Benchmarks

Yes, a few are slower – updated above. Compare with the earlier ROM here

Interesting comparisons.

the benchmarks are expressed in percentages (where 100%=SA202 RiscPC running RISC OS 4.02)

But I don’t get why the percentages in the RiscPC column are therefore not all 100%.

And what do the figures for Rect Copy of 125% on the ARMX6 and 1178% on Iyonix mean? Shurley shome mishtake, Ed?

Shome mishtake: probably not: ‘rectangle copy’ is a measure of graphics performance. The iyo has a dedicated old-school graphics card and AFAIK Castle went to some lengths to optimise the RISC OS interface. The ARMX6 number is disappointing, but I wouldn’t be surprised if its much superior processor and memory speed more than makes up for it in actual use.

Rectangle copy is a graphics acceleration test – graphics acceleration is not yet implemented on the ARMX6 which means that screen scrolling in the command line interpreter (after pressing f12) is noticeably slow. Initially beagleboard and pandaboard did the same. Original BBC computers had hardware scrolling for the screen so you just wrote the new line over the line that scrolled away and reset the pointer to screen start.

In actual use, it is only the text scrolling after f12 that is noticeably slow.

Thanks, chaps, for clarifications.

And the RiscPC percentages?

The RISCOSMark programme defined 100% for a Risc PC some time ago. Many measurements have been made aginst that arbitrary standard. Later Risc PCs were a little faster (faster processor). The benchmarks highlighted in yellow depend on screen resolution – faster where there is less resolution/colour depth (so lower resolution would get closer to 100% for Risc PC – other computers are less dependent on screen resolution). Now comparison is using a fairly high resolution so these Risc PC figures are sub-100%. The storage speed benchmarks are all compared to a nominal hard disc on a Risc PC – my own Risc PC gets around 60% rather than 100% – don’t know why.

Benchmarks do vary a bit (other than pure processor or memory speed) so you should not give that much weight to them. Real world performance is what matters. A few tasks will be pure processing and the benchmark captures relative performance well here. Most are processor, memory, video, hard disc all mixed together which is why I quote ROM unpack and compile time.

I have just turned off the Iyonix and turned on the ARMX6 (I shall be migrating things over soon) and the silence is lovely…

I have just turned off the Iyonix and turned on the ARMX6 (I shall be migrating things over soon) and the silence is lovely…

Interesting: The comment made by Iyonix users when they got one brand new was how they were virtually silent. As a long time RPC user I thought I’d gone deaf, then I switched the RPC back on…

It’s just the difference been virtually silent and silent…

It’s just the difference been virtually silent and silent…

Given the after effects of the 70’s concerts¹, probably about 10 years matching the receding hairline etc.

¹ Deep Purple, Nazareth, Quo, Ian Gillan Band, Uriah Heep, Rainbow, Hawkwind², Nils Lofgren & Tom Petty (same bill) and the “quiet” ones like Pink Floyd, Yes, Genesis, Camel

² It rained³ indoors in the Birmingham town hall

³ Look it was wet and condensing in the cloudy haze up above us.

Lovely how a thread can morph from performance into performances at the touch of a button :)

Thanks Chris for further enlightenment on benchmarking the shared RISC OS history.

The comment made by Iyonix users when they got one brand new was how they were virtually silent.

When I bought mine (brand new) virtually silent isn’t a term I recall coming instantly to mind.

It isn’t a term I recall coming ever to mind!

Mine has always been quite noisy, and the only time I’ve not noticed it is when I used to have it set up at shows, because all the background noise drowned it out.

The RISCOSMark programme defined 100% for a Risc PC some time ago.

I have posted a PandaVersion but it was ignored. ;-)

When I bought mine (brand new) virtually silent isn’t a term I recall coming instantly to mind.

Early IYONIXes were very quiet – the only fan was the PSU fan, and the early HEC units had a good quality fan. When new, harddiscs are also typically low-noise.

Later IYONIXes were very noisy because of the graphics card fan.

One of my IYONIXes is completely silent because it has an Amacrox Calmer fanless PSU and an SSD.

My Omega, on the other hand, has a big, non-regulated 12cm case fan that sounds like a military jet with afterburner.

Early IYONIXes were very quiet

Oh, mine was early. Very limited numbers at the Motorcycle museum.

Have updated the benchmarks to include ROM compile speed on the ARMX6 with a 500Mbyte RAM disc and to bring ARMX6 and Panda up to date with the latest ROMs. The benchmarks may be found here (click refresh in your browser – the updated figures have a light green background).

I have updated the RISCOSMark test programme here to include a display of file system and machine revision so that the context can be inferred from the output (useful when you are testing several machines).

Have updated the IGEPv5 benchmarks (running at 1000MHz) and added benchmarks for the OMAP5432EVM (running at 1500MHz). Here – note the gigabit networking on i.MX6 and OMAP5 are limited to 480Mbit/s.

Its not comparing the same thing though. IGEPV5 has the EtherUSB changes that appeared in September so all the other RO5 machines that use usb networking aren’t using the same EtherUSB.

In your sharefs tests do any of the machines in the test set ‘sharefswindow’? The speed tests will depend on this setting. The default is 2, but setting it to 4 on both machines in the test with the new EtherUSB can give an improvement.

Are they supposed to be OS benchmarks? If so shouldn’t LanMan be used instead of LanMan98?

The ARMX6 has the latest OS update from RComp and uses EtherTH 0.22 (30-Sep-2015). The IGEPv5 has the latest ROM from CJE Micros. The Pandaboard has the latest OS update from RComp (another update is in the pipeline but has not yet emerged) and uses EtherUSB 0.28 (24-Aug-2014). The Raspberry Pi has the latest RC14 release. The Iyonix has the latest OS that did not break things (5.18 broke !Writer due to a SCL fix and ROOL advice was to stop upgrading when things broke). ARMini is a bit behind though, I agree – however I may leave that as it is because I am trying to show how things have changed rather than what software that was not available when the machines were generally used could do now.

The problem with LanManFS (rather than LanMan98) was that until (very) recently it could not log on to my NAS. Sharefs to a VRPC was problematical, particularly for ARMX6. I don’t use sharefswindow at all.

From my point of view it would be more useful to see how the betas are progressing and if there are any problems. For example I improved EtherUSB after seeing your benchmarks on the PI showing slow networking. It may be that it wasn’t improved everywhere and your benchmarks don’t show that. ARMX6 and IGEPV5 are getting beta updates so why not the rest – just a thought.

Again with LanManFS, now that it should work if a machine doesn’t have a version that works it would highlight the fact and maybe someone will do something about it.

I just think it would be useful to see the state of play at the moment.

It all takes time – I’ll see what I can do.

I’ve updated benchmarks for IGEPv5 for the latest 1500MHz SmartReflex ROM.

Screen speed looks excellent.

Interesting that the vastly superior Memory speed compared to the ARMX6 doesn’t translate to better RAMFS performance.

USB networking isn’t too shabby when compared to the dedicated Ethernet of the ARMX6. Networking performance is dependant on the performance of both sides of the connection, in the lanman98 case you state that it is to a NAS, which machine do you connect to in the ShareFS test?

I’d be interested to know the ShareFS HD Read/Write speed between armx6 and igepv5 with ‘sharefswindow 8’ set on both machines. Fastest read I’ve seen was on a standard pi b+ to a beagleboard and that was 5.66MB/s. Read/Writes are generally 4.5 – 5MB/s when the remote share is in RAMFS – Just trying to get a feel for what ShareFS maxes out at.

USB networking isn’t too shabby when compared to the dedicated Ethernet of the ARMX6

Unfortuntaely the gigabit ethernet on ARMX6 is limited to 480Mbits/sec due to chip limitations (whatever that means).

For the sharefs test I connect to an ARMX6. Will do some more tests tomorrow.

Interesting that the vastly superior Memory speed compared to the ARMX6 doesn’t translate to better RAMFS performance.

Yes that is a bit strange!

AIUI having pages that are cacheable can increase the speed significantly! One to look into.

Re network speed. I think the limiting factor is no longer the hardware but software. IIRC someone mentioned that some networking is actioned by the centisecond timer which is aeons for an A-9 or A-15 CPU

Thanks Chris for new benchmarks.

The networking on the X6 (for inbound) does need some work from a driver level I believe.
I can get a solid 11MB/sec on mine, under the wimp using wimppoll
Using a centi or milli second timer is probably my comments, they were based on usage rather than any knowledge (i.e. I could be misunderstanding/misusing something – entirely possible)