..:: SiSoft SANDRA ::..
Well, now that we have covered the technical aspects of the “Prescott” processor, now it’s time to put all of the performance rumors to bed. We’ve heard countless statements flying around about how “Prescott” was “much slower” than an equally clocked “Northwood” chip, and so forth. To start off the benchmarks for today, we’ll begin with the usual software developed by SiSoftware, SANDRA. In the arithmetic benchmark, we can clearly see the result that the increased pipeline length has placed upon the performance of “Prescott.” All of these benchmarks were conducted with Hyper-Threading Technology enabled, and from the results we are seeing above, “Northwood” is capable of more computations than is “Prescott,” as we expected due to the pipeline increase.
Next up, we have the results from both the multimedia benchmark, and the memory bandwidth benchmark. For the multimedia benchmark, we see that once again the “Prescott” processor suffers a sizeable cutback in performance when compared against an equally clocked “Northwood” processor. Whether or not the other advantages of “Prescott” will be able to make up for this loss is yet to be seen. When we get to the memory performance benchmark, we finally see some advancement in performance due to chip and architectural changes. Unlike in the arithmetic and multimedia performance results, the memory bandwidth for “Prescott” has increased by a sizeable margin. This will further aid “Prescott” in being able to counter act the lengthened pipeline.
..:: MBReview PriBench ::..
Using our very own in-house benchmark, MBReview PriBench uses a computationally intensive algorithm to compute several hundred millions of prime integers. Unlike other synthetic benchmarks which rely on additional components within a computer system like the hard drive, PriBench does not rely on any of these exterior devices for the tests. PriBench is a system level benchmark, solely utilizing the performance of the processor and memory subsystem. It also focuses on comparing processor architectures and technologies and then seeing how much a core clock and/or FSB speed increase can be beneficial to the system. Our program is extremely accurate as we have witnessed time fluctuations of approximately .05 – .10 seconds.
Taking a look at the results, we can see that due to the various enhancements that have been made to the “Prescott” Pentium 4, it is easily able to wipe the floor with “Northwood” in each of the three varying intensity benchmarks. In the highest intensity calculations, we see a performance increase of between twenty five and thirty percent! PriBench is best at showing off the ALU and FPU performance of processors, and with the enhancements that Intel has made to “Prescott” in this area, we are seeing some large overall gains in computation capabilities. The question is, will this hold up in other benchmarks?
..:: SuperPI ::..
In the SuperPI tests, we run the program four times, once at one million, once at two million, once at four million, and you guessed it once at eight million. The numbers in the graph above show the time in seconds that it took the system to calculate pi to the set number of digits. In this benchmark, we once again see the effects of the enhancements that have been made. In the lower level benchmarks, there isn’t much of a performance difference, although when we proceed up the scale to more intensive calculations, “Prescott” begins to shine over the previous “Northwood” incarnation. The performance differences we are seeing here border on the 10% line at maximum, so they aren’t quite as high as those we received with our PriBench software, nonetheless they further exemplify the improvements between “Prescott” and “Northwood.”
..:: SPECviewperf 7.0 ::..
In the SPECviewperf 7.0 benchmarks, we received some dispersion of results. In each of the recorded benchmarks, the “Prescott” processor is able to overtake “Northwood” at 3.20GHz, but things get a little tipsy when it comes to comparing the 3.00GHz models. In the Light05 benchmark, we see that the 3.00GHz “Prescott” processor actually comes in slower than that of the “Northwood.” In each of the other benchmarks that we tested, “Prescott” was able to hold onto a fair lead, averaging between five and ten percentage points, depending on the benchmark in question.
..:: ScienceMark 2.0 ::..
For our ScienceMark 2.0 benchmarks, we decided to take a look at the performance results achieved with three of the built-in programs, those being MemBench, Cipher AES, and MolDyn. We’ll start off with the MemBench results. Much as what we achieved in SANDRA, MemBench once again shows the performance enhancements of “Prescott” over the “Northwood” processor, as “Prescott” is able to provide a higher level of bandwidth, lowered latencies, and lower necessary clock cycles.
When we moved on to the next benchmark in the ScienceMark 2.0 suite, Cipher AES, we began to see a downturn in the performance of “Prescott.” In the Cipher AES benchmark, the “Northwood” processor was able to best “Prescott” in computation time by roughly three and a half percent. These results were repeated when we moved on to the MolDyn benchmark. In MolDyn, “Northwood” is once again able to easily best “Prescott,” this time by a much larger margin of nearly fourteen percent. All that was looking so well for the “Prescott” processor is now starting to turn around the other way. Let’s move on to some encoding benchmarks to see if this performance holds up, of if “Prescott” sees some improvements.
..:: TMPGEnc & XMPEG ::..
For video encoding performance, we took a look at two programs, TMPEGEnc, and XMPEG. For XMPEG conversions, we utilized the latest DivX 5.1.1 codec, and for both of the above mentioned benchmarks, we used the default test file of Aero4 that come along with TMPEGEnc. The TMPEGEnc performance shows some very sizeable gains in performance between “Prescott” and “Northwood.” With the enhancements that have been made to the “Prescott” processor, we were expecting such performance gains, and were glad to see them with the results we just obtained in ScienceMark. We received less than thrilling results when we went about our benchmarking with XMPEG, since this time the performance gains were virtually null, one second for each respective clock speed.
..:: Quake III Arena ::..
Quake III Arena has been around for quite some time, and has been amicably used for benchmarking purposes ever since its inception. As time has progressed, less and less importance has been placed on this benchmark due to the ever increasing age of the software and graphics engine. The original Pentium 4 held a large lead over the competition for Quake III Arena performance, and from the results we received, this will continue to hold true for some time to come. The performance gains that we’re seeing here are far from what I was hoping for due to the added cache of the “Prescott” processor, and due to some of the results we achieved in earlier synthetic benchmarks. When it comes to real world performance, “Prescott” fails to shine as it is only able to overtake “Northwood” by one to three percent in Quake III Arena with that gain dropping as the resolution rises. Since most gamers run with higher res. settings than 640 x 480, the performance gain from “Prescott” will, in the end, be virtually null.
..:: Unreal Tournament 2003 ::..
Last up for today we have yet another real world performance benchmark for our motherboard reviews, Unreal Tournament 2003. We are using the built-in benchmarking utility with custom .ini files to make sure all settings are at high quality to allow for optimal benchmarking results for comparison. In the Unreal Tournament 2003 benchmarks, we see results that are even more disheartening than that of Quake III Arena. Now, we are seeing the “Prescott” processor actually fall behind the “Northwood” processor at an equally clocked speed. With all of the enhancements that have been made, and the added cache, we can clearly see here that they are ill-effective in combating the deeper pipeline. As the resolution grows, the performance loss by “Prescott” actually increases, another disheartening fact from some of the more encouraging results we achieved earlier.