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

[3dilettante]

VGLeaks said Orbis is based on FreeBSD 9.0.

What file system does PS4 use ?

I'm not certain about the HDD, but the firmware USB stick must use FAT32.

 

[Bagel seed]

Edit: near the end he says something about offloading parts of the Audio pipeline to proprietary hardware & even external hardware. I wonder what he mean by that?

Proprietary hardware = Trueaudio
External hardware = VR headset?

 

[patsu]

This is the video I was talking about in my last post.


Edit: near the end he says something about offloading parts of the Audio pipeline to proprietary hardware & even external hardware. I wonder what he mean by that?

I'll have to rewatch the last part. The speaker is pretty dull. Lost my attention in the last third.

 

[Michellstar]

The CPU clock is 1.6GHz I seen it in the audio presentation & he said something about the CPU being 8 vectors at 1.6GHz vs the GPU being 16 vectors at 800MHz.

Don´t you mean 18??

 

[Lucid_Dreamer]

Presentation by SCEA dude.

Stream link, 1 hr (You'll need to register on their site first to view it): https://vts.inxpo.com/scripts/Server.nxp?LASCmd=AI:1;F:SF!42000&EventKey=115505

Or search for 'audio engine for hsa' and click on the 4085 link.

Hmmm. So, there is only around a 10% difference between a Jaguar core and i7 (clock for clock). He, also, said there is only around a 25% difference between Jaguar core and a SPU (clock for clock). SPUs are really powerful, when it can be taken advantage of!

Hopefully, with the fixed function hardware, Jaguar won't have much lower total performance than PS3's Cell.

 

[patsu]

Hmmm. So, there is only around a 10% difference between a Jaguar core and i7 (clock for clock). He, also, said there is only around a 25% difference between Jaguar core and a SPU (clock for clock). SPUs are really powerful, when it can be taken advantage of!

Yap and the Cell processor runs at twice the clock speed.

Hopefully, with the fixed function hardware, Jaguar won't have much lower total performance than PS3's Cell.

GPGPU and dedicated h/w.

 

[Aeoniss]

Yap and the Cell processor runs at twice the clock speed.



GPGPU and dedicated h/w.

I thought this was talked about and how, in practical terms, the jaguar CPU is significantly better all around than Cell?

CPU performance for this gen sounds really, really shaky...

 

[3dilettante]

This has been gone over, and the message hasn't changed.

There are certain things for which Cell is very, very, good at.
For most things, Jaguar is more robust, but it will have to rely on more specialized hardware to catch up in the areas it isn't.
If no such measures are present, it has to avoid that workload or eat the performance shortfall.

 

[taisui]

There are people who believe Cell's more powerful because it simply got more FLOPS than your desktop i3s.

Having the CPUs not be in order is already a significant improvement versus the PowerPC processors.

 

[patsu]

I thought this was talked about and how, in practical terms, the jaguar CPU is significantly better all around than Cell?

Jaguar is much more versatile. The presenter was talking specifically about an audio processing job. The SPUs are great for that task. Clockspeed aside, the speaker speculated that the 25% performance gap may be due to a specific SPU instruction.

This has been gone over, and the message hasn't changed.

There are certain things for which Cell is very, very, good at.
For most things, Jaguar is more robust, but it will have to rely on more specialized hardware to catch up in the areas it isn't.
If no such measures are present, it has to avoid that workload or eat the performance shortfall.

^Yay at what he said.

 

[pjbliverpool]

Hmmm. So, there is only around a 10% difference between a Jaguar core and i7 (clock for clock). He, also, said there is only around a 25% difference between Jaguar core and a SPU (clock for clock). SPUs are really powerful, when it can be taken advantage of!

Hopefully, with the fixed function hardware, Jaguar won't have much lower total performance than PS3's Cell.

I don't know the context of the statement or the workload involved but 10% difference clock for clock vs an i7 is certainly not representitive of the normal performance difference. It's not difficult to find benchmarks comparing Jaguar to low power IvyBridges around the net where the IB's are coming in at 2x or more performance per clock and per core.

 

[onQ]

Don´t you mean 18??

No the SIMD are 16 vector wide.

 

[TheWretched]

He was referring to the CU count.

 

[onQ]

He was referring to the CU count.

He said 8 Vector clocked at 1.6GHz vs 16 Vector clocked at 800MHz.

 

[tunafish]

He said 8 Vector clocked at 1.6GHz vs 16 Vector clocked at 800MHz.

Which isn't really right. The CPU has 4-wide vector units operating on 8-wide registers. The GPU has 16-wide vector units operating on 64-wide registers. There's no sensible way to come up with 8 vs 16 from that.

 

[almighty]

I don't know the context of the statement or the workload involved but 10% difference clock for clock vs an i7 is certainly not representitive of the normal performance difference. It's not difficult to find benchmarks comparing Jaguar to low power IvyBridges around the net where the IB's are coming in at 2x or more performance per clock and per core.

Indeed, that 10% number is a load of tosh and one who games on PC will tell you the same.

AMD are still a good 30-40% ( Sometimes even more ) slower then Haswell with there current FX CPU's.

 

[TheWretched]

What I meant to say is, he thought you were referring to CUs and wanted to correct you.

EDIT: And if IB comes out at 2x perf per clock and core, then with an 8 core versus a 4 core system, you'll end up at the same speed, if you break down the twice as many cores... at least to a certain degree, if your code allows for full SMP.

 

[pjbliverpool]

EDIT: And if IB comes out at 2x perf per clock and core, then with an 8 core versus a 4 core system, you'll end up at the same speed, if you break down the twice as many cores... at least to a certain degree, if your code allows for full SMP.

Except for the 2x clock advantage. Or where single threaded performance is important. Or where SIMD throughput is key where Haswell at least has a 4x advantage per core and per clock.

 

[Pixel]

Guys down at PC Perspective http://www.pcper.com/live/ did a live teardown of PS4

Their measements of the PS4 apu diesize came to the same dimensions that MS listed for Xbox one's apu die size.

I'm not at all surprised, thats to be expected. And thats what all the leaks suggested.

Given how densely packed sram transitors are compared to cu's xbox will still take i'm guessing a 0.5-0.8 billion lead in transistor count.

 

[dumbo11]

Someone's taken one apart - there are high-res versions under 'View original'.

http://www.ifixit.com/Teardown/PlayStation+4+Teardown/19493

The back of the board image shows a couple more chips (USB? and something power-related).