Drobe :: The archives
About Drobe | Contact | RSS | Twitter | Tech docs | Downloads | BBC Micro

Iyonix processor benchmarked

By Chris Williams. Published: 3rd Mar 2003, 02:11:09 | Permalink | Printable

Side by side StrongARM RiscPC and VirtualAcorn'ed PC. Now to find an Omega

The Castle Iyonix has been out for what, 4 months now? It's arguably the biggest hardware advancement for RISC OS since the Acorn StrongARM launch, harnessing the Intel XScale processor and nVidia GeForce2. It's had hype, it's had controversy but it's not been properly benchmarked by us. Yet.

Tony Howat dropped us an email to this evening to let us know that he's ported a 32 bit compatible version of Byte magazine's famous nbench benchmarking software, which is engineered to "expose the capabilities of a system's CPU, FPU, and memory system" according to Byte. It essentially runs a number of real world algorithms to determine the raw processing power of a computer. Tests on actual real applications can be found here, courtesy of Steffen Huber.

And oh look, Tony's been so kind as to test drive his port out on a range of RISC OS systems from a 242.8MHz StrongARM RiscPC to a 600MHz XScale Iyonix. And a 1.1GHz AMD powered PC too but try not to get distracted too much by that. Of course, we just can't wait to get this sucka running on an Omega, believe me when I say we can hardly contain ourselves. We also double checked the Iyonix results, running the benchmark software through drobe reader John Bell's Iyonix (cheers mate). Citing from Tony's website, in answer to your question "so how much faster is an Iyonix?", our survey said:

242.8MHz StrongARM RiscPC
Athlon 1800XP emulating ARM3 A5000
600MHz Intel XScale 80321 Iyonix
JMB's Iyonix
1.1GHz AMD Duron DDR based PC
Numeric Sort
String Sort
FP Emulation
Neural Net
LU Decomposition
Integer Index
FP Index

Numbers numbers!
So what does the above mean? Well, the indexes refer to how much faster the machine is compared to the "base machine", which is a Pentium 90 system with 16MB of RAM and a 256KB cache. So an index of 2.0 means the system is twice as fast as this base specification. We see that, in terms of integer processing power, the StrongARM system is just a touch over twice as fast as this base line system whereas the Iyonix is five times faster. All the RISC OS machines score very badly in the floating point (FP) department as they lack hardware to do the FP for them and it all has to be done in software. This was discussed recently on drobe.co.uk, in David McEwen's article about Quake on the Iyonix. Ultimately, we see the Iyonix scores aproximately 2.5 times faster than the StrongARM RiscPC.

We'd be idiotic to say these benchmarks were all you need to compare the Iyonix with other systems; all these tests are performed in memory and rely on the processor being able to number crunch lots - something obviously demanded of a spreadsheet, mathematical or vector/3D graphics software for example. What they don't test is hard disc interface and other hardware speeds, all of which are essential for a desktop machine.

We're not going to draw a conclusion even if we ought to because it's obvious what we would say anyway, (of course the Iyonix is faster than a RiscPC, duh). The facts are here, you can run it for yourself. One thing to note is that the benchmark software highlights the fatal lack of FP hardware in RISC OS machines.

In pictures
Also, we thought we'd just sneak in a mention of the Iyonix's appearance at the European HobbyTronic show, lurking in amongst the RiscPCs and C64s.


Tony's nBench software and results

Previous: Software news
Next: RISC OS Select 3 unveiled


Viewing threaded comments | View comments unthreaded, listed by date | Skip to the end

Comparing the 242.8MHz SA RPc with the 600MHz Iyonix one would expect the Iyonix to be 600/242.8 = 2.47117 times faster. On integer results it's 5.066/2.038 = 2.48577 times and on FP 0.096/0.042 = 2.285714 times. So it appears that, on average, performance scales approx linearly with MHz

 is a RISC OS Usermavhc on 3/3/03 8:18AM
[ Reply | Permalink | Report ]

Doing the same thing with the Iyonix and the 1.1GHz PC you'd expect a 1.83 times advantage in favour of the AMD machine. Instead we see that the PC has a 4.20 times advantage (187.6 times for FP). Although, I expect well draw level when the 113GHz Iyonix is released later this year ;-)

It'd be interesting to see stats for the current Apple computers as well.

All said and done though, most risc os users say they are more productive using a risc os machine rather than any of the competition. So, until there is a 4.5GHz PC to run an emulator it looks like we'll be sticking with the Iyonix -- Spriteman.

 is a RISC OS UserSpriteman on 3/3/03 9:19AM
[ Reply | Permalink | Report ]

I notice that MicroDigital have a floating point processor listed as an optional Omega upgrade. It's price is TBA, and it probably doesn't exist at all yet, but I think it's something MD should seriously concentrate on (as well as actually releasing the computer itself!)

Could an FPA be fitted to the Iyonix, somehow?

 is a RISC OS Usermonkeyson on 3/3/03 9:56AM
[ Reply | Permalink | Report ]

No. FPA in terms of modern ARM chips (and indeed, pretty much most processors) is integrated into the processor. I don't expect either machine to get FPA in the forseeable future unless a suitable FPA ARM chip arrives.

Tony should have noted that I provided the first column of Iyonix benchmarks (bad monkey!)

-- Peter, drobe.co.uk

 is a RISC OS Usermrchocky on 3/3/03 10:20AM
[ Reply | Permalink | Report ]

So an FPTwister isn't on the cards then? Something that turns FP add, multiply, divide instructions into MOVs or LDRs containing the results?

FPGAs have been used for FP - see [link]

 is a RISC OS Usermonkeyson on 3/3/03 11:44AM
[ Reply | Permalink | Report ]

Not that I'm speculating that this is what MD are doing. I'm just wondering whether such a thing is possible - maybe as a third party add-on.

 is a RISC OS Usermonkeyson on 3/3/03 11:49AM
[ Reply | Permalink | Report ]

Well, this benchmark shows all the strengths and none of the weaknesses of the Iyonix. There are things that actually run slower on the Iyonix than on the RiscPC. AFAIK some graphics-stuff can be up to five times slower (see www.riscos.de's news section, it's the thing about TrackMan2, a fractal generator).

-- Julian G. F. Zimmerle

 is a RISC OS UserJGZimmerle on 3/3/03 12:10PM
[ Reply | Permalink | Report ]

It just shows, how pointless such benchmarks are.

-- Julian G. F. Zimmerle

 is a RISC OS UserJGZimmerle on 3/3/03 12:12PM
[ Reply | Permalink | Report ]

I'm not convinced, sorry, the way you fob the benchmark off as "pointless" and then cite Iyonix problems. The tests show off a computers raw processing power (and nothing else, as we stated). So basically what you're saying is, Julian, you won't run the tests on your Omega?


Chris, drobe.co.uk

 is a RISC OS Userdiomus on 3/3/03 12:43PM
[ Reply | Permalink | Report ]

> you fob the benchmark off as "pointless" and then cite Iyonix problems

Actually he cites Iyonix problems and /then/ fobs off the benchmarks...

As a potential Omega owner (ditto Iyonix) I'd be interested to see some Omega benchmarks too - even by e-mail if it's not to go public yet. :)

 is a RISC OS Usermonkeyson on 3/3/03 1:05PM
[ Reply | Permalink | Report ]

Hey, what trimmed the (pedantic)(/pendantic} tags in my reply? :|

 is a RISC OS Usermonkeyson on 3/3/03 1:06PM
[ Reply | Permalink | Report ]

Julian: we appreciate these benchmarks are far from comprehensive. However, if wish to claim that the Omega is faster at some tasks, we would really appreciate being pointed at some tests/numbers that show how and when.

monkey: Yes, I suppose. My FPGA knowledge is 5 years out of date. My feeling is that such a solution might be rather slow - the diference between ARM interger and FP instructions behaviour is sufficiently large that an "FPTwister" solution would be way out of the question. There are other FPGA solutions too - but I think it would simply be far better to use a proper FP ARM chip.

Nor do I do think further propogating these benchmarks "privately" is helping anyone. Let's get them out in the open

-- Peter, drobe.co.uk

 is a RISC OS Usermrchocky on 3/3/03 1:08PM
[ Reply | Permalink | Report ]

I think Julian has a point about pointless benchmarks, the Iyonix comes out on top of a RiscPC on all tasks in the table above, but like he said, in some real world tasks the RiscPC is faster (although I expect there is a pretty good reason for this). I agree with Mr Chocky about getting some benchmarks for the Omega in the open, maybe just run BatchFSI on a load of JPEGs or something. On the benchmark point, my brother's 550MHz G4 laptop kills my 200MHz RiscPC in all tests we have run, but for simply viewing a JPEG, the RiscPC can be faster or the same. So really I think the only benchmark worth anything is the actual task you're running.

 is a RISC OS Userthegman on 3/3/03 1:38PM
[ Reply | Permalink | Report ]

Where can we find this TrackMan2 mentioned in the riscos.de news? I take it that it /is/ 32 bit and not running through Aemulor?

 is a RISC OS Usermonkeyson on 3/3/03 1:41PM
[ Reply | Permalink | Report ]

It would have been more interesting if you had compared the test with a A5000, a RiscPC 600 and a 600MHz PC like I did with my tests.

When I recently did my set of benchmarks which included the Iyonix pc there was a lot of people who wanted to know how it compared with their current machines.

Amazing how many RISC OS users are not even using a StrongARM machine!

Of course what we all want to know is why is it only 2.5X faster?

Peter Bonder at Acorn promised us a Phoebe ([link] for some of you youngsters) that would be twice as fast as a RiscPC. And that was with both of those machines using a StrongARM.

What went wrong with the Iyonix?

Unfinished Compiler? :-)

 is a RISC OS Userquatermass on 3/3/03 1:45PM
[ Reply | Permalink | Report ]

And of course MicroDigital have promised an SA Omega that is 2.5x-3x faster than an RPC...

Has anyone tried overclocking an Iyonix?

 is a RISC OS Usermonkeyson on 3/3/03 2:00PM
[ Reply | Permalink | Report ]

Nothing went wrong with the Iyonix. You're taking these results as fact that the Iyonix is only 2.5 times faster than the RiscPC. I'll add more fuel to the fire by showing you all the results I got from my Kinetic 266. Numeric sort - 2.06 String sort - 2.63 Bitfield - 3.27 FP Emulation - 3.25 Fourier - 0.13 (a real iyonix killer!) Assignment - 1.86 Idea - 3.47 Huffman - 1.77 Neural net - 0.08 LU Decomp - 0.08 Integer index - 2.527 FP index - 0.093


 is a RISC OS UserThe Doctor on 3/3/03 2:02PM
[ Reply | Permalink | Report ]

Trackman2 can be found in the downloads section on www.riscos.de. According to Detlef, it does run without Aemulor. I have not verified his results myself, nor have I had a closer look at the software.

@Chris: Sorry, I only just looked at my emails. I have just run nbench on the Omega and all results are very close to what would be expected.

-- Julian G. F. Zimmerle

 is a RISC OS UserJGZimmerle on 3/3/03 2:27PM
[ Reply | Permalink | Report ]

I've just run !Trackman2.Trackman through ARMalyser, and can confirm that it is 32-bit safe. :)

 is a RISC OS Usermonkeyson on 3/3/03 2:32PM
[ Reply | Permalink | Report ]

Okay, so you want some more "Omega numbers"? The VIDC emulation of the Lightning seems to be more than five times as fast as the original VIDC, plus 2D and 3D acceleration. And of course there is also the powerful DCT/iDCT unit of the northbridge chip, wich should give give JPEG and MPEG coding and decoding a boost. Unfourtunately the software to make use of these facilities is not included in my machine, so numbers on their performance will have to wait.

The mass storage interface is about twenty times faster than the RiscPC's (about 50 to 70 MBytes/s in normal desktop use, if you have a fast enough device connected to it).

More to follow, when I get round to it.

BTW, the Omega's PCI-documentation is on its HDD. :-)

-- Julian G. F. Zimmerle

 is a RISC OS UserJGZimmerle on 3/3/03 2:57PM
[ Reply | Permalink | Report ]

Sorry, thta should have been "more than five times as fast as the original VIDC20 of the RiscPC".

-- Julian G. F. Zimmerle

 is a RISC OS UserJGZimmerle on 3/3/03 3:00PM
[ Reply | Permalink | Report ]


 is a RISC OS Usermonkeyson on 3/3/03 3:00PM
[ Reply | Permalink | Report ]

Julian: Very interesting info, when you 'close to what is expected' does that mean that the Omega is 166% the speed of an Iyonix?

monkeyson: I concur, WOOT!

 is a RISC OS Userthegman on 3/3/03 3:33PM
[ Reply | Permalink | Report ]

nooooo, not the evil that is benchmarks.

cue the 'But look how much faster the PC is' posts. cue the 'I'm much faster using RISC OS' posts

cue the 'It burns my eyes' post from the flibble.

 is a RISC OS Userflibble on 3/3/03 3:49PM
[ Reply | Permalink | Report ]

It burns my eyes

 is a RISC OS Userflibble on 3/3/03 3:49PM
[ Reply | Permalink | Report ]

thegman: If you got 166% from 1000/600 (MHz), then I'm not sure if that's the case because the Omega doesn't have an XScale yet.

I've just tried Trackman2 on my Iyonix - now to try it on my RiscPC. Results to follow... -- Simon Wilson, Boulder, Colorado

 is a RISC OS Userksattic on 3/3/03 3:58PM
[ Reply | Permalink | Report ]

ksattic: I heard from somewhere, I've forgotten where, that the Omegas shipped had XScales, maybe Julian will comment. If they don't then I worry how long it'll be before buyers have them, somehow I think full release of the Omega is further away than I thought.

 is a RISC OS Userthegman on 3/3/03 4:46PM
[ Reply | Permalink | Report ]

If the ARMTwister stuff works it would be pointless to sell a StrongARM machine to try to compete against the Iyonix. My guess is they want to get everything working that they previously were going to leave until after launch so it's a decent competitor, which is what's causing the new delays.

Iyonix was obviously designed to beat Omega to market.

[duplicated comment removed - chris]

 is a RISC OS Usermavhc on 3/3/03 7:18PM
[ Reply | Permalink | Report ]

It was reported on Drobe (see Omega in Germany) than Julian does have an XScale but his machine has a preproduction motherboard.

-- Steve Knutson

 is a RISC OS Userknutson on 3/3/03 7:21PM
[ Reply | Permalink | Report ]

FPGA's aren't great at floating point but could still blow a software implemenatation out of the water. I'd like to see a reconfigurable co-processor tightly coupled to the ARM (like Garp for the MIPS) and a clever compiler which does some opcode-chaining and vectorisation. Thinking about it... does anyone know how good the compilers are for XScale? -- Rob

 is a RISC OS Userrob on 3/3/03 8:34PM
[ Reply | Permalink | Report ]

I can't agree with the Trackman2 results that Julian mentioned (though he has said that the results didn't come from him). I tried generating a similar fractal image a number of times using a Kinetic RiscPC (ViewFinder diasabled) and it rendered the image in approx 5.5 seconds. My Iyonix rendered the same in 4.0 seconds. Not the difference I would have expected, but faster all the same.

Looking at the code, it looks like it does direct writes to screen memory.

I don't know of a single thing that is slower on the Iyonix than on a RiscPC (except DMA IDE transfers with a card). Does anyone know of anything that I can test? -- Simon Wilson, Boulder, Colorado

 is a RISC OS Userksattic on 3/3/03 8:47PM
[ Reply | Permalink | Report ]

@ksattic: something must be wrong with both your RiscPC and your Iyonix, then. Detlefs Iyonix renders the fractal in about 0.42 seconds and his 202 MHz RiscPC does it in 0.16 seconds. My 233MHz-Kinetic-RiscPC can render it in 0.14 seconds.

-- Julian G. F. Zimmerle

 is a RISC OS UserJGZimmerle on 3/3/03 10:32PM
[ Reply | Permalink | Report ]

Julian - I zoomed into a particularly complex area on both machines.

It seems like you have a particularly past RiscPC! Are the times you quoted for the initial fractal screen? -- Simon Wilson, Boulder, Colorado

 is a RISC OS Userksattic on 3/3/03 11:06PM
[ Reply | Permalink | Report ]

Oh dear. I really should know better than to get into a benchmark battle, but...

The Trackman2 results intrigued me, so I had a quick look into it. Taking the program as it is, the top-level fractal renders in 0.16 seconds on my Risc PC and 0.42 seconds on my Iyonix. However, this program is unusual in that it does all its calculations in screen memory, constantly reading and writing screen memory bytes, which is not a sensible thing to do on an Iyonix since the screen memory is the other side of the PCI bus, and the graphics card is not optimised for screen memory reads.

I then updated the code to plot into a buffer in main RAM and only copy the results to the screen when it had finished. Result - the top-level fractal now takes 0.06 seconds to plot, and generally the plotting times now seem to be 2.5ish times faster than the Risc PC, which is what you'd expect given the difference in clock speeds.

What does this tell us? 1) Benchmarks will tell you anything you want them to. 2) A little time spent optimising (less than one hour) can yield large benefits (a 700% speed increase in this case). 3) Er, that's it.


P.S. Trackman2 is not 32bit safe. It contains three LDM Rx,{...,PC}^ instructions. I fixed these too.

 is a RISC OS Userjbyrne on 3/3/03 11:27PM
[ Reply | Permalink | Report ]

Even PCs don't have decent PCI/AGP video card read facilities (it's a software issue in Windows, not sure about Linux). Makes sense. :o) -- Simon Wilson, Boulder, Colorado

 is a RISC OS Userksattic on 3/3/03 11:30PM
[ Reply | Permalink | Report ]

2.5 times faster is 250% faster or 350% of the original speed but not 700% ???

 is a RISC OS Userdanielbarron on 4/3/03 10:05AM
[ Reply | Permalink | Report ]

Read it again. It took 0.42 seconds the first time and 0.06 after the re-write. Thats 700% increase. Cheers!

 is a RISC OS UserThe Doctor on 4/3/03 11:01AM
[ Reply | Permalink | Report ]

Incidently, on my genetically enhanced Kinetic the top level draws in 0.12. Thats a whole 0.02 faster than a standard Kinetic! Woot! Makes all that work seem worthwhie :-/ (Thinks: Need more speed!) Cheers!

 is a RISC OS UserThe Doctor on 4/3/03 11:44AM
[ Reply | Permalink | Report ]

On a RiscPC with a PC card fitted is it

[edited mangled comment, I suspect WXL strikes again - chris, 5/4/2003 1:39AM]

 is a RISC OS Userjlavallin on 5/3/03 12:19AM
[ Reply | Permalink | Report ]

Could someone send me the benchmarking software so I can compare an A7000+?


-- Andrew Harmsworth, Cambridge. www.gcse.com owner and author

 is a RISC OS Userharmsy on 5/3/03 10:31PM
[ Reply | Permalink | Report ]

There's a link to it at the bottom of this news article. The address is: [link] -- Simon Wilson, Boulder, Colorado

 is a RISC OS Userksattic on 5/3/03 11:28PM
[ Reply | Permalink | Report ]

Anyone tried the Iron Dignity demo on an Omega/Iyonix yet? That would give a more useful benchmark.

 is a RISC OS Userrpozz on 13/3/03 9:10PM
[ Reply | Permalink | Report ]

Please login before posting a comment. Use the form on the right to do so or create a free account.

Search the archives

Today's featured article

  • Let's publicise RISC OS in the Open Directory Project - follow our step by step guide of how you can help!

     5 comments, latest by piemmm on 22/5/03 8:00PM. Published: 20 Apr 2001

  • Random article

  • Omega XScale support in 2004?
    Didn't we ask this last year?
     49 comments, latest by rdenk14 on 24/05/04 10:09PM. Published: 20 May 2004

  • Useful links

    News and media:

    Top developers:
    RISCOS LtdRISC OS OpenMW SoftwareR-CompAdvantage SixVirtualAcorn

    CJE MicrosAPDLCastlea4X-AmpleLiquid SiliconWebmonster


    RISCOS.org.ukRISCOS.orgRISCOS.infoFilebaseChris Why's Acorn/RISC OS collectionNetSurf

    Non-RISC OS:
    The RegisterThe InquirerApple InsiderBBC NewsSky NewsGoogle Newsxkcddiodesign

    © 1999-2009 The Drobe Team. Some rights reserved, click here for more information
    Powered by MiniDrobeCMS, based on J4U | Statistics
    "As their reports are full of errors, [omissions] and mis-reporting - so readers you have been warned... unfortunately Drobe articles cannot be taken at face value"
    Page generated in 0.2651 seconds.