This Article was last modified in July, 1998.
The K6-2 is AMD's next-generation K6 processor. For all practical purposes, the K6-2 will outperform a Pentium II at the same clock speed, and even a Pentium II at higher clock speeds in many cases. While the K6-2 does feature a number of improvements over the original K6, the main improvement is the addition of AMD's 3DNow! technology. The AMD K6-2 also features superscalar MMX™ technology for leading-edge two dimensional multimedia performance and enables support for a 100MHz system bus (the first CPU to "officially" support the 100MHz bus speed) to speed access to L2 cache and main memory by up to 50 percent leading to a significant increase in PC system performance.
AMD officially released the K6-2 on May 28, with initial clock speeds of 266 and 300 megahertz and plans to move to 333 and 350 megahertz soon.
AMD's original K6 could match the Pentium II in many areas - mainly business areas, but could not outperform the Pentium II in any significant ways. Secondly, the original K6 was no way near the Pentium II in terms of FPU performance, which made it a very poor choice for gamers. To address this issue, AMD created the K6-2. The K6-2, like the second-generation K6 cpus, is a .25 micron processor, and is currently available in clock speeds of 266 and 300MHz. The main improvement of the K6-2 vs. the K6 is it's 3D performance, which is addressed with the addition of 3DNow! technology.
In a nutshell, MMX was developed primarily to improve integer based multimedia performance, but it does not effectively address the needs of floating point intensive 3D applications. The growing trend towards 3D applications requires even more powerful cpus with specific enhancements to accelerate 3D performance. 3DNow! moves beyond MMX to deliver the 3D performance needed for today's emerging 3D applications, namely, games. Both the AMD K6 and AMD K6-2 cpus include MMX technology. The AMD-K6-2 is the industry's first processor to implement 3DNow! technology. 3DNow! significantly improves the 3D performance of the K6-2, and finally gives the Pentium II some competition in the gaming area. Basically speaking, 3DNow is a special set of 'SIMD' (Single instruction, multiple data) instructions which greatly improve the speed of FPU calculations used in applications such as games.
For games and applications to take advantage of these new 3DNow! instructions, the must be specifically designed for them. Fortunately, Microsoft's soon to be released Direct X 6 will make full use of AMD's 3DNow! technology, meaning that any game which uses Direct X will be taking advantage of your K6-2. Many games also available right now take advantage of 3DNow!, such as Unreal, Incoming, Quake 2 (with the addition of AMD's drivers), and a few others. Pre-release versions of Direct X 6 are also floating around on the web, and finding a copy shouldn't be too hard.
So how much of a difference does 3DNow! really make?
To enable and disable 3DNow! in Quake II, CPU-Central simply switched the MiniGL driver (3dfxgl.dll) from AMD's 3DNow! MiniGL driver to the default MiniGL driver. When used in combination with Quake II and a Voodoo II card, AMD's 3DNow! technology gives nearly a 25% performance boost! Strangely, with 3DNow! enabled, Quake II actually performs faster at 800x600 than it does at 640x480.
When used in combination with software rendering (with a Matrox Millennium II), in Quake II, AMD's 3DNow! technology again gives nearly a 25% boost at all resolutions. Unfortunately, even with 3DNow!, frame rates with a Matrox Millennium II aren't very playable.
Installing a K6-2 is relatively simple - and requires basically the same amount of configuration as any other Socket 7 CPU. First of all, the K6-2 requires a split voltages, and requires a core-voltage of 2.2v - make sure the motherboard that you purchase supports this voltage. Secondly, the K6-2 runs at a bus frequency of 100MHz, and a multiplier of 3x. The K6-2 will run at lower bus speeds, however, overall performance will be significantly reduced.
Most people would expect that since the K6-2 is manufactured using a .25 micron manufacturing process, and uses such as low voltage, that the K6-2 should be very overclockable. However, the K6-2 that CPU-Central received was not as overclocking-friendly as we had originally expected - the absolute highest speed we could get it to boot up at was 333MHz (112x3 or 83x4). And even at this speed, the system was very flaky, and not stable enough to use for practical purposes. Anything higher than 333MHz (with voltages from 2.1v to 2.6v) would just get us a blank screen when we turned the computer on.
This may sound very disappointing (and it is for us), but from what we've read from various newsgroups, many people have been able to get their K6-2 300 up to 350MHz running very stable, and a few people even up to 372MHz (with much less stability).
Processor Comparative Chart
|Processor Features ||Performance Benefits ||Celeron ||AMD-K6 ||Pentium II ||AMD K6-2 with 3DNow! technology |
|Die size (mm2) ||Smaller die size=lower cost ||203 ||162 (0.35 micron) |
68 (0.25 micron)
|203 (0.35 micron) ||81 |
|Clock speed (MHz) ||Faster clock speed generally means faster processing and apps launching ||266MHz ||233MHz |
|L1 cache ||Built-in feature that helps the chip retrieve even faster ||32K |
|L2 cache ||Augments L1 cache, making data retrieval even faster ||No ||Yes ||Yes ||Yes|
|Processor bus (max bus speed) ||Moves data between chip and memory ||Slot 1 |
|Socket 7 |
|Slot 1 |
|MMX(TM) Instruction Set ||Enhances multimedia applications and runs other apps 10% faster ||Yes ||Yes ||Yes ||Yes|
|3DNow!(TM) Technology ||Enables lifelike graphics for a more realistic experience ||No ||No ||No ||Yes|
|100 MHz Bus Support ||Moves data between the chip and the memory. L2 cache can¹t improve performance without a fast bus to move the data. Faster bus eliminates the data bottleneck ||No ||No ||Yes ||Yes|
|Accelerated Graphics Port Support ||Speeds up 3D graphics ||Yes ||Yes ||Yes ||Yes|
|X86 compatibility ||Standard industry architecture, essential for running standard PC applications ||Yes ||Yes ||Yes ||Yes|
Processor Comparative Benchmarks
To see just how well the K6-2 performs against a Pentium II, here are some comparative benchmarks taken from AMD's website:
As you can see, the AMD K6-2 with 3DNow! technology will easily outperform a Pentium II processor at a similar clock speed, and will even challenge the Pentium II 400. For 3D Winbench to take advantage of 3DNow! of course, the system was equipped with a pre-release version of DirectX 6, so results with the final release will most likely vary. For more information on system configuration, visit AMD's website.
Although business performance wasn't marketed as the main feature of the K6-2, as you can see, the K6-2 will still perform neck-and-neck with a Pentium II at an equivalent clock speed, and will produce most that satisfactory business results. For more information on system configuration, visit AMD's website.
Quake II (for the very few of you who don't know) is a very demanding 3D action game from ID Software, and also makes very valuable real-world benchmarks. All tests were performed using AMD's 3DNow! miniGL driver.
Looking at this chart you'll notice some strange results. First of all, running at 90x3.5 (315) or 83x4 (333), doesn't run Quake II as fast as running at 75x4 (300). Why is that? 75MHz is a much lower bus speed than 90 or 83 MHz, and the overall CPU speed is also lower. The only logical explanation that CPU-Central could come up with is the PCI bus speed. Quake II makes heavy use of the Voodoo II card (which is on the PCI bus), and when running at 75MHz FSB, the PCI bus along with the Voodoo II card, are being overclocked. On the other hand, when running at 83 or 90 MHz FSB, the PCI bus is actually being underclocked.
As like the rest of the benchmarks, running at 112x3 (336) is the absolute fastest speed by far. The only problem was stability. Quake II suffers from frequent crashes and lock-ups at this speed, as do just about every other program.
CPUmark32 is a 32-bit Windows processor benchmark provided by Ziff-Davis Labs designed to measure the performance potential for running 32-bit applications.
CPUmark32 would not complete testing at 75x4.5 or 112x3 because of instability, making the default speed of 100x3 the highest overall speed. As you can see, CPUmark32 scores generally get higher as the bus speed increases.
Incoming is another 3D action game, from Rage Software. Incoming is also capable of making very valuable real-world benchmarks.
As usual, 112x3 (336) is by far the fastest speed. Unfortunately, as with other programs, Incoming is not very stable at this speed. Incoming does not seem to benefit specifically from a higher bus speed, PCI bus speed, or higher CPU clock speed, but rather, a combination of all of these things.
FPUmark is a 32-bit floating point test. It is a component of WinBench 98.
By looking at this graph you can clearly see that FPU WinMark gives better scores as the CPU's clock speed increases. It does not seem to benefit significantly from the bus speed, or PCI bus speed. Like the other tests, 112x3 (336) is the fastest setting, and in this benchmark, is actually tied for first with 75x4.5 (337).
All benchmarks were performed on a system with the following components:
- AMD K6-2 300MHz
- Soyo 5EH w/1MB L2 Cache
- 64MB Corsair Microsystems PC100 SDRAM
- Diamond Monster 3D II
- Matrox Millenium II (PCI)
- Quantum Fireball SE 6.4GB
- 21 instructions
- Support for SIMD floating-point and integer operations
- Specific SIMD integer instruction to enhance MPEG decoding
- New PREFETCH instruction to eliminate extra data retrieval time
- FEMMS (Fast Entry/Exit Multimedia State) instruction to reduce switching time between MMX(TM) and x87 code
- Open-standard support of IEEE 754 single precision data type
- Fully pipelined dual execution resources
- Unlimited storage of floating-point numbers in memory
- Execution of up to two 3DNow! instructions per clock
- Total of four floating-point calculations (add, subtract, multiply) per clock (Enables potential peak performance of 1.2 Gigaflops at 300MHz vs. potential peak performance of 0.3 gigaflops for 300MHz cpus without 3DNow! technology)
- Common floating-point stack; eliminates task switching between AMD-3Dnow! and MMX operations
3DNow! technology -- MMX Technology Relationship
- Different technologies, yet with similar encoding and simultaneous instruction execution
- MMX developed to improve integer-intensive operations in rendering such as 2D graphics
- 3DNow! technology developed to speed up floating-point-intensive front-end operations designed to boost 3D graphics and multimedia performance
We at CPU-Central believe that the K6-2 is an excellent Pentium II alternative, and is one of the fastest performing low-cost CPU available today.
- Gamers who want great performance and who don't want to spend excess money on a Pentium II will definitely find the K6-2 an excellent processor.
- People who use mainly business applications and again, don't want to spend excess money on the Pentium II will find the K6-2 to offer an excellent price/performance value.
Manufactured By: Advanced Micro Devices (AMD)
Places to Buy: TC Computers, Micro Trends