• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

(PS3Linux) Open Platform for PS3 - site open with instructions

BlueTsunami

there is joy in sucking dick
Panajev2001a said:
I personally think that we will not be stuck with a frame-buffer device for long and that we will have an OpenGL driver and Cg programming capabilities soon enough :).

:) :) :) I cannot wait for this day
 

Blimblim

The Inside Track
cedric69 said:
Ok, could anybody try to play a full screen 1080i video under Fedora and see what happens?
I'd wait for a better PS3 optimized version of Linux than a stock ppc64 kernel. As soon as I have my PS3 I'll install a gentoo with all ps3 stuff enabled. Though ffmpeg will only use Altivec stuff for the moment.
 

Blimblim

The Inside Track
kikonawa said:
Blim, when do you receive your ps3? still held at customs?
Hopefully tomorrow, more likely on wednesday. The customs called me on friday to tell me they were releasing the parcel, but that I'd have to pay 131 euros. Yikes.
 

gofreak

GAF's Bob Woodward
Blimblim said:
I'd wait for a better PS3 optimized version of Linux than a stock ppc64 kernel. As soon as I have my PS3 I'll install a gentoo with all ps3 stuff enabled. Though ffmpeg will only use Altivec stuff for the moment.

For anyone else who's also interested, there's some vague-ish instructions on installing gentoo here:

http://overlays.gentoo.org/dev/lu_zero/wiki/PlaystationNotes

Luca Barbato continues to document progress here, also:

http://planet.gentoo.org/developers/lu_zero.php

He doesn't have access to a PS3 so he's relying on others to test things. At the moment it seems they've an issue with gcc memory usage (though he should provide a way to avoid that shortly)..if you do go about installing gentoo, for now things may not be perfect.
 

antiloop

Member
Windows XP running on PS3 in Linux (emulated):

xp_ps3.png


I can't begin to imagine how sluggish XP is running in PS3 Linux with 200 MB ram. :lol

But still kinda cool.

http://www.beyond3d.com/forum/showpost.php?p=876894&postcount=30
http://rian.s26.xrea.com/
 

antiloop

Member
BlueTsunami said:
:lol

Is that even real? Jesus.

Looking at the screen it looks to be using QEMU for emulation...

Seems legit to me.

He did a Dhrystone benchmark too: (in Linux)

Dhrystone
自宅メインマシン: 3485.742 (His home PC)
PS3 Cell linux: 1361.608 (Cell - only the PPC core)
 

Blimblim

The Inside Track
antiloop said:
Seems legit to me.

He did a Dhrystone benchmark too: (in Linux)

Dhrystone
自宅メインマシン: 3485.742 (His home PC)
PS3 Cell linux: 1361.608 (Cell - only the PPC core)
Too bad he doesn't mention his PC specs.
 

cedric69

Member
Blimblim said:
I'd wait for a better PS3 optimized version of Linux than a stock ppc64 kernel. As soon as I have my PS3 I'll install a gentoo with all ps3 stuff enabled. Though ffmpeg will only use Altivec stuff for the moment.
Sure, but in case it would run smoothly... well, then we can give for granted that with a properly optimized version... media player supremo FTW!!! ;)
 

Blimblim

The Inside Track
BlueTsunami said:
Thanks Blim!

Yep what iapetus said.

I would guess, once you hit Decoding/Encoding and other clock heavy stuff, that PS3 is going too crawl.
Well, as long as there is no SPE involved yes, but once we get some code that uses them it will be quite different :)
 

gofreak

GAF's Bob Woodward
I'd be wary of comparing numbers when we don't know how he compiled and ran that. I'm not sure what the state of cell support is with the compiler in FC5 - without that you'll get a ppc build that doesn't account for its IOE nature etc.

Not that I'd expect much from the PPE in the first instance, nor do I know how much performance a targetted build would add, but some context would be nice.

BlueTsunami said:
I would guess, once you hit Decoding/Encoding and other clock heavy stuff, that PS3 is going too crawl.

It'll eat that for breakfast, if you code for the SPEs at least.
 

BlueTsunami

there is joy in sucking dick
gofreak said:
I'd be wary of comparing numbers when we don't know how he compiled and ran that. I'm not sure what the state of cell support is with the compiler in FC5 - without that you'll get a ppc build that doesn't account for its IOE nature etc.

Not that I'd expect much from the PPE in the first instance, nor do I know how much performance a targetted build would add, but some context would be nice.

On B3D patsu states that FC5 on the PS3 is very buggy.

gofreak said:
It'll eat that for breakfast, if you code for the SPEs at least.

Oh yeah :D For sure
 

antiloop

Member
If you want to compare with that list at least go with the not optimized numbers. ;)

I know that these score can vary a lot depending on OS and optimizations.
 

iapetus

Scary Euro Man
antiloop said:
If you want to compare with that list at least go with the not optimized numbers. ;)

Only if you're interested in Dhrystone performance. If you're interested in what the PS3 can actually run under Linux, then at the moment the optimised numbers make more sense IMO. Right now you can expect performance roughly equivalent to a 600MHz Duron without a 3D graphics card - that should give an idea of what emulators/video codecs you can expect to get results from.
 

gofreak

GAF's Bob Woodward
Does optimisation in that context refer to actual code modification or automatic compiler optimisation?
 

Queeg

Member
iapetus said:
Only if you're interested in Dhrystone performance. If you're interested in what the PS3 can actually run under Linux, then at the moment the optimised numbers make more sense IMO. Right now you can expect performance roughly equivalent to a 600MHz Duron without a 3D graphics card - that should give an idea of what emulators/video codecs you can expect to get results from.

While I doubt thats a fair indication of what it can do, as we have no idea what was run or how (I just did a Dhrystone test and it wouldn't make with register support so I got a figure of 6.6 million instead of the 10 million+ I should get), it's still around a 68030 @ 33 Mhz emulation speed for the Amiga IIRC :D
 

gofreak

GAF's Bob Woodward
Blimblim said:
Compiler of course.

I doubt there was an such optimisation here for Cell then (I don't think the version of gcc with FC5 can target cell specifically?) - and iapetus is correct that this will weigh on every application for the time being, but we should be talking about a matter of days here (I assume YDL's gcc will have cell-specifics). Again, not that I expect much from the PPE alone on a general benchmark, even with a cell build..
 

Blimblim

The Inside Track
gofreak said:
I doubt there was an such optimisation here for Cell then (I don't think the version of gcc with FC5 can target cell specifically?) - and iapetus is correct that this will weigh on every application for the time being, but we should be talking about a matter of days here (I assume YDL's gcc will have cell-specifics). Again, not that I expect much from the PPE alone on a general benchmark, even with a cell build..
Well of course this benchmark was only using the PPE. It's your average ppc64 architecture so I doubt some further optimizations on this part only would change much of the score.
It's all that's available for the moment for all Linux applications, until people hand optimize applications for SPEs. And it won't be a trivial task.
 

wsippel

Banned
Panajev2001a said:
I personally think that we will not be stuck with a frame-buffer device for long and that we will have an OpenGL driver and Cg programming capabilities soon enough :).
Why would Sony be interested in granting access to the GPU, if they took the measures and created the hypervisor to prevent access? And, even more important, why would Nvidia release drivers for RSX, even if Sony relases an updated hypervisor that removes the access restriction, especially considering that they only created the x86, amd64 and ia64 Linux drivers for professional users in the first place and never bothered to release Linux/ PPC drivers?
 

RuGalz

Member
gofreak said:
I doubt there was an such optimisation here for Cell then (I don't think the version of gcc with FC5 can target cell specifically?) - and iapetus is correct that this will weigh on every application for the time being, but we should be talking about a matter of days here (I assume YDL's gcc will have cell-specifics). Again, not that I expect much from the PPE alone on a general benchmark, even with a cell build..
gcc 4.3 (experimental) version supports cell architecture. That's what I'm planning to use to compile kernel and then other stuff once I have gentoo all installed. (Yes it's taking forever because I was actually compiling gcc and glibc on the system and was also hitting a few bumps. gcc 4.1 took like 2 hours to compile lol.) Anyhow, hopefully after it's all said and done it was worth the trouble.
 

wsippel

Banned
Blimblim said:
Well of course this benchmark was only using the PPE. It's your average ppc64 architecture so I doubt some further optimizations on this part only would change much of the score.
It's all that's available for the moment for all Linux applications, until people hand optimize applications for SPEs. And it won't be a trivial task.
Dhrystone is an integer-only benchmark, anyway, and CELL sucks at integer math. Big time. The SPEs are floating point processors, too, so they wouldn't help. Whetstone would be much more interesting...
 

Blimblim

The Inside Track
wsippel said:
Dhrystone is an integer-only benchmark, anyway, and CELL sucks at integer math. Big time. The SPEs are floating point processors, too, so they wouldn't help. Whetstone would be much more interesting...
I'll do this benchmark on wednesday then ;)
 

Panajev2001a

GAF's Pleasant Genius
wsippel said:
Why would Sony be interested in granting access to the GPU, if they took the measures and created the hypervisor to prevent access?

It is all related to the average selling price of shrimps at the fish market of St. Petersburg.

Seriously, the hypervisor does not ban all access to anything beyond running a calculator. You could have said the same things few weeks ago "why would Sony be interested in granting access to the SPE's, if they took the measures and created the hypervisor to prevent access?" following the same kind of reasoning you are presenting here, yet close to the metal access to the CELL Broadband Engine has been granted.

Besides GPU access, PLAYSTATION 3 is GREATLY more open than PlayStation 2 was even when you consider PlayStation 2 Linux: you are not limited to a Sony provided distribution (they have given people the tools to install the OS of their choice: if you can boot something, you can install it) and you have USB/HDD/optical drive access.

Also, giving an OpenGL driver does NOT mean opening up all specifications of RSX or releasing trade secrets or letting you code for RSX down to the metal.

Providing basic 2D and 3D acceleration would also be in their best interest not to utterly embarrass their Open Platform offering by staying stuck to a frame-buffer device dragging also CELL performance down.



And, even more important, why would Nvidia release drivers for RSX, even if Sony relases an updated hypervisor that removes the access restriction, especially considering that they only created the x86, amd64 and ia64 Linux drivers for professional users in the first place and never bothered to release Linux/ PPC drivers?

First, because the potential user-base would still be quite large.

Second, nVIDIA does not have to do all the work. I think they can allow access to the RSX and maybe the Game OS RSX driver (the Game OS IS running on the side or else how do you explain those 60+ MB of RAM unavailable to Fedora Core 5 ;)).

The Hypervisor is not there to ban access, but to control it which is not always exactly the same thing :).
 

cedric69

Member
iapetus said:
Only if you're interested in Dhrystone performance. If you're interested in what the PS3 can actually run under Linux, then at the moment the optimised numbers make more sense IMO. Right now you can expect performance roughly equivalent to a 600MHz Duron without a 3D graphics card - that should give an idea of what emulators/video codecs you can expect to get results from.
Are you referring to native Linux applications or the Windows emulation people was discussing before?
If you're talking about Linux, that's really depressing.
 

iapetus

Scary Euro Man
cedric69 said:
Are you referring to native Linux applications or the Windows emulation people was discussing before?
If you're talking about Linux, that's really depressing.

Native Linux applications that aren't optimised for CELL. And why is it particularly depressing? That's plenty to run most of the things I want - emulators for everything up to 16-bit and most of MAME at full screen full frame rate, a reasonable set of media apps and standard internet software.
 

cedric69

Member
iapetus said:
Native Linux applications that aren't optimised for CELL. And why is it particularly depressing? That's plenty to run most of the things I want - emulators for everything up to 16-bit and most of MAME at full screen full frame rate, a reasonable set of media apps and standard internet software.
Because it won't be able to do the single thing that XBMC can't do at the moment: HD video playback.
 
how in the world do i format my usb external HDD to Fat 32? I right click and select format, and it only gives me the option to format in NTFS... I tried booting from the windows CD, and the only thingt it'll let me do is partition the drive... it won't let me near formatting... Is there a third party app i can use? help!

edit: format Z: /FS:FAT32 /V:My_Label /X

voila! works... cmd line FTW.
 

gofreak

GAF's Bob Woodward
Blimblim said:
Well of course this benchmark was only using the PPE. It's your average ppc64 architecture so I doubt some further optimizations on this part only would change much of the score.

According to the guy working on the Gentoo overlay, that it is in-order alongside various other characteristics means you'll certainly be suffering a performance penalty using a generic ppc compile versus one targetting cell. How significant, I don't know..
 

wsippel

Banned
Panajev2001a said:
Seriously, the hypervisor does not ban all access to anything beyond running a calculator. You could have said the same things few weeks ago "why would Sony be interested in granting access to the SPE's, if they took the measures and created the hypervisor to prevent access?" following the same kind of reasoning you are presenting here, yet close to the metal access to the CELL Broadband Engine has been granted.
The thing is, the hypervisor doesn't restrict access to the SPEs, but it does restrict access to the GPU. That was obviously a design choice. Why should Sony change that after they decided to go that route in the first place?

Also, giving an OpenGL driver does NOT mean opening up all specifications of RSX or releasing trade secrets or letting you code for RSX down to the metal.
Depends. There are three ways to get an accellerated, full-featured OpenGL driver:

- Nvidia relases a driver. Highly unlikely.
- Nvidia releases the specifications and Sony removes the hypervisor restriction. Not in a thousand years.
- Sony removes the hypervisor restrictions and the community tries to create it's own drivers from scratch, without specs. Even with direct hardware access, nobody managed to create a working, full-featured driver for Nvidia GPUs on Linux in years - even though there were many projects trying and many people interested. There's still a project working on it, and they're making progress, but their driver is far from complete, let alone usable.

The best bet would be to simply ignore RSX and optimize the hell out of MESA on CELL. Not as fast as real GPU access, but probably good enough.

Providing basic 2D and 3D acceleration would also be in their best interest not to utterly embarrass their Open Platform offering by staying stuck to a frame-buffer device dragging also CELL performance down.
Basic 2D accelleration, XV and EXA, should be sufficient to keep the embarrassment to a minimum... :)

First, because the potential user-base would still be quite large.
Still, Nvidia would need Sonys permission, and an updated hypervisor. Even then I doubt it'll happen. PS3 Linux isn't suited for applications that would need OpenGL, anyway. It's mostly about "toys", emulators and such, and maybe an OpenGL-accellerated desktop. But there's still software rendering, so that would only be an option. It's not like you could or would like to run Softimage|XSI or Pro/E Wildfire on PS3 Linux, anyway. :D

Second, nVIDIA does not have to do all the work. I think they can allow access to the RSX and maybe the Game OS RSX driver (the Game OS IS running on the side or else how do you explain those 60+ MB of RAM unavailable to Fedora Core 5 ;)).
A driver running on GameOS with some sort of pseudo-driver for Linux that interfaces with the real driver? Sounds complicated, but might be possible.
 

Blimblim

The Inside Track
gofreak said:
According to the guy working on the Gentoo overlay, that it is in-order alongside various other characteristics means you'll certainly be suffering a performance penalty using a generic ppc compile versus one targetting cell. How significant, I don't know..
Does OOO really matter for a pure math performance benchmark?
 

Blimblim

The Inside Track
JeffDowns said:
how in the world do i format my usb external HDD to Fat 32? I right click and select format, and it only gives me the option to format in NTFS... I tried booting from the windows CD, and the only thingt it'll let me do is partition the drive... it won't let me near formatting... Is there a third party app i can use? help!
Start / settings / control panel / Admin tools (whatever the english name is) / computer management / hard drive
delete the partition and create it again.
wsippel said:
The thing is, the hypervisor doesn't restrict access to the SPEs, but it does restrict access to the GPU. That was obviously a design choice. Why should Sony change that after they decided to go that route in the first place?
As it been confirmed that RSX is blocked by the hypervisor? It could simply be a matter of not having the drivers for Xorg yet.
 

gofreak

GAF's Bob Woodward
Blimblim said:
Does OOO really matter for a pure math performance benchmark?

Maybe not, but I'm sure the comparison will be made..

BTW, if gcc 4.3 is the first version with cell support, what of the version YDL is shipping? IIRC it is an earlier version, but it does come with the Cell SDK.
 

Panajev2001a

GAF's Pleasant Genius
wsippel,

I do not see why they would need to modify the Hypervisor. In the documentation they basically said that they reverted to the frame-buffer device because they do not have a compatible driver+X server combination supporting RSX ready not that the Hypervisor was filtering any kind of access for Guest OS operation.

I think they might choose the slightly complex way of allowing RSX access through an abstraction layer made available to the Guest OS could be the best alternative: no extra work done by nVIDIA, no specification opened and no big security hole.
 

wsippel

Banned
Blimblim said:
As it been confirmed that RSX is blocked by the hypervisor? It could simply be a matter of not having the drivers for Xorg yet.
That's what I think after browsing the Sony documentations. Maybe I got it wrong. Someone quoted the passage in question in this very thread if I remember correctly.

And "simply a matter of having Xorg drivers" is quite optimistic, not only because you'll actually need two drivers, kernel and Xorg. There are none, and they would be extremely hard to write from scratch. Even after several years of hard work, there still is no open source driver for Nvidia GPUs with full hardware 3D accelleration. And Nvidia only releases drivers for Linux on x86, amd64 and ia64.
 

Blimblim

The Inside Track
wsippel said:
That's what I think after browsing the Sony documentations. Maybe I got it wrong. Someone quoted the passage in question in this very thread if I remember correctly.
Yeah that was me, but in another thread :)
Sony said:
Although the GPU is connected directly to CBE, no direct access by guest OSes to the GPU is allowed currently.
 
Top Bottom