Benchmarks?


cb88

The mighty inverse laplacinator
Joined
Apr 7, 2008
Messages
758
Location
I'm in ur loop
Website
gcadvsg.rr.nu
have any raw hardware benchmarks been done on the hardware?

Drhystone MIPS?
Whetstone MIPS?
bogo MIPS?!
http://freespace.virgin.net/roy.longbottom...e%20results.htm
SD interface proformance?

i doubt the SGX has been tested at all yet....

from what i read TI is claiming 1200 dhrystone MIPS for 600mhz which would place it alongside the PIII at 600mhz

if anybody has linux running on a MK0 how about a cat of /proc/cpuinfo that would be neat too....
 
Maybe you could say, PentiumIII at 600Mhz is simular to Cortex-A8 at 1050MHz. If you were only comparing DMIPS scores.

SGX 530 performance indications were provided from the SGX 535 figures released recently.
 
cb88 said:
have any raw hardware benchmarks been done on the hardware?

Drhystone MIPS?
Whetstone MIPS?
bogo MIPS?!
http://freespace.virgin.net/roy.longbottom...e%20results.htm
SD interface proformance?

i doubt the SGX has been tested at all yet....

from what i read TI is claiming 1200 dhrystone MIPS for 600mhz which would place it alongside the PIII at 600mhz

if anybody has linux running on a MK0 how about a cat of /proc/cpuinfo that would be neat too....



I am also interested in benchmarks. It would be nice if someone could lets say supply some source code. Then we can all run the same thing on various devices : GP2x, Nokia N800, Pandora and even desktops etc and compare what can we really expect.

Actually let us make a small repository. We can classify benchmarks into several categories :

a) Console based programs not using DSP or SGX. Basically simple stuff like matrix multiplication, some sorting benchmarks etc. This can be run on several platforms without problems.

b ) SDL based benchmarks. Useful for comparing basic game performance b/w GP2X, N800 and Pandora etc.
Again these will not use DSP or SGX.

c) Some specialized problems which require DSP or SGX. Will only run on the Pandora as far as the handhelds are concerned but can be compared with desktops.

Lets do a small shootout. We can specify problems and developers will submit a version optimized for their platform of choice (Gp2x, desktop linux, n800 or pandora). For each platform, anyone can submit any number of solutions. This will help people understand how to code for Pandora and also tell us what the platform is really worth.

If someone has Mk0 units and if they can also participate in some benchmarks before Pandora goes for pre-order, that will also be very helpful for people like myself who are as yet undecided.
 
Last edited by a moderator:
Before anyone dives into benchmarking, bear in mind that GCC doesn't produce ARM code that's anywhere as efficient as the x86 code it produces. The best performing programs for GP2X, for instance, all use some amount of assembly.
 
Exophase said:
Before anyone dives into benchmarking, bear in mind that GCC doesn't produce ARM code that's anywhere as efficient as the x86 code it produces. The best performing programs for GP2X, for instance, all use some amount of assembly.
Yes. So I think instead of "single source code" type of benchmark we can probably try "single problem multiple submissions" types? Thats not always a fair comparision but can still be fun :)
And plus it allows people to understand how to write faster code.

So for example, we can give setup routines and verification routines for a matrix multiplication problem. The actual multiplication can be left blank. Then people can submit any number of matrix multiplication samples.
Sort of like the "programming language shootout" benchmarks but instead of varying the language we are varying the platform with C with platform specific assembly allowed. So for example, on PC one might submit a SSE optimized version while for Pandora another dev might submit a NEON optimized version.

Even if code is not optimized for particular platforms, we can still gain "something" out of comparison of Pandora vs Gp2x, N800 or openmoko etc (if available) though not that much in comparion to x86.
 
Last edited by a moderator:
@icurafu I the numbers I have read projected 2000MIPS Dhrystone for the cortex-A8 i had thought that those where for the 1000mhz model... while the lower ones (1200 MIPS were our 600mhz model) perhaps the were projecting for an even faster clock?

this looks interesting they have optimised compilers for the 3530 ... and also a custom micro kernel (maybe even wifi support) that they say is royalty free... (BSD license perhaps?) might be a solution for emulators that need more system resources than can be had running a linux desktop without causeing too much of a pain to the dev?

actually there OS looks pretty interesting... heh personally I would find a port of it more interesting than a ubuntu port....LOL guess im a nerd....meh...

LOL forgot the link: http://www.ghs.com/products/TI_OMAP_development.html
 
where are you getting your PIII bechmarks? the ones i found at that link in the first post would place the pandora smack between a 600mhz PIII and a n athlon
 
cb88 said:
where are you getting your PIII bechmarks? the ones i found at that link in the first post would place the pandora smack between a 600mhz PIII and a n athlon
A P3 at the same clock speed is probably going to beat a Cortex-A8, not the other way around. Possibly by a good margin. Out of order, speculative execution and more executional units are in its favor.
 
Last edited by a moderator:
Back
Top