BlueSmoke - Review : AMD AthlonXP 1800+

		Date	: Feb 12th, 2002
		Genre	: Processors
		Developer	: AMD
		Author	: Tom Smith

In late 1999 AMD release their new processor, the Athlon. This processor really allowed AMD to establish it's self in the processor market, with its excellent performance. As the last 2 years have gone by, AMD has released 2 updates to the Athlon Processor line. First was the thunderbird which sported on die cache that ran at the full speed of the processor and at that point AMD transitioned over to a socket based processor. The latest version of the processor is based on the palomino core. Also AMD has released a SMP version, MP and have updated the Duron with the Morgan core. The Athlon has a new name to boot. XP. XP stands for Extreme Performance. Does the processor live up to it's new name? Read on to find out.

Today we have the XP 1800+ on the test bed. It is not the latest in AMD's line up but it is one of their top three. So with that lets check out what is new in the latest line.

Since the release of the Thunderbird and the XP there have been some significant changes to the processor. One of the most important changes is they have reduced the die size to .18. They have also gone from the Ceramic Grid Pin Array to the Organic Grid Pin Array. This should help in reducing the amount of broken processors and allow the XP to scale more efficiently. AMD has also added SSE and 3D NOW! professional. The kicker on the other hand is QuantiSpeed™ Architecture.

In the beginning of the processor days all of the processors basically performed the same amount of work in a given time. Now is not the same back them . Cyrix was the first to go to a "rating" system. This was supposed to equal the performance of an Intel processor. More often than not that was not the case, back then. With the release of the XP line of processors AMD went back to a performance rating system. Yes Intel has the raw megahertz, but AMD felt that they could do the same amount of work as an Intel processor but at a lower Megahertz. Hence the 1800+ name with the processor today. It means that even though the processor is only clocked at 1.533 GHz it will do the same amount of work as an Intel 1.8 GHz processor. This is where the QuantiSpeed™ Architecture comes in.

QuantiSpeed is :

1. Nine-issue, superscalar, fully pipelined micro-architecture

At the heart of QuantiSpeed Architecture is a nine-issue, superscalar, fully pipelined core. This provides more pathways to feed application instructions into the execution engines of the core, simply allowing the processor to complete more work in a given clock cycle (high IPC). The delicate balance between the depth of the pathways (pipelines) and operating frequency of the processor produces high levels of performance. Longer pipelines alone translate into lower IPC, but high operating frequencies. However, shorter pipelines alone result in increased IPC, but lower operating frequencies. The AMD Athlon XP processor is designed to maintain a balanced approach between pipeline depth and processor frequency to provide extraordinary levels of overall processor performance.

2. Superscalar, fully pipelined Floating Point Unit (FPU)

QuantiSpeed Architecture features a superscalar, fully pipelined FPU, which completes more floating point operations per clock cycle than competitive x86 processors and permits high operating frequencies. The end result is the most powerful x86 FPU available today. The AMD Athlon XP processor has ample computing power to tackle the most computationintensive software applications.

3. Hardware data prefetch

Prefetching instructions from system memory to the processor’s Level 1 Instruction Cache is a common technique used to increase the processor’s work throughput (IPC) and therefore overall performance. This feature of QuantiSpeed Architecture prefetches data from system memory to the processor’s Level 1 Data Cache, which reduces the time it takes to feed the processor critical data, increasing work throughput. As a result, application performance is automatically enhanced when using the AMD Athlon XP processor with QuantiSpeed Architecture.

4. Exclusive and speculative Translation Look-aside Buffers (TLBs)

The TLB structures in QuantiSpeed Architecture keep the maps to critical data close to the processor. This is designed to prevent the processor from stalling or waiting when future data is requested. These TLB structures are now larger, exclusive between caches, and speculative. Larger TLB’s give the AMD Athlon XP processor access to additional data maps. The exclusive nature of these structures removes the duplication of information, freeing- up more space in the Level 2 cache for other useful data to be used by the processor. The aspect of speculation allows the AMD Athlon XP processor to generate future maps of critical data quickly. These three enhancements to the TLB structures further increase the work completed per clock cycle, thus improving real-world application performance of the AMD Athlon XP processor.

As a result of QuantiSpeed Architecture, the AMD Athlon XP processor has an optimum balance of IPC and frequency to achieve high levels of real world PC application performance.

3DNow! Professional is also a new feature, or rather a updated feature.

3DNow! Professional Technology

21 original 3DNow!™ instructions
19 Additional Instructions to Enable Improved Integer Calculations
5 DSP Instructions
52 SSE Instructions

Also the XP includes an internal thermal diode, but I have yet to see a motherboard that takes advantage of it. Most motherboards still use a probe that resides in the middle of the socket. The XP is also supposed to reduce power consumption and heat out put. The voltage is still 1.75 but there is a significant reduction in processor core temperature. This reduction in temperature will also allow for better scalability and overclocking.

Speaking of overclocking. You will notice that the gaps in the L1 bridges are laser cut. The ol' pencil to connect the bridges will no longer work. In order to connect the bridges you will need to fill the gaps with a non conductive filler, like High Speed PC's XP unlocking Kit (A review of it is in the works as I am writing this).

OK enough of this technical mumbo jumbo lets get on to the benchmarks.

The Intel system that I got to borrow was a Dell Dimension 4400. I was only allowed to borrow it for a few days, but I think the tests will show who comes out on top.

I used the following software to test the two systems.

• Sisoft Sandra 2002 TE
• 3D Mark 2000
• 3D Mark 2001
• Quake III Arena
• Unreal Tournament

Both Systems were built with a fresh install of Windows 2000. Installed the latest VIA and Intel drivers. DirectX 8.1 was used along with the latest nVidia Drivers.

A few comments first about the way that I benchmark. I do not turn off the sound in the games. I do not set the level of detail or anything else to a setting lower than what I would use to play the game. I believe that when you benchmark a product the manufactures the & end users want to know what the everyday performance of this product is.

Now on with the show.

First off we have the Sandra CPU Arithmetic enchmark. You can see that the XP clearly dominates the scene here in this benchmark. The XP takes about a 3.5 % lead over the Intel P4.

Here we have the Multimedia Benchmark, again with the XP clearly in the lead. We see the XP taking the lead with about a 5% difference.

Here are the memory scores. As you can see here the memory scores are very, very close. The Intel P4 leads the XP by less than 1 %, but this could very well be a limitation of the VIA chipset.

Here we see the the two processors duking it out in 3D Mark 2000. Again the XP leads by a small to medium margin. Remember these scores will not be really high, as the system I had put together has a GeForce 2 MX400, so the system is definitely video card limited.

Again in 3D Mark 2001 we see the XP lead the way by a nice comfortable margin. Remember the XP is only clocked at 1.533 GHz, 66 MHz less than the 1.6 and the XP is still giving the P4 a beating.

In Quake III the story has not changed. The XP still leads the P4 in all resolutions.

The same goes for Unreal Tournament. The P4's slower IPC just cannot keep up. Now we are seeing why AMD decided to go with the naming convention vs. the MHz rating.

So what have we seen here today. Something that we all knew, clock for clock the AMD XP beats Intel's P4. That is why AMD poised it's naming conventions with a rating system. The XP core is not ready to scale to 2 GHz. They will have to shrink down to the 13 micron process before they hit 2 GHz.

Now recently I have read some web sites that state the XP rating system is used to compare the XP and the Thunderbird line of processors. I have read the white paper and have seen what they say.

The Intel P4 is not a bad performing chip, but from a cost/performance ratio the XP beats it about the head and shoulders. Granted today the 1.6 is not as expensive as it once was, but lets look at the price of a 2.0 GHz P4 against a XP 2000+, you tell me which gives you more bang for your buck. AMD has had a winner on thier hands since they released the original Athlon series of processors, the XP is the latest and greatest, for now at least, in AMD's line up. Be on the watch out for some new stuff from AMD around mid-year. The XP comes Highly Recommended!