Tom's Hardware P4 ‘performs horribly’ in MPEG 4

The trouble with reviews: they're a moving target. New benchmarks come out, new bios revisions, patches, drivers, extensions, and there's always, always new tests to do.

Witness Dr Thomas Pabst, the most famous hardware reviewer of all. On Monday, Pentium 4 launch day, Tom published a comprehensive rundown of the new Intel CPU platform. Gazillions of specs, comparisons and benchmarks later, he gave the chip the thumbs up, albeit with caveats over price and early performance.

Today, he revisits his assessment, with an Important Pentium 4 Evaluation Update, following correspondence with reader Toby Hudon, expressing doubts over the validity of Tom's MPEG 4 benchmarks.

Essentially, Tom had used the default MMX settings of CPUs in his initial review, in which the Pentium 4 had outshone the rest of the gang. However, in real-life applications, CPUs would utilise "IEEE reference full precision FPU iDCT", to run MPEG 4 applications, Hudon points out.

Tom reran the tests using IEEE high quality iDCT with Flask MPEG. Guess what? The Pentium 4 is a dog. His tests show that it takes "12 hours for an average movie to be encoded at high quality with Pentium 4 and only seven hours with a fast Athlon processor".

New results means new conclusions. You can read them here.

Andrew Orlowski writes Now, wise heads will recall that the two new cores that Intel has introduced in the past ten years - the P5 in 1993 and the P6 (repackaged as Pentium II)in 1995, didn't notch up performance over their predecessors right away, and indeed not for a good few months. For example, the 486 saw a good two years of life while its sibling got into its stride.

But what might be worrying - and should be deeply troubling over at Satan Clara - is that the problems in the new P7 core are deep and structural, and won't be magicked away by manufacturing process improvements. For example, floating point performance is clearly underwhelming. And the deep, deep pipelines which Chipzilla bizarrely touts in some weird new pissing competition, only lend themselves to massive performance penalties in non-predictable tasks such as building software.

And these really are fundamentals. We haven't yet mentioned that SMP is officially Considered Harmful, or that for the next twelve months, the Pentium WhatIV will be tied to its deceased Siamese twin: Rambus memory.

We can without an excess irony describe Intel as a great, if not the greatest engineering company in the world. As it proved when it engineered its way out of being steamrollered by RISC. So two quick questions, guys. How did you get here? And how will you get out of this one? ®