Fba2x

[Squidge]

Discussed in another thread previously was this:

239 630 225 231 535 377 199 206 199 200 = 304 fps avg
188 227 188 198 182 178 182 188 200 179 = 191 fps avg
176 184 176 178 199 224 234 435 455 191 = 245 fps avg
207 196 197 187 237 207 187 173 169 187 = 194 fps avg
176 172 175 162 183 208 245 233 236 229 = 201 fps avg
575 310 202 201 197 196 197 208 202 203 = 249 fps avg
165 192 201 191 193 185 221 237 420 483 = 248 fps avg
190 207 197 196 187 216 199 204 185 170 = 195 fps avg
177 198 199 180 174 184 178 167 192 210 = 185 fps avg
247 283 605 195 209 197 196 195 212 223 = 256 fps avg
191 183 206 245 274 623 202 210 196 196 = 252 fps avg
195 203 199 207 209 219 239 394 510 188 = 256 fps avg
209 196 197 188 215 199 205 184 170 177 = 194 fps avg

Changing a few hardware registers now gives us this:

649 296 463 546 250 249 247 231 271 242 = 344 fps avg
229 212 223 235 231 211 223 215 221 252 = 225 fps avg
297 488 492 249 243 243 229 292 235 215 = 298 fps avg
206 222 209 212 199 222 269 295 276 280 = 239 fps avg
703 275 258 243 244 248 244 252 207 227 = 290 fps avg
251 236 222 272 281 683 296 257 243 244 = 298 fps avg
248 245 242 224 212 234 237 216 216 225 = 229 fps avg
200 233 271 283 637 331 257 243 243 254 = 295 fps avg
269 228 235 287 345 664 238 250 243 236 = 299 fps avg
261 246 248 271 282 670 309 258 243 244 = 303 fps avg
248 246 242 223 212 233 238 218 216 229 = 230 fps avg
205 234 273 281 652 315 257 244 243 256 = 296 fps avg
268 225 237 289 376 629 240 249 244 234 = 299 fps avg

Still not optimised yet for the omap

(For people who have not read the original: http://www.gp32x.de/board/index.php?showt...;p=579936&# )

 

[Peter R]

Awesome.

 

[fishybawb]

:blink:

The words "fucking", and "hell" come to mind. Nice work Squidge, this is going to be the handheld if this works out.

 

[fusion_power]

Interesting. At how many MHz was the test?
In line 8 there are 10 FPS more than before. (195->295)
In line 11 there are now 22 Frames less than before (252->230)
Not bad. :lol:

 

[fishybawb]

Even accounting for blips, that's an average of 228 vs 280 fps by my (possibly dodgy) calculations. Very impressive considering the lack of full optimisation

 

[Tobriand]

That's pretty amazing, really - although to be fair, we already knew the OMAP3430 would kick serious ass.

The next question of course, is what speed is it when it *is* optimised for the chip? And - in due time, presumably after release - how is optimisation different on Pandora to GP2x or GP32?

 

[Squidge]

Interesting. At how many MHz was the test?
In line 8 there are 10 FPS more than before. (195->295)
In line 11 there are now 22 Frames less than before (252->230)
Not bad. :lol:

Well, of course there are going to be differences like that. The FPS is simply output once a second for 10 seconds, then the average is taken. So if a different number of frames are executed in one line, every line afterwards will be different also as it will be running a different part of the game by then. You can't simply compare each lines average. The only thing you can really do is take the average of the first line (the first 10 seconds), and then take the average of all 13 lines and compare - 228 vs 280.

 

[Rivroner]

¿The tests have been taken at default Omap Speed:625mhz? Thank you

 

[fusion_power]

Interesting. At how many MHz was the test?
In line 8 there are 10 FPS more than before. (195->295)
In line 11 there are now 22 Frames less than before (252->230)
Not bad. :lol:

Well, of course there are going to be differences like that. The FPS is simply output once a second for 10 seconds, then the average is taken. So if a different number of frames are executed in one line, every line afterwards will be different also as it will be running a different part of the game by then. You can't simply compare each lines average. The only thing you can really do is take the average of the first line (the first 10 seconds), and then take the average of all 13 lines and compare - 228 vs 280.

Ah, OK, I didn't know that.

¿The tests have been taken at default Omap Speed:625mhz? Thank you

This sounds like we could underclock the Pandora around 300MHz and still have over 100FPS. That would result in save alot of Battery-Power -in theory-

 

[Pickle]

Squidge,

Any possabilit you can get some more numbers on other emu's?

 

[Squidge]

¿The tests have been taken at default Omap Speed:625mhz? Thank you

These ones? No. Neither 'batch' were run at 625Mhz.

Squidge,

Any possabilit you can get some more numbers on other emu's?

Sure, but I'd rather get an LCD working first.

 

[fusion_power]

Uh, is the used MHz value some kind of secret? :blink:
Everyone knows the OMAP specs so I can't see the reason. ^^"""

 

[Pickle]

¿The tests have been taken at default Omap Speed:625mhz? Thank you

These ones? No. Neither 'batch' were run at 625Mhz.

Squidge,

Any possabilit you can get some more numbers on other emu's?

Sure, but I'd rather get an LCD working first.


oh, i assumed you had more of the complete setup. A screen might be useful...

Did you build your own toolchain or did it come with TI's kernel? Are you even using TI kernel? Or do they just provide kernel patches and you built up eveything from scratch?

 

[Rivroner]

¿The tests have been taken at default Omap Speed:625mhz? Thank you

These ones? No. Neither 'batch' were run at 625Mhz.



So, they were run at ?

 

[Squidge]

Did you build your own toolchain

I'm using GCC for the tool chain, nothing more, nothing less.

So, the were run at ?

Two different frequencies

 

[atomicthumbs]

So hardware registers are used to change the frequencies...

hmm

 

[Squidge]

So hardware registers are used to change the frequencies...

hmm

and also on the GP2x, GP32 and lots of other ARM cores...

Your point being?

 

[javaJake]

What does Fba2x represent, by the way?

 

[Exophase]

So hardware registers are used to change the frequencies...

hmm

and also on the GP2x, GP32 and lots of other ARM cores...

Your point being?


I think his point is that it kinda goes without saying that changing the CPU frequency would make it benchmark faster, and it was already implied that you didn't do the first test at "maximum" clockspeed (actually, it was implied that you did it at well under the speed listed on openpandora.ca). And that by posting it and saying changing "some hardware registers" brought about the change you were suggesting that it was something less obvious, making it look better per-MHz than it previously was.

That just may be how I feel though. Seems a little misleading.

 

[Rivroner]

Did you build your own toolchain

So, the were run at ?

Two different frequencies


:huh: NO real answer? :rolleyes: