PS4s crashing under those conditions, so lagging isn't so bad considering the alternative negative behaviors.

I'm sure given proper amount of time to focus on the issues, they'll have things sorted out. It shouldn't be laggy because of drive access or seeks.

EDIT: I just remembered something, perhaps they're using some of that 'virtual memory' that is paged to disk?

 

Ah, that would make sense. If MS do that I expect they'll do it the same way they do it with hardware storage encryption on PC, which doesn't require a bridge chip.

Regards,
SB