'Performance Matters More Than MHz'
This heading is directly taken from the front page of an AMD presentation that I received right after I had been the first to report about AMD's plans to use the launch of the 'Desktop Palomino' for the introduction of a new performance rating system in late August 2001.
In times as hard as this, after experiencing a constant business downturn all year and then a terrible terrorist attack that killed thousands of innocent people and that made the economy even weaker, AMD's future revolves around "Performance Matters More Than MHz" and the question if customers are going to understand and accept it.
This article will be a bit different than the usual stuff you know about processor releases. The real beef of AMD's AthlonXP-launch lies not in processor architecture, benchmark numbers and overclocking climaxes. It's all about AMD trying, or rather struggling, to change people's perception of computer performance, which - rightly or not - has been based on the good old 'megahertz' for decades. Trying to change old habits is a very difficult task and AMD is fully aware of that. That's why AMD is starting with a somewhat halfhearted attempt of replacing 'megahertz' with, well, something that looks like 'megahertz' as well. We'll get to that a bit later.
AthlonXP - The Technical Specs
New Name
First and foremost, the name 'AthlonXP' is not just following the current trend to call everything after Microsoft's new operating system WindowsXP (like e.g. NVIDIA's new originally named driver set 'DetonatorXP'), but the 'XP' that is supposed to hype the new 'experience' you might have with WindowsXP, is supposed to suggest the 'extreme performance' of AthlonXP.
New Features
AthlonXP does indeed offer 3-7% more performance per clock than previous desktop-Athlons based on the "Thunderbird"-core, because it is based on the new "Palomino"-core that we already know from AMD's multi-processor offering AthlonMP and the Mobile Athlon4 processors.
I'll list Palomino's added features over Thunderbird once more:
- New design to reduce power consumption over Thunderbird core by 20%
- Implementation of the full Intel SSE instruction set. The SSE processor flag is set (if the motherboard BIOS supports Palomino) so that software can recognize AthlonMP as a SSE-capable processor. AMD calls its SSE-implementation '3Dnow! Professional'.
- Hardware auto data pre-fetching unit
- L1 Data TLB (Translation Look-Aside Buffer) was increased from 32 to 40 entries, the architecture of the data and instruction L1 and L2 TLBs was made exclusive and TLB-entries can be written speculatively.
- Implementation of a thermal diode to monitor processor temperature
New Packaging
AthlonXP comes in a new packaging, which is based on the same material as found in Intel's Pentium III, Celeron and Pentium 4 processors. The only difference is that the material used with AthlonXP is brown, while the Intel processors are green. The new material is supposed to have better thermal behavior and more elasticity than the previously used ceramic. Besides this important feature, the new organic material is also lighter and cheaper than the previous packaging material.
Clock Speeds
Today AMD introduces AthlonXP processors at 1333, 1400, 1467 and 1533 MHz, all using a FSB clock of 133/266 MHz. AMD doesn't want to see or hear MHz-numbers anymore though, so what we are actually presented with today is AthlonXP 1500+, AthlonXP 1600+, AthlonXP 1700+ and AthlonXP 1800+. AMD calls the number with the '+' at the end behind the 'AthlonXP' name 'model numbers', as Tom's Hardware already reported more than two months ago.
How seriously AMD is taking this 'Model Number' issue, is shown in the picture above. The BIOS of a motherboard that is supposed to get approval from AMD is not allowed to display the actual clock frequency, but only the model number! WindowsXP reports both, the 'model number' as well as the actual clock frequency, as you can see in the picture of a German WindowsXP below.
AMD has got to be consistent, and that's all it is by trying to keep the display of actual clock frequencies of AthlonXP at a minimum.
Benchmarks
Before we get into discussions of the pros and contras of AMD's model number policy, I suggest we have a look at the benchmark results scored with AthlonXP 1800+, which is in actual fact running at 1533 MHz. I want to be frank with you and admit that I am not really satisfied with the benchmarks we did, which is why I will supply you with the charts only and spare you any of my comments. I have requested a lot more benchmarks from our lab and the guys there have started running them. Please feel free to send in your own benchmarking suggestions. They will be posted as soon as possible.
Benchmark Setup
Intel Systems |
Socket 423
|
CPU |
Intel Pentium 4/ 2000 MHz (400 MHz QDR FSB) |
Motherboard |
ASUS P4T (Intel 850) Revision: 1.06 |
Memory |
2 x 128 MB, RDRAM, 400 MHz, Viking |
Socket 478
|
CPU |
Intel Pentium 4/ 2000 MHz (400 MHz QDR FSB) |
Motherboard |
Shuttle AV40V12 (VIA P4X266) Revision: 1.0 |
Memory |
2 x 128 MB, DDR-SRDAM, 133 MHz, CL2, Micron |
AMD Systems |
Socket 462
|
CPU |
AMD Athlon XP 1800+ (266 MHZ DDR), clock speed at 1533 MHz AMD Athlon 1400 MHz (266 MHZ DDR) |
Motherboard 1 |
Gigabyte GA-7DX (AMD 760) Revision: 4.0 |
Motherboard 2 |
Epox EP-8KHA+ (VIA KT266A) Revision: 2.0 |
Memory |
256 MB DDR-SDRAM, CL2, PC2100, Micron |
Common Hardware |
Graphics Card |
GeForce 3 Memory: 64 MB DDR-SDRAM Memory Clock: 200 MHz Chip Clock: 250 MHz |
Hard Disk |
40 GB, 5T040H4, Maxtor UDMA100, 7200 rpm, 2 MB Cache |
Drivers & Software |
IDE (Intel) |
Intel Chipset Installation Utility Production Release V3.00.029 IAA Version 3.1.2017.0 |
AGP (Intel) |
Intel Ultra ATA Storage Driver Production Release v6.10.011 |
IDE (AMD) |
VIA 4 in 1 Version 4.33 FINAL |
AGP (AMD) |
Miniport Driver 4.80 (Win98SE) Miniport Driver 5.22 (Win2K) |
Graphics Driver |
Detonator 4 Series V21.81 |
DirectX Version |
8.0a |
OS |
Windows 2000 Pro SP2 (Build 2195) Windows 98 SE, Version 4.10.2222 A |
Benchmarks and Settings |
Quake III Arena |
Retail Version 1.16 command line = +set cd_nocd 1 +set s_initsound 0 Graphics detail set to 'Normal' Benchmark using 'Q3DEMO1' |
3DMark2000 |
Version 1.1 Build 340 - default Benchmark |
3DMark2001 |
Build 200 - default Benchmark |
SiSoft Sandra 2001 |
SE Pro |
CINEMA 4D |
CineBench R6 |
Unreal Tournament |
Version 4.36 , UTBench |
Sysmark 2000 |
Patch 5 |
Sysmark 2001 |
Patch 3 |
Dronez |
|
Evolva |
V1.2 Build 944 |
Benchmark Results
Sysmark
3D Games - Windows 98 SE: Quake 3 Arena
3D Games - Windows 98 SE: Unreal Tournament
3D Games - Windows 98 SE: Dronez
3D Games - Windows 98 SE: Evolva
3D Games - Windows 2000 Professional: Quake 3 Arena
3D Games - Windows 2000 Professional: Unreal Tournament
3D Mark 2000
3D Mark 2001
Cinema4D - Windows 2000 Professional
Sandra Scores
New Pricing
Model |
Official 1KU OEM Price |
Street Price |
AthlonXP 1800+ (1533 MHz) |
$252 |
$225 |
AthlonXP 1700+ (1466 MHz) |
$190 |
N/A |
AthlonXP 1600+ (1400 MHz) |
$160 |
$150 |
AthlonXP 1500+ (1333 MHz) |
$130 |
$125 |
Athlon 1400 MHz |
$130 |
$105 |
Athlon 1333 MHz |
$125 |
$95 |
Athlon 1200 MHz |
$120 |
$85 |
On the street AthlonXP is currently more expensive than the old Athlon, even though Athlon 1400 is most likely superior to AthlonXP 1500+, because e.g. for the FPU clock frequency is more important than Palomino's new features.
Let's have a look at Pentium 4's pricing:
Model |
Street Price |
Pentium 4 2000 MHz |
$550 |
Pentium 4 1900 MHz |
$370 |
Pentium 4 1800 MHz |
$240 |
Pentium 4 1700 MHz |
$180 / $200 |
Pentium 4 1600 MHz |
$160 |
Pentium 4 1500 MHz |
$130 |
If the compare Pentium 4 to AthlonXP and take the according P4-MHz vs. AthlonXP model number, you get very similar prices. However, AthlonXP performs better, while it is clocked lower than its model numbers suggest.
AMD's New Model Policy
The benchmarks show quite clearly that AthlonXP 1800+ is able to beat Intel's flagship Pentium 4 2 GHz in the majority of benchmarks. From that point of view, the AthlonXP at 1533 MHz would have rather deserved the model number 1900+ or even 2000+. However, AMD was basing its model rating on the expected performance of the next Pentium 4 core 'Northwood', which will be performing better per clock than current Pentium 4 CPUs.
AMD's problem is the fact that Athlon processors are able to do a lot more work per clock cycle than Intel's Pentium 4. We just saw that AthlonXP at 1533 MHz is able to reach or even beat Pentium 4 running at an almost 33% higher clock speed. Normal buyers won't care about the 'work per clock' issue. They just ask for 'megahertz' and that's AMD's trouble. Hence, AMD didn't have any other choice than introducing a measure that would reflect the actual performance of its processors. The new model numbers are a first step in this direction.
AMD's new model rating is based on 14 benchmarks that represent 34 applications of the three fields 'visual computing', 'gaming' and 'office productivity'. AMD was trying to be as fair as possible, with the effect that the current model numbering of AthlonXP processors is a bit overly humble, as you can see from our benchmarks. Once Intel releases Pentium 4 processors based on Northwood however, the model numbers may become more realistic.
The model numbering is only an intermediate solution. AMD is trying to help establish an independent institution that creates a new performance measure that is fairer and more realistic than the current MHz-rating in 2002. This institution is supposed to benefit every PC customer, regardless if he buys Intel or AMD processors, by providing a reliable way of determing performance.
We will see how successful AMD will be in helping to establish this institution. Intel won't have any interest in giving aid, because it benefits from the clock frequency (hype) lead of Pentium 4. Microsoft will have to join in, since a majority of software comes from the software giant. OEMs might or might not be interested. Especially the old buddies of Intel won't be willing to help or recognize AMD's initiative. The media could help a lot, but certainly not the crazy fanatics found in so many Internet publications. It will be a long way and this institution will need a strong spokesperson that is able to take a lot of abuse from the AMD as well as Intel fanatics across the globe. The noble idea behind this institution might not be good enough to make it real. Let's see what happens.
There will be a lot of criticism of AMD now. Yes, it is true that the model numbering might confuse customers right now, since it is indeed difficult to know if Athlon 1400 is now slower or faster than AthlonXP 1500+. However, these are transitional times. AMD's new model number system deserves better than being compared with the old and confusing P-Rating. AMD hasn't got an alternative right now and so I respect the new system and accept it, even though I might not like it too much. I suggest that we all give it a chance.