How fast is PS3 running Linux compares to PC?

[David Hwang]

As title. Has anyone tried to run a benchmark or do a comparison?

[goblinlordx]

:/... it is kind of difficult to compare apples and oranges. The PS3 processor isn't made to perform like a desktop CPU. If you benchmark it with a program that was meant for benchmarking desktop PC... it would probably show it is easily underperforming its competition in that manner. Though I think it would probably perform at least at a comparable rate to a processor around ~1-2GHz or so.

At the moment you probably won't even be able to benchmark the other processors. All the processors are RISC processors and have special sets opcodes that differ from normal processors. RISC (Reduced Instruction Set Computer[as the name infers it has less opcodes]) processors are designed to have opcodes that would take a normal CISC (Complex Instruction Set Computer) processor (such as a IA-32, IA-64, x64, etc.) multiple cycles (3+?) and complete them in one for the same task. This means in certain situations only one of the 6 processors running at 3.2 GHz would be completing tasks at what would need to be a rate of 9.6+GHz on any CISC processor (or even much more. There is also some differences because of the way desktops handle branching... but I am not even going to go into that :/.

So, if it were assigned a task that it was not meant for it may take several cycles more than a CISC processor. If a benchmark program were made to use the full potential of the RISC processors... it would absolutely destroy any benchmark done on a common desktop PC or even a dual or quad processor using current technology. My point is... they are designed completely different... you really can't compare them by normal means (such as a simple benchmarking tool... which would give absolutely stupid results and give people the terribly wrong idea).

[chris-007]

The german c't magazin tested the PS3 using a SPEC benchmark. The result was comparable to an Athlon with 1,33 GHz. So not really fast compared to todays machines: SPECint2000 about 400 (PPC Core, SPEs unused).

[_zaphod_]

You got it all wrong.

RISC means there aeLESS opcodes, and they take less cycles.
Whereas a CISC cpu has opodes tat take 3 or more cycles, a RISC cpu has more basic opcodes, which take only 1 or two cycles.

By making the ASM simpler poele get to figure out their own ways of rlling te other opcodes by hand in themost efficient manner they can.

In theory it's easier to optimize because of the lower number of possibilities.

[goblinlordx]

I never said there were more or even the same amount of opcodes... but I think that would be obvious from the name. Though... you are right... not opcodes... cycles... my mistake... its been a while since I looked at stuff like this.

[johnzbesko]

Upon boot, Linux systems calculate bogomips. Could someone post:

$ cat /proc/cpuinfo

[goblinlordx]

Upon boot, Linux systems calculate bogomips. Could someone post:

$ cat /proc/cpuinfo

>_>... again... you are talking about very one-dimensional statistics. They really won't tell you how fast it will or can run certain programs. I will post the info though in a while. Besides... lol... do you even know what bogomips are? If you did... you prolly wouldn't even ask about it.

[goblinlordx]

Upon boot, Linux systems calculate bogomips. Could someone post:

$ cat /proc/cpuinfo

Here is the info you wanted:

[ps3@localhost ~]$ cat /proc/cpuinfo
processor : 0
cpu : Cell Broadband Engine, altivec supported
clock : 3192.000000MHz
revision : 5.1 (pvr 0070 0501)

processor : 1
cpu : Cell Broadband Engine, altivec supported
clock : 3192.000000MHz
revision : 5.1 (pvr 0070 0501)

timebase : 79800000
machine : PS3PF
[ps3@localhost ~]$ cat /var/log/dmesg|grep "delay loop"
Calibrating delay loop... 158.72 BogoMIPS (lpj=317440)

[barf]

Only 2 processors are available in Linux?

[goblinlordx]

Only 2 processors are available in Linux?

=_=... this is why I hate posting stuff like this... it confuses people. First before anyone starts speculating... IGNORE the BogoMIPS... they mean absolutely nothing of any importance when it comes to running any programs.

No... all the other cores are accessable via the main core.
No... there are not 2 CBEs
That is the information the hypervisor is sending to the "otherOS" via virtualization.