System setup, notes, and MT discussion
Hardware
| Athlon X2 System | Athlon FX System | Pentium 4 System | |
| Processors | AMD Athlon X2 4800+ | AMD Athlon FX-55 AMD Athlon FX-53 |
Pentium 4 570J Pentium 4 660 Pentium 4 3.46GHz XE Pentium 4 3.76GHz XE |
| Mainboards | ASUS A8N-SLI | ASUS P5AD2-E | |
| Memories | 2 x 512MiB Corsair XMS3200 XL Xpert 2-2-2-5 @ DDR400 |
2 x 512MB Crucial Ballistix DDR2-667 4.0-4-4-12 @ 710MHz for Extreme Editions 3.0-3-3-10 @ 533MHz for all others |
|
| Graphics Cards | NVIDIA GeForce 6800 Ultra 256MiB PCI Express 16X |
||
| BIOS Versions | 23rd March 2005 | 1002.005 - 22nd October 2004 | |
| Hard Disks | Western Digital Raptor 74GB SATA, 8MiB buffer, 10000rpm |
||
| Operating Systems | Windows XP Professional 32-bit Service Pack 2 |
||
| Core Logic Drivers | NVIDIA nForce4 Platform Driver Version 6.39 |
Intel INF Update Utility 6.3.0.1007 | |
| Graphics Drivers | NVIDIA ForceWare Unified Driver 71.20 |
||
Software
The 32-bit build of Microsoft Windows XP Professional was used, along with the venerable Service Pack 2. The software used for benchmarking was as follows:ScienceMark 2.0
HEXUS Pifast
HEXUS Crypto
MainConcept MPEG Encoder
LAME 3.96
Cinebench 2003
3D Studio Max v6
Kribibench
Realstorm 2004
picCOLOR v4.0
Notes
The Athlon X2 4800+ was benchmarked against Athlon FX-55, FX-53 and a range of recent Pentium 4 Processors. The reader is encouraged to visit Tarinder's recent look at the Intel Pentium Extreme Edition 840 for comparison results with that processor. Note that due to differences in Tarinder's test platform and software, compared to the suite used for this article, the results aren't directly comparable. I'll update the article with full comparison numbers in due course, as I rebenchmark the XE 840 on a fully comparable platform. Apologies for having to suffer doing so, as we resync testing platforms inside HEXUS.In addition, a selection of benchmarks from our professional benchmark suite was chosen to highlight multi-threaded application performance on a multi-threaded operating system, against Athlon FX-53 (the single-core processor with the same clock frequency and L2 cache size).
Click for a larger version
CPU-Z Processor Information
CPU-Z Mainboard Information
Multi-Threading Discussion
Obviously, the benefits of a dual-core processor are only really felt in multi-threaded applications and an operating system environment that's able to take advantage of it, too. Microsoft's Windows XP Professional kernel and thread scheduler is HyperThreading aware, and will schedule threads across both virtual cores on that class of SMP processor, during very basic use of the operating system. That's where proponents of HyperThreading, of which I'm one, find their best arguments for choosing a processing with HT Technology. Using a dual-core processor gives you those benefits across the board, in terms of operating system responsiveness and that smooth feel that HT users like to comment on. It's not psychological, either.However, the major benefits on a multi-processor system, which dual-core CPUs implement, are in applications that can use multiple threads of code to speed up their operation, by utilising the execution resources of more than a single processor. The possibilities for performance improvements entirely depend on the application. Parallelising software isn't a simple task, unless the problem you're trying to solve in code is an inherently parallel one to begin with. For applications that aren't parallel in nature, there's not much to be gained by engineering in parallelism in an attempt to gain small speed increases. The effort to do so negates the slight performance improvements you might see.
However, for the classes of applications that are inherently parallel, such as media encoding (you're doing the exact same operations millions of times on media data, which often don't have to be done in order, so can be done as fast as possible on any CPU available), image processing (you're doing the same operations on each pixel or pixel group, and they often aren't dependant instructions, so you just run them on whatever you've got, as fast as possible) and offline 3D rendering (you're processing scanlines usually, and you can do so as quick as you can, in any order, on any CPU you have available), there's definite scope for developing and optimising for multi-processor systems.
Those applications have been accelerated by multi-processing systems, without dual-core processors, for years. They'll therefore go quicker on dual-core processors, too, as I'll show you soon, by default.
It's the games question that's probably the most pressing, in terms of deciding whether you need a dual-core processor, for the people that usually read HEXUS. At the time of writing, it's a clear answer of "the software isn't ready, yet". That 'yet' is crucial. With dual-core processors about to hit the mainstream x86 space en-masse and all of the major next-generation consoles sporting multiple execution cores for the games developers to utilise in their next generation games for those systems, games development, which has a long tradition of driving the PC industry forward especially in terms of graphics, will absolutely be focussed on multi-processing systems in the future.
And therein lies the biggest barrier to picking up a dual-core system in the near future. For the gamer, yes, you'll want a dual-core processor at some point, without any shadow of a doubt. For the user keen on media processing or the catch-all term of content creation, they'll likely interest you immensely too. For the average PC user, well, you'll be interested in one less than the other guys, but if the cost is right, why wouldn't you choose one, just for the less-perceptible, but no less there, improvements on the desktop.
Future processor architecture is firmly entrenched in going sideways with parallelism, rather then upwards with clock speed.
That all said, there is the potential for speedup in inherently single-threaded applications like games, on a dual-core or other multi-processor system, while actually using that application on an operating system that's working hard on other tasks at the same time. Think about it. On a single processor system, that one processor has to feed both the OS and all your running applications with its single set of execution resources. Your system-hogging application, like a game, has to share those resources with everything else that's running. Add a second processor, or more, and you've got spare resources to devote to running the OS and all other applications and tasks that you usually have open at the same time as the game.
Who really shuts down everything else while they play a game or run an intensive application? You're not using a computer properly if you do. It opens up a can of benchmarking worms, but it's a can that needs to be opened should we wish to evaluate the real performance improvements to be seen on multi-processor systems, using both single and multi-threaded applications.
Therefore we'll revisit the X2 range and Intel's Pentium D and Extreme Edition 840 range with tests designed to measure performance under a standardised operating system load with other applications running, to see what gains are to be had. Absolute results where the application has the CPU to itself (relatively speaking) will have to sit side-by-side with a new class of benchmarks.
And you, the HEXUS reader, can give us a hand with that. Keep an eye on the forums for the thread where I solicit your opinions on how we go about it.
