wbmw,
If you were to redesign the benchmark, what would you include?
do you agree that there may be such a thing as a meaningless benchmark? How about a diskcopy benchmark from one floppy to another, with one computer using the fastest Athlon, the other the fastest P4. Would you agree that this is a meaningless CPU test?
How do you thik they are serving these pages? Through Internet connection, as a user would? The differences between CPU would be in 1/100s of a percent at best. The content is just just sitting on a local hard disk, most likely, or they must be going through a network that is at least Fast Ethernet, with data sitting on a local server. This has nothing to do with the experience of the WWW, therefore calling it a webmark is meaningless.
Let me put it another way. I upgraded from Celeron 366 with 66 MHz SDRAM (the damn thing was not reliable at 550/100) to 1.33 GHz Athlon with PC-2100 DDR, and I can see the difference on CPU intensive tasks, but I see no difference while browsing the Web. The new processor is about 4x faster than the old one, memory bandwidth went up by the same margin. I use cable and DSL.
How can there be a 22% difference on more or less equally performing processors, when I can see no difference in processors so different? It is because just about all of the delay is caused by the internet latency, web server latency, bandwidth limit, not the client CPU. Coming up with a test that is more or less independent of CPU speed and calling it CPU Webmark test is a complete joke.
I have nothing against real life tests on which P4 actually beats the fastest Thunderbird. For example the Winstone Content Creation benchmark is not bad. Also, Flask DVD decoding / encoding is not a bad test. P4 wins both. On the other hand, CD ripping and encoding is just a theoretical test, since practically, the CPU's outrun the CD reader by a wide margin, even on the most complex encoding such as Monkey Audio.
Clearly, the applications used are real world applications that are popular, and highly used. The usage model was researched, and comes with a real understanding of the user environment, something that few other benchmarks can claim.
How about a CPU test of an analog modem dialer? Dialers are real world, popular, highly used, well researched, comes with real understanding of user environment. But let's add a twist to it. Doing it with actual modem will produce equal result for all CPUs. How about a twist. How about designing a modem simulator that will intercept the calls to the modem, and reply at speed of the CPU. Now let's call the dialing procedure a few million times, so that we get a result that is measurable in at least seconds.
Would you agree that this is a total joke? This is the concept behind Webmark, slightly exaggerated. Drawing any conclusions on what will happen in real life usage based on the results of WebMark or my modem dialer test are meaningless.
AMD Zealots obviously don't like the Bapco benchmarks because Intel outperforms in them,
The problem most people have with this organization is their conflict of interest. Hmm... I just noticed that Webmark is made my BAPCo. Interesting, wouldn't you say?
Essentially, in a multitasking environment, such as one the new Bapco benchmarks set up, very little can be considered repetitive, or predictable.
There was a benchmark suite that measures just that, it has been around before P4 release (and before BAPCo made "adjustments" to their suite to make P4 look good) In this benchmark, Athlon beats P4: anandtech.com
Bapco just picks apps, or tasks within apps that favor P4). This benchmarking business is not rocket science. It was a lot more difficult for Bapco to slant their benchmarks one way or another in the past, since Piii and Athlon were very similar. They could just tinker around the edges to perhaps de-emphesize FP and emphesize SSE.
P4 is drastically different than Athlon and Piii, it has obvious strengths and weaknesses, so it is a lot easier for someone with conflict of interest to design the benchmark suite to satisfy the sponsor.
Or, you can simply believe what you want. There will still be people on this board arguing that the Pentium 4 is a dud
I have not said it lately. Based on 1.4 GHz Thunderbird vs. 2 GHz P4 comparison. I think the dud comments were based on the past and some roadmaps. The difference today is that Intel executed well, and AMD did not. So today, I say P4 is not a dud comparatively speaking. If the AMD's mainstream processors today were 1.4 to 1.7 GHz Palominos, the story would have been different.
Joe |
