RISC OS news and information from Drobe


Beta! \| About us \| Contact \| RSS	Webspace \| Tech docs \| Downloads \| BBC Micro \| Gallery \| Wallpaper

We're dreaming of an Omega Christmas

By Chris Williams. Published: 27th Nov 2002, 02:13:08.

Omegad! Microdigital says Omega in production [Updated] ARMTwister explained

Well, what a turn of events. Until Castle revealed the Iyonix, their 32 bit 600Mhz XScale machine, Microdigital had been regularly updating the RISC OS community on the Omega's development progress, on their Newsdesk pages. The Newsdesk section of their website recently stopped being updated and people began fearing the worst; had Microdigital given up in the face of the Iyonix?

Never! And why would they? As analysed previously on drobe.co.uk, the Omega is a 306Mhz 26 bit RISC OS computer that boasts a punch of impressive features including optional XScale 2nd processor and Microdigital's Lightning+ video chipset. Although it's been a long time coming, the clouds have tonight parted and news has fallen unto the earth. According to a shock and sudden update on the Microdigital Newsdesk (see below for a link), the Omega is apparently in production and will be shipping in time for Christmas 2002. Also, 1Ghz 80200 XScale 2nd processor cards should be shipping in January and Microdigital's technical director is expected to publish an overview of their ARMTwister technology later this week. Microdigital also hinted at 1.2Ghz ARM10 processors.

Now, be aware that this announcement has knocked us for six. Microdigital, unlike Castle, have given us release dates which can only mean that their confidence is high. Here's to a very, very merry RISC OS Christmas this year.

Update
Microdigital have informed us that they've updated their Newsdesk pages to explain (simply) how their multiprocessor ARMTwister works.

Links
Omega ships for this Christmas

Discussion

Viewing threaded comments | View comments unthreaded, listed by date

Let battle commence!

32bit OS, hmmm

mavhc on 27/11/02 2:31AM [ Reply | Report ]

Chris, drobe.co.uk

diomus on 27/11/02 4:36AM [ Reply | Report ]

What with a 600MHz Xscale actually giving a realworld equivalent performance of a 300MHz StrongARM.

It looks like to be an interesting time ahead as Castle and MD compete.

quatermass on 27/11/02 9:23AM [ Reply | Report ]

Excellent to see MD pushing on with the Omega, particularly if they ship the 1Ghz chip in January, it'll give Castle something to think about. I'll be very interested to see what specialist applications can be made for the unified memory arcitecture, the SGI O2 has this, and found it's way into medical imaging and all kinds of niche areas.

thegman on 27/11/02 9:39AM [ Reply | Report ]

quatermass: FUD. The 600Mhz XScale behaves precisely as you'd expect a 600Mhz XScale to perform compared to a StrongARM (much faster) - which will become immediately clear if you've used an Iyonix for any length of time, or run any kind of benchmark.

While it's good news that MD will finally get the machine out the door (and drobe can do a proper comparison) to clear the air as it were, I am concerned at the "should"s and "talking to"s in MD's announcement. I would not hold your breath for either machine until you see them shipping.

mrchocky on 27/11/02 9:59AM [ Reply | Report ]

Saying that a 600MHz XScale actually gives a realworld equivalent performance of a 300MHz StrongARM is a lot of nonsense. The CPU itself, clocked twice as fast, will execute instructions twice as fast (more or less). It's also got a lot more cache.

Real world performance depends on lots of other factors however, and the Iyonix and Omega have completely different architectures, so how they will actually compare remains to be seen.

-- James Byrne

jbyrne on 27/11/02 10:03AM [ Reply | Report ]

Anyone have any idea if processors like a 1GHz XScale or a 1.2GHz ARM10 would be crippled by Omega's 133 SDRAM? Especially considering the bandwidth is shared with the video?

-- Sendu Bala

senduran on 27/11/02 10:46AM [ Reply | Report ]

Quite simply, they have to be. Yes.

ian on 27/11/02 11:54AM [ Reply | Report ]

Any design will cripple something. A ferrari car will cripple comfort while a ford car will cripple performance. I don't care if an XScale or ARM10 will becrippled by the 133 SDRAM as long as the performance gain is worth the money.

lss on 27/11/02 12:08PM [ Reply | Report ]

I think you've got the wrong end of the stick there, Leo. Omega's performance would be far higher if MD had put "proper" dual-port video RAM in Omega. They haven't, for whatever reason, so the graphics chip and CPU are in a constant fight for access to the RAM. You don't think Acorn put VRAM in the RiscPC just for a laugh, do you ?

ian on 27/11/02 2:13PM [ Reply | Report ]

Maybe my side off the stick is as nice as yours. If MD had put in "proper" (what is proper?) video RAM ,what would be the state of the Omega now? what would be the price? As I said before, designing a device always is a compromise and as long as you can live with the compromise there are no sides to the stick. And if you can not live with the device you should look for other options (may be you want to save and wait your whole live for that red ferrari (or do you want it silver grey?))

lss on 27/11/02 2:31PM [ Reply | Report ]

Proper = dual port. Your analogy about cars is void, because as far as video performance goes, Omega is worse than RiscPC and that was built 10 years ago! If it's worth spending 1300quid on an Omega, it surely should be at least AS GOOD as a 10-year-old RiscPC that can be picked up on eBay for 100quid now?

ian on 27/11/02 2:46PM [ Reply | Report ]

I think saying video is worse than the RiscPC without ever having used an Omega is a little premature (unless you are one of the few who have used one). Aside from cost there are good reasons to share memory, the Silicon Graphics O2 and O2+ which I mentioned above have UMA, and SGI knows how to make computers with good graphics.

thegman on 27/11/02 3:27PM [ Reply | Report ]

Bottom line of any of this must be that both machines are considerably quicker for the majority of tasks compared to even a StrongARM Kinetic RiscPC with Windfall graphics.

If not, spend the money on ice cream.

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

harmsy on 27/11/02 4:06PM [ Reply | Report ]

I believe Ian is refering to what he saw at the Scottish Omega presentation.

However, we'll only have to wait another month to see Microdigital's Lightning+ chipset in action in the end users' hands.

Chris, drobe.co.uk

diomus on 27/11/02 4:14PM [ Reply | Report ]

Yes, I've used an Omega. The video performance was awful, as is fully to be expected when the GPU is fighting with the CPU. As for "sharing memory" don't get confused between having separate (non-addressable) graphics ram and dual port video ram... the latter is what the RiscPC has, and for a very good reason! If you bother to do the math and see just how much bandwidth refreshing a measly 1280x1024x32bit screen @ 80Hz actually requires, you'll see it's an awful drain on the memory bus. If you can't understand why having the GPU hogging the memory bus gives a vastly underperformant machine, then you'd best step out of the discussion.

Anyway, stuff Lightning. It's an antique architecture. What I'm hoping for is that the rest of the machine (sound, PCI, UDMA etc) is stonking, and then we can add a decent PCI graphics card (ViewFinder 3 ? :-)

ian on 27/11/02 5:47PM [ Reply | Report ]

Wow, I had'nt realised what an emotional issue this for some people. I've never used an Omega, and I don't know the specifics of their system, but as they call the system they have 'UMA' and SGI has something called the same thing, I assume that they bear a passing resemblence. I have used SGI's UMA based O2, and this has excellent graphics performance with very large images compared to it's peers. Anyway I expect we'll see soon enough what the graphics performance is of this machine.

thegman on 27/11/02 9:29PM [ Reply | Report ]

Hm. Like I said. Step out of the discussion if you refuse to look at the facts. SGI/UMA is very different to Omega, it's four times faster and most importantly *ALL* the ram is dual-ported.

ian on 27/11/02 9:38PM [ Reply | Report ]

Depending on the application, both systems have advantages. E.g. for games which are written in 26bit mode, the Omega would be the fastest machine, after all it doesn't have run an emulator.

Competition is a good thing!

* This is of course pure speculation based since I haven't seen either machine and a lot of time passed so MD could have made significant improvements since the last public outting :-)

-- Steve Knutson Palmerston North

knutson on 27/11/02 9:51PM [ Reply | Report ]

Hmmm, the Xscale may be clocked faster and its got fast memory, etc. but its still much more inefficent than a StrongARM. Because a lot of the Xscale instructions take more clock cycles than a StrongARM to perform.

Which is why when PocketPC PDAs started using a 400MHz Xscale instead of a 200MHz StrongARM its users said it had gotten slower!

Hopefully the ARM10 will be better still.

quatermass on 27/11/02 10:53PM [ Reply | Report ]

Sure Stu, whatever. XScale instructions may take more cycles (but certainly not double), the processor is still twice as fast and there's the longer pipeline.

"This liquid fuel rocket is bigger and therefore slower than this little firework"

Sure.

Chris, drobe.co.uk

diomus on 27/11/02 11:30PM [ Reply | Report ]

Surely no-one is taking Microdigital seriously anymore? How many "in production" claims and release dates have they issued?

Please!! -- Michael Stubbs, Leeds

arenaman on 27/11/02 11:37PM [ Reply | Report ]

quatermass: more FUD. The PDA business relates to 400Mhz XScales and is irrelevant to the 600Mhz processor, which is clearly shown if you'd actually used one (as I said above), instead of spouting uninformed rubbish.

The slow PDAs are for real, but argument about efficiency is highly questionable.

mrchocky on 27/11/02 11:38PM [ Reply | Report ]

If we (drobe.co.uk) didn't think the announcement was serious then we wouldn't have published an article on it. As it happens, we're oh so excited that this machine is going to see the light of day at the end of this year, the year that saw RISC OS 4.33 and the XScale Iyonix. You should be more positive

And oh look, they've uploaded the mobo pics too. What more could you possibly want?

Chris, drobe.co.uk

diomus on 27/11/02 11:52PM [ Reply | Report ]

Ian, before you further upset yourself, can you please look at what I have said. I'm not disputing what you are saying, nor am I denying that dual-port RAM is better, I'm merely suggesting that we wait and see what the final result is from MD.

thegman on 28/11/02 12:26AM [ Reply | Report ]

Pictures of a mobo with /chips/ on it? I really don't think those mobos will do very much...

-- Sendu Bala,

senduran on 28/11/02 12:28AM [ Reply | Report ]

Just like to add to the little SGI/UMA snippet. Even with dual-ported RAM on the SGI O2, you still get a noticeable performance increase when you switch the graphics chipset off. Ian is quite right in pointing out that anything that uses main memory for graphics as well is going to suck ass when you want to use big modes, unless you have 10 million billion MHz RAM

.

-- Ian Hawkins (g0tai) [link] [link] [link] (etc)

piemmm on 28/11/02 9:22AM [ Reply | Report ]

Ever turned your hand to overclocking ram, g0tai? :-)

-- Spriteman.

Spriteman on 28/11/02 11:12AM [ Reply | Report ]

IIRC the XScale can have a higher clock rate since it is more deeply pipelined. If the pipeline never stalls then (best case) it will run twice as fast. Every time the pipeline does stall (branch/mispredicted branch) the impact is greater. When people talk about 'XScale optimised' I presume they mean pipeline scheduled by the compiler (to reduce the number of stalls). The latest Pentiums have a 20 stage pipeline which enables them to run at >2GHz... In order to get any performance benefit they rely on complicated branch prediction and will inevitably sometimes have to throw out 20 partially executed uops (think about this in terms of power efficiency). The Acorn scene has always ranted (correctly) that MHz is a pretty meaningless measure of computing power... so why is everyone so keen on it now? -- Rob

rob on 28/11/02 12:07PM [ Reply | Report ]

Because we're very fickle?

Chris, drobe.co.uk

diomus on 28/11/02 6:36PM [ Reply | Report ]

The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the The Omega I have seen and used at the

JGZimmerle on 30/11/02 10:30AM [ Reply | Report ]

Uuups. That was WebsterXL's fault. Oregano and Browse did not work at all for posting a comment.

-- Julian G. F. Zimmerle

JGZimmerle on 30/11/02 10:38AM [ Reply | Report ]

The Omega I have seen and used at the RISC OS Expo 2002 in the Netherlands certainly showed very good video performance. What else do you expect from a UMA where the CPU only needs a maximum of 252MB/s of the 1024MB/s memory-bandwidth. That leaves 772MB/s for the graphics.

If the Omega has a dual channel memory-controller it would even have a memory bandwidth of up to 2048MB/s.

If the Lightning has a 9MB cache it would only have to update its cache on changes (max. 30 times per second) and so would only need a maximum bandwidth of 270MB/s (if the complete content of the screen should change 30 times per second).

What I want to say: When the GPU is integrated with the chipset, there are lots of ways to minimise the bandwidth needed by the GPU.

-- Julian G. F. Zimmerle

JGZimmerle on 30/11/02 10:42AM [ Reply | Report ]

Ha! Now the official explanation is on MD-Newsdesk.

The Omega has 2nd-level-cache for the CPUs! :-)

-- Julian G. F. Zimmerle

JGZimmerle on 30/11/02 4:18PM [ Reply | Report ]

Thanks for the calculation of the bandwidth requirements. It takes me back to Hardware Design papers at University :-)

It reminds me why I prefer the C compiler to a soldering iron...

I am interested in a couple of "real" apps that could do with the extra grunt. Compare a 300Mhz RiscPC to an IYONIX compiling with GCC, creating an MP3 file and the screen rate in DOOM. I'd like to see the same for the OMEGA too.

-- Steve Knutson, New Zealand

knutson on 30/11/02 9:56PM [ Reply | Report ]

Yes, you're not the only one who wants a side by side comparasion. We want to do one, it'll be fun.

Chris, drobe.co.uk

diomus on 2/12/02 7:18AM [ Reply | Report ]

Erm, how many times has Omega gone into production according to MD?

I suspect that pretty soon Mr Atkins will announce a chip manufacturer has let them down or they've decided to make some improvements to the motherboard design, thus another few months will be required.

Well, the Omega seems pretty irrelevant to me. I'll just continue to enjoy reading the Microdigital NewsDesk pages - the best comedy around! -- Michael Stubbs, Leeds

arenaman on 3/12/02 8:18PM [ Reply | Report ]

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

Create a new account
Forgot your password?

Search this website