Handtops, UMPC - OQO Model 01+ v/s 02

Much has been made of the relative performance of the VIA C7M CPU v/s the Intel Pentium M/Core Solo CPU's, based on various benchmarks. I am not at all sure that this is of any great significance. The current crop of Intel CPU's are big, expensive, power hungry and hot compared to the VIA CPU's. The big question is – Is the OQO Model 02 any better than the Model 01+?

The answer to that question subjectively would seem to be an unequivocal YES. Every report I have come across to date on the Model 02 has been positive, with the performance described as “fast” and “snappy”. We all know the architecture of the Model 02 is much improved on that of the Model 01+, but how do the CPU's stack up in a “real world” performance test on a benchmark that is not optimized for an Intel CPU?

One of the oldest and most reliable benchmarks around that measures “real world” performance (at least of integer operations) and that can be ported from machine to machine without incurring architecture based errors in the results is the Prime Number generator. In the simplest terms, this is a program that finds all the prime numbers between 1 and maximum number, and measures how long it takes to perform the task. The faster the task is finished, the more efficient the computer.

I have a Prime Number benchmark that operates on this principle, and that essentially pre-dates the PC era. It measures the amount of time taken to find all the prime numbers between 1 and 1 million. A friend and I wrote it in BASIC to test the performance of his then brand new original Apple computer in 1979 or there about. Back then, it took a mere 27 or so hours to run (And stopped at 10000), but back then it was quite an achievement.

I re-wrote the program in GW basic for my first PC in 1982 (increasing the maximum number to 1 million), and then again in Turbo Pascal in 1984 or there about. The version I still use to this day was written using fundamentally unchanged core code with an improved windows interface in Delphi 1 back in 1996, and re-compiled in Delphi 7 for 32 bit windows two years ago. As you will see, the run time is down from 27 hours on the original Apple to seconds on the latest Laptops.

I sent the program to Kenrick, and he was was very kind and ran my Prime Number benchmark on this OQO Model 02. He also included for comparison the results from his tc1100. The results are shown below. Thank you Kenrick.

I have listed the results from my 01+ under XP and under XP Tablet. Just for the sake of interest, I have also listed the results from my HP Compaq nw8000 2 GHz Pentium M machine.

This benchmark has the unfortunate drawback of being limited to testing single core CPU's, but other than that, it has, in my experience over the years, given a very accurate reflection of actual CPU and memory systems performance. The test is affected by OS, which is also a “real world” factor.

Results (shorter time is better):

Original Apple 8 bit PC running embedded OS – 27 hours or so (1 to 10000).
OQO Model 01+ running XP - 7.050 Seconds (1 to 1 million)
OQO Model 01+ running XP Tablet - 7.631 Seconds (1 to 1 million)
OQO Model 02 running XP Tablet - 7.125 Seconds (1 to 1 million)
OQO Model 02 running Vista Business - 7.110 Seconds (1 to 1 million)
Sony PCG-U101 600Mhz Mobile Celeron with 1GB of RAM running Win XP Pro SP2 - 6.519 Seconds (1 to 1 million)
TC1100 Pentium M 1 GHz running Vista - 5.097 Seconds (1 to 1 million)
Sony UX running Vista - 3.094 Seconds (1 to 1 million)
HP Compaq nw8000 2 GHz Pentium M running XP – 2.103 Seconds (1 to 1 million)
Athlon 64 X2 3800+, 2GB RAM desktop (don't know OS) - 1.437 seconds (1 to 1 million)

As you can see from the above, the Tablet version of XP is slower on the Model 01+ than straight XP. As you can also see, the OQO Model 02 is a bit faster than the model 01+ running the Tablet version of XP. It is interesting to note that the Model 02 is fractionally slower running the Tablet version of XP than the Model 01+ is running straight XP. This would seem to indicate that the tablet version of XP is not the best OS for either the Model 01+ or the Model 02.

The Pentium M CPU powered HP is predictably faster, in line with other benchmarks. From this one can see the test seems to scale well.

The loading time of the test is not measured, and the graphics component is small, so it should have a minimal impact. Taking all the above into account, what can one deduce?

Well, the first thing is that the Model 01+ runs integer programs surprisingly well once they have actually loaded. The latencies that slow down the Model 01+ operations seem to come with loading the program etc. Once a program is loaded, it is run efficiently.

The second thing is that the Tablet version of XP causes a big hit in performance. If you are interested in squeezing the most performance out of your OQO, regardless of model, don't use the Tablet version of the OS.

The third thing is that the VIA C7M CPU in the Model 02 runs the test significantly better than the Model 01+ on the same OS. 0.5 of a second is fairly significant on this test, and represents between 50000 and 100000 prime numbers on the OQO.

A point to note is that the Pentium M CPU, clock speed for clock speed, is about 3 times more efficient than the VIA C7M CPU. This is more or less as expected.

What does all of this say about the OQO Model 02 and its VIA C7M CPU?

At first glance, it tells us that the VIA C7M CPU is no speed demon, although it performs a little better than the Transmeta Crusoe CPU in the Model 01+. But the VIA C7M gives us that extra performance while using less battery power, producing significantly less heat, and permitting a better architectural design. And this is where the Model 02 really scores over the Model 01+. The architecture supporting the CPU is so much better than that of the Model 01+, that even with a CPU that measures only a bit faster in a “real world” test, the whole machine seems to be a lot faster and more “snappy” to quote some of the reviews. Based on that, I would have to conclude that the Model 02 is a significant improvement overall (so long as one is not running XP tablet), and that it should be a success.

For those of you not as yet ready to ditch your trusty OQO Model 01+, take heart. While it may not be quite as good as the Model 02, it is still a decent performer once a program is loaded.

One final thought – VIA has produced a great little CPU that gets the job done without wasting power or producing too much heat. The much talked about and anticipated Intel UMPC CPU's that are expected this month, if clocked at 600 MHz as reported, are not going to perform any better than the VIA C7M CPU's at 1.5 GHz. OQO may well be looking to the Intel CPU for the Model 02+ or the Model 03, but for now, the Model 02 is looking good.

* EDIT - Kenrick was very kind and ran the test on Vista Business, I have included the results in the table. It seems to answer the question that everyone has been asking: What will be the penalty for running Vista? The answer to that would seem to be a little v/s XP Pro, and better than XP Tablet.

*EDIT - MeanSquare has also been very kind and taken the time to run the benchmark on his Sony UX loaded with Vista. Thank you MeanSquare. I have added the results to the table. They have confirmed once more the Sony UX's current position as performance champion of the Handtop world. Again, as a very CPU intensive test, the benchmark has scaled well, with the Sony UX completing the test in less than half the time taken by the OQO Model 02. With the Sony UX's faster sub-systems and drive, this should also translate to similar performance figures, until the OQO Model 02 is fitted with a SSD at least.

As MeanSquare was quick to point out, the Prime Number generator is not a good overall system benchmark, as it is extremely processor and memory intensive. I agree. It is a real world CPU/memory test however, and one of the few benchmarks that can be tested across different systems without the results being skewed by having to work on different architectures etc.

The above is not intended as a comparison between various different types of machines with various architectures. I am pleased that the other machine's results have been added, but this test is intended to help answer the primary question, is the Model 02 better than the Model 01+?

Anyone interested in a copy of this simple test is welcome to PM me and I will e-mail a copy to you.