Date:
Sunday , February 01, 2004
Author:
Kyle Bennett
Google +1

Prescott Arrives

Intel debuts their new Prescott core today by launching 4 new CPUs while also scaling legacy architectures. We benchmark them all.

continued...

Test Setup

Pentium 4 Extreme Edition 3.2GHz & 3.4GHz, Penitum 4 3.2C & 3.2E - Intel D875PBZ (i875P w/ .1012 Intel Inf Drivers ), 1GB (2x512MB) Kingston PC3500 HyperX DDR400 (2,2,2,5), ATI 9800XT (ATI Catalyst 3.8 drivers), 40GB Maxtor ATA133 HDD, Windows XP w/SP1 and DX9B.

Athlon64 3400+, A64 3000+ - MSI K8T Neo (VIA K8T800 w/ 4.51 Hyperion Drivers), 1GB (2x512MB) Kingston PC3500 HyperX DDR400 (2,3,2,5), ATI 9800XT (ATI Catalyst 3.8 drivers), 40GB Maxtor ATA133 HDD, Windows XP w/SP1 and DX9B.

AthlonFX-51 - Asus SK8V (VIA K8T800 w/ 4.51 Hyperion Drivers ), 1GB (2x512MB) Muskin DDR400 ECC Registered DDR400 (2,3,2,5), ATI 9800XT (ATI Catalyst 3.8 drivers), 40GB Maxtor ATA133 HDD, Windows XP w/SP1 and DX9B.

SiSoft Sandra Memory Bandwidth Benchmark

Article Image

Keep in mind that our Athlon64 CPUs use single channel memory, while the AthlonFX and Pentium 4 CPUs take advantage of dual channel memory subsystems.

We see all of the Pentium 4 systems come in where you would think they would, all inside of the realm of error. You might also notice our new FX-51 score. This is using the new Asus SK8V / VIA K8T800 chipset and Muskin DDR400 Registered DIMMs capable of running some very aggressive CAS2 timings. Overall, nothing surprising in the P4 camp.

dBpowerAMP Music Converter

Article Image

Considering this is our first real Prescott benchmark out of the gate, it might be a bit disheartening, but it actually is a very good way to start. Clock for clock in this specific benchmark, using what is probably the best known MP3 encoder to the enthusiast, we see the Prescott take it on the chin compared to the Northwood core. Looking at the Extreme Editions, or Gallatin cores, we see that everything else falls in line where it should be. The Prescott is still no slouch at encoding MP3s, it quite simply is not as good at it as the Northwood. This is a good example of a specific instance in which the Prescott core changes can have a negative impact on a specific application, but rest assured this is not the norm.

DVD2AVI DivX encoder w/ codec 5.1.1

Article Image

DVD2AVI has been our standard front end for DivX encoding for some time now. It is a widely used program and it is also freely available on the Net. For this set of benchmarks, we have moved to the newest version of the DivX codec, so these scores are NOT comparable to previous scores shown in our reviews. The above times represent an entire movie being encoded.

What we see above we do not like. Again we see the 3.2GHz Prescott fall behind its 3.2GHz Northwood brother. Not good at all. Looking at what the Extreme Editions bring to the table, we see that extra cache being put to good use as they almost catch AMD's single memory channel flagship. Still, across the board, the AMD CPUs defeat the Intel CPUs new and old.

XMPEG DivX encoder w/ codec 5.1.1

Article Image

We have added another widely used and freely available DVD encoding tool, and after you see the results you will know why. This tool also has the ability to process a specific number of frames, so the results are not comparable to the DVD2AVI results. The times above represent the duration of the encoding process for 10,000 frames of video to DivX. All scores represent the same 10,000 frames. (Note: If you want to try this at home, we suggest you do not use the tool's built in timer as it is not reliable.)

This encode is very different, as you can see. The Pentium 4s show victory for all CPUs across the board, and more importantly we see the Prescott not being beat out by the equally clocked Northwood.

Intel did have some insight as to why we were seeing things this way. Also keep in mind that this version of XMPEG was released months before the release of Prescott, so I would not think there has been Prescott specific coding going on. It seems that this particular front end program has been built with HyperThreading in mind and that the Intel core is being utilized to its fullest in this particular situation instead of decoding a frame and then encoding the frame with DivX. It is doing some of both at the same time.

If anything, these two DivX benchmarks really represent a bigger picture when it comes to CPUs and is something that we have been seeing in the VPU/GPU world for a while now. Raw synthetic benchmarks mean little to nothing anymore as it all depends on individual applications and how well they make use of the technology today. I think this will become more and more apparent over the next decade if the technologies continue to diverge the way they have in the last couple of years.