Benchmarks
Pages: 1 2 3 4 5 6 7 8 9 10 11 12 13 ... 18
Trevor Johnson (329) 1645 posts |
There’s some code on the beagleboard.org mailing list for 1GHz, which also bypasses SmartReflex. I thought it might be of interest for comparison purposes. |
Trevor Johnson (329) 1645 posts |
Linux benchmarking thread, for reference. I’ve not investigated which language it’s written in, but perhaps it’s C++. If so, would there be any merit in trying to compile under RISC OS for comparison? |
Jeffrey Lee (213) 6048 posts |
If it is C/C++ then there’s not much point compiling it under RISC OS, because it relies heavily on floating point, for which we’re still using FPEmulator/softfloat. |
Trevor Johnson (329) 1645 posts |
Until VFP/NEON under GCC is ready? |
Jeffrey Lee (213) 6048 posts |
Correct. Hopefully I’ll be able to get back to working on that this weekend, as long as no more big SCSI/USB/etc. bugs pop up. |
Andrew Conroy (370) 740 posts |
Would it be a silly question to enquire if any progress has been made towards running at 1GHz ‘officially’ yet? |
Jeffrey Lee (213) 6048 posts |
That’s the next thing on my todo list; if all goes to plan I should be able to start work on it next week. |
Trevor Johnson (329) 1645 posts |
Micahel, here’s some data for you: Note: screenmode 1024x600 16M (TouchBook) FFNew_test CPU speed (MHz) Time (ms) Tot it Efficiency (it/ms/MHz) 125 17030 93781789 44.05 250 8440 93781789 44.45 500 4210 93781789 44.55 550 3820 93781789 44.64 600 3500 93781789 44.66 720 2920 93781789 44.61 FracVFP =======Single====== =======Double====== Single Double ========Single======= ========Double======= CPU speed (MHz) Time (ms) Tot it Time (ms) Tot it =Efficiency (it/ms/MHz)= Calc speed Disp speed Calc speed Disp speed 125 124100 177944574 14534 177936156 11.47 97.94 1433.881 1433 12242.752 1224 250 61550 177944574 7208 177936156 11.56 98.74 2891.057 2891 24685.926 2468 500 177944574 177936156 550 177944574 177936156 600 177944574 177936156 720 21270 177944574 2491 177936156 11.62 99.21 8365.988 8365 71431.616 7143 FracNEON CPU speed (MHz) Time (ms) Tot it Efficiency (it/ms/MHz) Calc speed Disp speed 125 8970 177944574 158.70 19837.745 19837 250 4450 177944574 159.95 39987.545 39987 500 2210 177944574 161.04 80517.907 80517 550 177944574 600 177944574 720 1540 177944574 160.48 11554.843 115548 Comparisons elsewhere won’t be ideal because I used the Touch Book (screen resolution stated above). Note that in ‘FracVFP’ and ‘FracNEON’ the displayed time figure is cs not ms – which confused me when I started looking at the figures and realised it’d taken much longer to calculate than the figures I’d tabulated. I’d have a look at your code for the factor of 10 speed issue… but I’m not ready for ARM assembler yet! |
Trevor Johnson (329) 1645 posts |
Chocolate teapots per second (CTPS), anyone (without OpenGL ES, of course)? |
Trevor Johnson (329) 1645 posts |
Does anyone have any thoughts about CoreMark? (Perhaps RISC OS doesn’t yet have all the necessary functions for this to be meaningful.) |
Jeffrey Lee (213) 6048 posts |
A quick skim over the CoreMark website suggests that they’ve designed it to be easily portable. I doubt it would take long to get it running. |
David R. Lane (77) 766 posts |
Just demo’ed the rotating cubes and Mandelbrot fractal programs on the BeagleBoard at SASAUG (Surrey and Sussex Acorn User Group) — they worked brilliantly. We used a resolution of 1280 × 1024 for the rotating cubes, the minimum resolution for which it works, but it was on a 1920 × 1080 screen. So the cubes were distorted. Has anyone got an mdf that includes a 16 to 9 resolution that will work for the cubes on a BB? |
Trevor Johnson (329) 1645 posts |
Cross-reference to The truth about VIDC, ViewFinder and VPod at c.s.a.hardware. BeagleBoard comparison results in Matthew Phillips’s post. |
rob andrews (112) 200 posts |
Hi any news about SmartReflex driver for xM or a new opp.bas Jeffery |
Jeffrey Lee (213) 6048 posts |
No news yet, but a SmartReflex driver is the next thing on my todo list once I sort out the USB/SCSI CD driver. |
Chris Gransden (337) 1207 posts |
These are the results for romark 1.01 running on a Pandaboard ES at 1200MHz.
|
Steve Revill (20) 1361 posts |
That’s nice, but what do the percentages mean? |
Jeffrey Lee (213) 6048 posts |
That’s the performance relative to the base system (202MHz StrongARM, RISC OS 4.02) |
Steve Revill (20) 1361 posts |
Ta. |
Chris Gransden (337) 1207 posts |
Looks like the CPU in the Pandaboard ES runs stable at 1.5GHz. So here are the romark 1.01 results,
|
Jeffrey Lee (213) 6048 posts |
You’ll all be glad to hear I finally got round to writing the SmartReflex driver for the BB-xM and adding 1GHz support. Much as you’d expect, some benchmarks I ran showed that CPU-heavy stuff now runs about 20-25% faster than it did at 800MHz. The driver is in CVS, so expect to see it in ROMs starting from tomorrow. Note that anyone who’s been using the ‘override’ entry in the CPUClk HAL device to set up custom OPPs/clock speeds will need to be aware that the SmartReflex driver uses a different table format to the non-SmartReflex driver. See the header and code for details. Also killing the Portable module won’t (yet) disable SmartReflex, so if you’re using your own code to set the clock speed manually you’ll have to deal with that. |
Rob Heaton (274) 515 posts |
Nice work! Other than replacing the ROM, is there anything else needed to enable the BB-xM to run a 1Ghz? |
Jeffrey Lee (213) 6048 posts |
Yes.
No. |
Andrew Rawnsley (492) 1445 posts |
Brilliant work, Jeffrey, thankyou :) This is the one thing that was missing! |
Chris Gransden (337) 1207 posts |
Here’s another comparison between the Beagleboard xM and Pandboard ES. The figures are from the SDL Quake 1 ‘timedemo demo1’ benchmark running fullscreen at a resolution of 800×600 and 640×480.
|
Pages: 1 2 3 4 5 6 7 8 9 10 11 12 13 ... 18