The new .18 micron Intel Pentium III processor, codename Coppermine, is the latest member of the Intel P6 family. A family that started off with the Intel Pentium Pro a few years ago, and which has been the foundation of a whole new family of Intel CPU’s. Originally targeted at the server market, the Pentium Pro came in a wide variety of clockspeeds and L2 cache configurations. It’s well known successor, the Intel Pentium II, was the logical follow up as it added MMX technology to the P6 core as well as an external L2 cache running at half the clockspeed of the CPU. To be able to keep the CPU and the external cache as close together as possible, Intel moved away from the traditional socket approach for interfacing the CPU with the motherboard and chose a cartridge design, Slot 1, that mounted the CPU and the external cache onto a PCB which interfaced with the motherboard using the Slot 1 connector.
The original Intel Pentium III, manufactured in a .25 micron process wasn’t that much different from the Intel Pentium II either. It added a few enhancements to the CPU core, such as SSE, Streaming SIMD Extensions for enhanced floating point and 3D application performance. Which basically meant that some geometrical or integer intensive calculations could be handled more efficiently and thus be processed faster by using these instructions. As well as the Intel Processor Serial Number, a feature that remains a topic of much controversy as it allows the user to be identified by the serial number of its CPU.
The new Intel Pentium III however heralds the return of the L2 cache running at clock speed, much like the original Pentium Pro. Whereas with the Intel Pentium Pro the L2 cache was mounted inside the CPU package, with the new Intel Pentium III the L2 cache is actually on-die, thereby reducing cost as well as improving L2 cache latency and throughput.
What’s Intel Inside?
Before we start looking at the actual real world performance of this new Intel Pentium III lets take a closer look at some of the new features and their influence on the overall performance.
0.18 Micron Process
The new Intel Pentium III is manufactured in a .18 micron process, which means that the minimum trace width on the CPU is 0.18 micron. Shrinking the die size of a CPU has multiple advantages, one of the most beneficial being that it postpones the need for a total CPU re-design, which can be far more expensive and time consuming than upgrading the fabrication process to the new die size. If we look at the entire family of P6 Intel CPU’s we’ve gone from .35 Micron for the Klamath (Pentium II / 266 … 300) to .25 micron for the Deschutes and Katmai (Pentium II / III 450 … 600) and are now using .18 micron for the Coppermine (Pentium III / 500 … ). As mentioned earlier all these CPU’s are based on the same P6 core and are basically enhancements over the earlier generation.
The reason that a CPU can be clocked to a higher speed if we use a smaller die is the fact that the individual transistors as well as the length and width of the interconnecting traces is reduced. The reduction in size causes for far less propagation and switching delay in the transistors, allowing the CPU to be clocked at a higher speed. Thus by simply shrinking the die size as well as having a very scalable CPU design, they’ve been able to clock the same modified P6 core to much higher speeds than the earlier generations.
The next advantage of shrinking the die is that you can now have more transistors in the same amount of space. If we look at the original 0.25 micron Pentium III it featured 9.5 million transistors on a 140mm^2 die, whereas the new Pentium III has 28 million transistors, including the L2 cache, on a 106mm^2 die.
One other significant advantage of shrinking the die is that the voltage requirements of the CPU are lowered. As all the transistors and interconnecting traces have become smaller there is also less power needed to operate them, thus we can lower the voltage and therefor decrease the power consumption of the CPU.
Advanced Transfer Cache
A CPU’s cache memory usually is made up out of a L1 and a L2 cache. The L1 cache is located within the CPU core and has a very low latency as it is used extensively for all sorts of purposes, such as data fetching, data shifting and data loops, whilst storing only small portions of data. The L2 cache is used for storing larger portions of data, usually coming from the L1 cache, and thus having a size that is a multiple of the L1 cache. An often used L2 cache size configuration is between four and sixteen times that of the L1 cache depending on the configuration and desired latency.
The biggest advantage of moving the L2 cache on die and having it run at clockspeed is the fact that both the L1 and the L2 cache can be run in parallel and can be accessed concurrently resulting in a further reduction of latency, which is a huge performance booster.
The new Pentium III uses a 256 bits data path to the on die 256KB cache as opposed to the 64 bits data path as used on the original Pentium II / III. This alone offers a fourfold improvement over the original configuration, but there is more, as the new Pentium III offers a L2 latency that is four times as small as the original Pentium III, boasting a combined 11.7Gbyte/s data throughput for a 733MHz Pentium III.
Advanced System Buffering
Intel also increased the number of Fill Buffers from four to six. Fill Buffers are used as an interface between the CPU and the much slower main memory. This to ensure that the CPU will not have to stall when it wants to write data back in the much slower main memory or when the CPU wants to read multiple data locations in the main memory.
To further up the performance of interfacing with the main memory they’ve also increased the number of Bus Queue Entries from four to eight. Bus Queues are used to hold outstanding bus or memory operations and are often used in the fetching and pre-fetching of data across the system bus.
The number of Writeback Buffers has also been increased; the new Pentium III now has four whereas the original Pentium III had only one. Writeback Buffers are used to reduce data blocking during cache data replacement operations and they also provide faster deallocation times for Fill Buffers as data doesn’t need to be transferred along the system bus, but can be allocated from the Writeback Buffer.
Defining the Benchmarks
For testing the new Pentium III we could have run a whole slew of benchmarks at various resolutions and color depths, starting from 640×480@16bits and up to 1600×1200@32bits. We however chose not to do so, as we want to look at the real world performance and not at the maximum performance at 640×480@16bits as nobody is going to actually run its system at that resolution, same applies for 1600×1200@32bits. Upon careful consideration we decided to look at the performance using a 1024768@16bits and 1024×768@32bits resolution as these resolutions are most often used.
What is far more interesting however is looking at the performance difference between a 100MHz and a 133MHz system bus, as the Pentium III is optimized to make maximum use of the 133MHz bus. It would be interesting to see if the 33% increase in system bus frequency also gives us a similar percentage gain in performance.
Benchmarking software is also a topic of much controversy, as most benchmarking software outputs a number that actually doesn’t say anything about how fast the system is without a comparison to another system. We feel that game benchmarks do offer a real world benchmark as they focus around the actual game that is going to run on the system. If we use that game to measure the system’s performance, we can immediately see what the implications are of for example switching to another color depth.
To be able to rule out bottlenecks such as, but not limited too: fillrate limitations, AGP bus saturation, and also be better able to analyze the game benchmarks, we think it is also worthwhile to compare these game benchmarks to some business benchmarks. This to be able to make a sound judgement of what is actually going on and whether a certain bottleneck is driver, software or hardware related. Without doing this double check you can only guess about what is actually going on, although you might guess correctly based on the information about the systems capabilities. Doing a double check on the benchmark data will reveal exactly what is the matter and will allow you to do a far better analysis of the actual systems performance.
For doing the game benchmarks we choose to use Quake II v3.20 and the Quake III Arena Demo v1.05. Both games have built in means to determine system performance, although heavily focussed on FPU and graphic performance. For Quake II we use the Demo1 and Crusher benchmark as both are a good indication of what a minimum and a maximum load during a deathmatch session consists of. We’ve set all of the graphics detail and texture settings to maximum whilst running these benchmarks. If you want to run some of these benchmarks on your system, you’ll need to do the following :
Demo1 :
– switch to resolution X with colordepth Y on your desktop
– start up Quake II
– set the appropriate graphics settings in the game
– bring down the console by pressing [~]
– type [timedemo 1] [enter]
– type [demomap demo1.dm2] [enter]
Crusher :
– download the Crusher.zip file
– unpack it into your Quake II directory, see the attached Readme
– switch to resolution X with colordepth Y on your desktop
– start up Quake II
– set the appropriate graphics settings in the game
– bring down the console by pressing [~]
– type [timedemo 1] [enter]
– type [demomap crusher.dm2] [enter]
The Quake III Arena Demo uses a very similar approach as the Quake II demo as it also measures fps but the entire game is based around a new 3D engine that stresses the system and its components to far greater lengths than the Quake II timedemo’s. Again, we’ve set all of the graphics detail and texture settings to maximum whilst running these benchmarks. If you want to run some of these benchmarks on your system, you’ll need to do the following :
Quake III Arena :
– download Quake III Arena Demo v1.05
– switch to resolution X with colordepth Y on your desktop
– start up Quake III Arena
– set the appropriate graphics settings in the game
– bring down the console by pressing [~]
– type [timedemo 1] [enter]
– type [demo q3test1] [enter]
Choosing Benchmark Applications
As mentioned we will also be looking at the business performance and for that purpose we use SiSoft Sandra ’99 to determine CPU. FPU and memory performance and Ziff Davis WinBench 99 to determine CPU, FPU and Business Application performance.
The reason for using these benchmark packages is because they can be downloaded off of the internet free of charge, thus enabling everyone that wants to benchmark his or her own system to do so. Furthermore both packages feature online support and software updates and also offer a benchmark database so you can verify your results.
SiSoft Sandra ’99 uses a whole slew of modules to be able to determine system contents and system performance. In this review we’ll use CPU Benchmark, CPU Multi-Media Benchmark and the Memory Benchmark modules to be able to determine the performance of the CPU and the memory subsystem.
Ziff Davis WinBench 99 also uses a whole slew of modules that can be used to determine system performance. In this review we’ll use CPUmark 99, FPUmark 99 and Business Graphics Winmark 99 to be able to determine the performance of the CPU and the graphics subsystem.
If you want to get a hold of a copy of these programs they can be found here:
SiSoft Sandra ’99 : http://www.sisoftware.co.uk/sandra
Ziff Davis WinBench 99 : http://www.zdbop.com
Benchmark Results Page 1
We’ve been using two types of videocards in our benchmarking, the Matrox G400 and the Voodoo3/3000, to be able to compare them, as well as to look at their scalability at higher clockspeeds. We’ve chosen the Matrox G400 and the Voodoo3/3000 as we feel that they both offer more than adequate performance and some of the best features at a very good price/performance ratio. Also, they are available now and have been bestsellers for months, and we’d rather focus on real world benchmarks on a typical setup. Not everybody has the resources to upgrade to the fastest videocard the moment it becomes available.
Matrox G400 Benchmarks at 100 MHz FSB
Test system :
Intel Coppermine ES
Abit BX6 Rev. 2.0 MW BIOS
Corsair 128 MB PC-133
Matrox G400 DH 32 MB at 1/1 FSB
Generic 10 GB HD
Generic 32X CDROM
Windows 98 SE with DirectX 7.0
CPU Clockspeed : 500 MHz / 5×100
Sandra CPU : 1349 mips/673 mflops
Sandra MEDIA : 1205 mmx/731 fpu
Sandra MEM : 313 cpu/284 fpu
CPUMark 99 : 48.1
FPUmark 99 : 2700
WinMark 1024x768x16 : 239
WinMark 1024x768x32 : 232
Q2 3.20 DM1 1024x768x16 : 65.7fps
Q2 3.20 CRU 1024x768x16 : 48.1 fps
Q2 3.20 DM1 1024x768x32 : 61.9 fps
Q2 3.20 CRU 1024x768x32 : 46.6 fps
Q3 Test 1024x768x16 : 47.8 fps
Q3 Test 1024x768x32 : 39.0 fps
CPU Clockspeed : 600 MHz / 6×100
Sandra CPU : 1620 mips/811 mflops
Sandra MEDIA : 1446 mmx/877 fpu
Sandra MEM : 313 cpu/284 fpu
CPUMark 99 : 56.4
FPUmark 99 : 3240
WinMark 1024x768x16 : 280
WinMark 1024x768x32 : 272
Q2 3.20 DM1 1024x768x16 : 65.7 fps
Q2 3.20 CRU 1024x768x16 : 51.3 fps
Q2 3.20 DM1 1024x768x32 : 61.9 fps
Q2 3.20 CRU 1024x768x32 : 49.2 fps
Q3 Test 1024x768x16 : 48.1 fps
Q3 Test 1024x768x32 : 39.9 fps
CPU Clockspeed : 700 MHz / 7×100
Sandra CPU : 1890 mips/952 mflops
Sandra MEDIA : 1687 mmx/1023 fpu
Sandra MEM : 319 cpu/285 fpu
CPUMark 99 : 64.4
FPUmark 99 : 3780
WinMark 1024x768x16 : 321
WinMark 1024x768x32 : 308
Q2 3.20 DM1 1024x768x16 : 65.7 fps
Q2 3.20 CRU 1024x768x16 : 52.3 fps
Q2 3.20 DM1 1024x768x32 : 61.9 fps
Q2 3.20 CRU 1024x768x32 : 49.9 fps
Q3 Test 1024x768x16 : 48.2 fps
Q3 Test 1024x768x32 : 39.9 fps
CPU Clockspeed : 800 MHz / 8×100
Sandra CPU : 2160 mips/1097 mflops
Sandra MEDIA : 1928 mmx/1170 fpu
Sandra MEM : 320 cpu/294 fpu
CPUMark 99 : 71.9
FPUmark 99 : 4320
WinMark 1024x768x16 : 360
WinMark 1024x768x32 : 341
Q2 3.20 DM1 1024x768x16 : 65.7 fps
Q2 3.20 CRU 1024x768x16 : 52.8 fps
Q2 3.20 DM1 1024x768x32 : 61.9 fps
Q2 3.20 CRU 1024x768x32 : 50.1 fps
Q3 Test 1024x768x16 : 48.3 fps
Q3 Test 1024x768x32 : 40.2 fps
Benchmark Results Part 2
Matrox G400 Benchmarks at 133 MHz FSB
Test system :
Intel Coppermine ES
Abit BX6 Rev. 2.0 MW BIOS
Corsair 128 MB PC-133
Matrox G400 DH 32 MB at 2/3 FSB
Generic 10 GB HD
Generic 32X CDROM
Windows 98 SE with DirectX 7.0
CPU Clockspeed : 533 MHz / 4×133
Sandra CPU : 1440 mips/719 mflops
Sandra MEDIA : 1285 mmx/780 fpu
Sandra MEM : 343c pu/332 fpu
CPUMark 99 : 51.2
FPUmark 99 : 2880
WinMark 1024x768x16 : 256
WinMark 1024x768x32 : 253
Q2 3.20 DM1 1024x768x16 : 65.5 fps
Q2 3.20 CRU 1024x768x16 : 50.4 fps
Q2 3.20 DM1 1024x768x32 : 61.7 fps
Q2 3.20 CRU 1024x768x32 : 49.2 fps
Q3 Test 1024x768x16 : 47.8 fps
Q3 Test 1024x768x32 : 38.9 fps
CPU Clockspeed : 600 MHz / 4.5×133
Sandra CPU : 1620 mips/811 mflops
Sandra MEDIA : 1446 mmx/877 fpu
Sandra MEM : 345 cpu/332 fpu
CPUMark 99 : 57.5
FPUmark 99 : 3240
WinMark 1024x768x16 : 283
WinMark 1024x768x32 : 280
Q2 3.20 DM1 1024x768x16 : 65.5 fps
Q2 3.20 CRU 1024x768x16 : 51.8 fps
Q2 3.20 DM1 1024x768x32 : 61.7 fps
Q2 3.20 CRU 1024x768x32 : 49.6 fps
Q3 Test 1024x768x16 : 48.1 fps
Q3 Test 1024x768x32 : 39.0 fps
CPU Clockspeed : 666 MHz / 5×133
Sandra CPU : 1800 mips/905 mflops
Sandra MEDIA : 1606 mmx/975 fpu
Sandra MEM : 350 cpu/336 fpu
CPUMark 99 : 63.4
FPUmark 99 : 3600
WinMark 1024x768x16 : 311
WinMark 1024x768x32 : 306
Q2 3.20 DM1 1024x768x16 : 65.5 fps
Q2 3.20 CRU 1024x768x16 : 52.4 fps
Q2 3.20 DM1 1024x768x32 : 61.7 fps
Q2 3.20 CRU 1024x768x32 : 50.0 fps
Q3 Test 1024x768x16 : 48.1 fps
Q3 Test 1024x768x32 : 39.1 fps
CPU Clockspeed : 733 MHz / 5.5×133
Sandra CPU : 1980 mips/998 mflops
Sandra MEDIA : 1767 mmx/1072 fpu
Sandra MEM : 359 cpu/337 fpu
CPUMark 99 : 68.5
FPUmark 99 : 3960
WinMark 1024x768x16 : 338
WinMark 1024x768x32 : 331
Q2 3.20 DM1 1024x768x16 : 65.5 fps
Q2 3.20 CRU 1024x768x16 : 52.7 fps
Q2 3.20 DM1 1024x768x32 : 61.7 fps
Q2 3.20 CRU 1024x768x32 : 50.1 fps
Q3 Test 1024x768x16 : 48.2 fps
Q3 Test 1024x768x32 : 39.2 fps
CPU Clockspeed : 800 MHz / 6×133
Sandra CPU : 2160 mips/1097 mflops
Sandra MEDIA : 1928 mmx/1170 fpu
Sandra MEM : 359 cpu/340 fpu
CPUMark 99 : 74.6
FPUmark 99 : 4320
WinMark 1024x768x16 : 364
WinMark 1024x768x32 : 355
Q2 3.20 DM1 1024x768x16 : 65.5 fps
Q2 3.20 CRU 1024x768x16 : 52.8 fps
Q2 3.20 DM1 1024x768x32 : 61.7 fps
Q2 3.20 CRU 1024x768x32 : 50.2 fps
Q3 Test 1024x768x16 : 48.3 fps
Q3 Test 1024x768x32 : 39.3 fps
Benchmark Results Page 3
3dfx Voodoo3/3000 Benchmarks at 100 MHz FSB
Test system :
Intel Coppermine ES
Abit BX6 Rev. 2.0 MW BIOS
Corsair 128 MB PC-133
3dfx Voodoo3/3000 AGP at 1/1 FSB
Generic 10 GB HD
Generic 32X CDROM
Windows 98 SE with DirectX 7.0
CPU Clockspeed : 500 MHz / 5×100
Sandra CPU : 1349 mips/673 mflops
Sandra MEDIA : 1205 mmx/731 fpu
Sandra MEM : 313 cpu/284 fpu
CPUMark 99 : 48.1
FPUmark 99 : 2700
WinMark 1024x768x16 : 219
WinMark 1024x768x32 : 210
Q2 3.20 DM1 1024x768x16 : 54.1 fps
Q2 3.20 CRU 1024x768x16 : 42.3 fps
Q3 Test 1024x768x16 : 37.9 fps
CPU Clockspeed : 600 MHz / 6×100
Sandra CPU : 1620 mips/811 mflops
Sandra MEDIA : 1446 mmx/877 fpu
Sandra MEM : 313 cpu/284 fpu
CPUMark 99 : 56.4
FPUmark 99 : 3240
WinMark 1024x768x16 : 258
WinMark 1024x768x32 : 253
Q2 3.20 DM1 1024x768x16 : 54.1 fps
Q2 3.20 CRU 1024x768x16 : 42.8 fps
Q3 Test 1024x768x16 : 56.7 fps
CPU Clockspeed : 700 MHz / 7×100
Sandra CPU : 1890 mips/952 mflops
Sandra MEDIA : 1687 mmx/1023 fpu
Sandra MEM : 319 cpu/285 fpu
CPUMark 99 : 64.4
FPUmark 99 : 3780
WinMark 1024x768x16 : 296
WinMark 1024x768x32 : 287
Q2 3.20 DM1 1024x768x16 : 54.1 fps
Q2 3.20 CRU 1024x768x16 : 42.9 fps
Q3 Test 1024x768x16 : 57.3 fps
CPU Clockspeed : 800 MHz / 8×100
Sandra CPU : 2160 mips/1097 mflops
Sandra MEDIA : 1928 mmx/1170 fpu
Sandra MEM : 320 cpu/294 fpu
CPUMark 99 : 71.9
FPUmark 99 : 4320
WinMark 1024x768x16 : 329
WinMark 1024x768x32 : 319
Q2 3.20 DM1 1024x768x16 : 54.1 fps
Q2 3.20 CRU 1024x768x16 : 43.0 fps
Q3 Test 1024x768x16 : 57.8 fps
Benchmark Results Page 4
3dfx Voodoo3/3000 Benchmarks at 133 Mhz FSB
Test system :
Intel Coppermine ES
Abit BX6 Rev. 2.0 MW BIOS
Corsair 128 MB PC-133
3dfx Voodoo3/3000 AGP at 2/3 FSB
Generic 10GB HD
Generic 32X CDROM
Windows 98 SE with DirectX 7.0
CPU Clockspeed : 533 MHz / 4×133
Sandra CPU : 1440 mips/719 mflops
Sandra MEDIA : 1285 mmx/780 fpu
Sandra MEM : 343 cpu/332 fpu
CPUMark 99 : 51.2
FPUmark 99 : 2880
WinMark 1024x768x16 : 236
WinMark 1024x768x32 : 233
Q2 3.20 DM1 1024x768x16 : 54.1 fps
Q2 3.20 CRU 1024x768x16 : 42.8 fps
Q3 Test 1024x768x16 : 56.4 fps
CPU Clockspeed : 600 MHz / 4.5×133
Sandra CPU : 1620 mips/811 mflops
Sandra MEDIA : 1446 mmx/877 fpu
Sandra MEM : 345 cpu/332 fpu
CPUMark 99 : 57.5
FPUmark 99 : 3240
WinMark 1024x768x16 : 262
WinMark 1024x768x32 : 257
Q2 3.20 DM1 1024x768x16 : 54.1 fps
Q2 3.20 CRU 1024x768x16 : 42.9 fps
Q3 Test 1024x768x16 : 57.2 fps
CPU Clockspeed : 666 MHz / 5×133
Sandra CPU : 1800 mips/905 mflops
Sandra MEDIA : 1606 mmx/975 fpu
Sandra MEM : 350 cpu/336 fpu
CPUMark 99 : 63.4
FPUmark 99 : 3600
WinMark 1024x768x16 : 288
WinMark 1024x768x32 : 281
Q2 3.20 DM1 1024x768x16 : 54.3 fps
Q2 3.20 CRU 1024x768x16 : 43.0 fps
Q3 Test 1024x768x16 : 57.8 fps
CPU Clockspeed : 733 MHz / 5.5×133
Sandra CPU : 1980 mips/998 mflops
Sandra MEDIA : 1767 mmx/1072 fpu
Sandra MEM : 359 cpu/337 fpu
CPUMark 99 : 68.5
FPUmark 99 : 3960
WinMark 1024x768x16 : 316
WinMark 1024x768x32 : 309
Q2 3.20 DM1 1024x768x16 : 54.6 fps
Q2 3.20 CRU 1024x768x16 : 43.5 fps
Q3 Test 1024x768x16 : 57.9 fps
CPU Clockspeed : 800 MHz / 6×133
Sandra CPU : 2160 mips/1097 mflops
Sandra MEDIA : 1928 mmx/1170 fpu
Sandra MEM : 359 cpu/340 fpu
CPUMark 99 : 74.6
FPUmark 99 : 4320
WinMark 1024x768x16 : 345
WinMark 1024x768x32 : 337
Q2 3.20 DM1 1024x768x16 : 54.8 fps
Q2 3.20 CRU 1024x768x16 : 43.7 fps
Q3 Test 1024x768x16 : 58.1 fps
Analyzing the Results
Looking at graphs and benchmarks isn’t as interesting as reading a good review filled with descriptions of crisp and clear images, breathtaking graphics or mindboggling framerates. Unfortunately these colorful descriptions don’t tell you much about the real world performance of the actual system, as they are all very subjective. In order to be able to make a sound judgement about the performance of Intel’s newest Pentium III we need to look at the graphs and benchmarks and draw our conclusions.
The graphs for the CPU performance are quite self-explanatory, as the CPU performance simply increases linearly with the clockspeed. The same applies for the Winmark 99 performance, as this benchmark is closely related to the clockspeed of the CPU.
Looking at both Winmark graphs at 100 and 133 MHz busspeed for the Matrox G400 as well as for the 3dfx Voodoo3/3000, we see that the business performance increases with clockspeed, and thus there is a performance increase when running at a higher clockspeed.
If we then look at the graphs titled Matrox G400 DM1 100 MHz, Matrox G400 DM1 133 MHz and Matrox G400 Q3A 100 MHz, Matrox G400 Q3A 133MHz we see that the Matrox G400 isn’t all that CPU dependant, as its framerate has already evened out at 500 MHz. Running the CPU at a higher clockspeed will not further increase the framerate.
Looking at both the graphs titled Matrox G400 CRU 100 MHz and Matrox G400 CRU 133 MHz we see that the performance steadily increases until 700 MHz, above which point the framerate stays within tenths of the previous value. Running the CPU at an even higher clockspeed will not increase the framerate significantly.
If we look at the graphs titled V3/3000 DM1 100 MHz and V3/3000 DM1 133 MHz we see that the same applies here; the Voodoo3/3000 has already reached its top performance and running the CPU at higher clockspeeds will not increase the framerate further.
Looking at the graphs titled V3/3000 CRU 100 MHz and V3/3000 CRU 133 MHz, we see that the performance doesn’t increase more than a few tenths over the entire range. Thus running the CPU at a even higher clockspeed will not increase the framerate significantly.
Looking at the two graphs titled V3/3000 Q3A 100 MHz and V3/3000 Q3A 133 MHz, we see that the performance steadily increases until 700 MHz, above which the framerate stays within tenths of the previous value. Running the CPU at higher clockspeeds will not increase the framerate significantly.
What is interesting to see, however, by looking at the graphs that pit the Matrox G400 against the 3dfx Voodoo3/3000, is that the Matrox G400 offers the most CPU-independent framerate. The difference between a 500 MHz and a 800 MHz Pentium III is negligible, whereas the Voodoo3/3000 is somewhat more susceptible to clockspeed changes.
Another thing that is somewhat expected, as we saw the same with the much-anticipated move from 66 MHz to 100 MHz system bus about a year ago, there is virtually no performance gain when running your busspeed at 133 MHz over the 100 MHz system bus we’ve been using.
One other thing that is quite interesting to see is that the Matrox G400 outperforms the Voodoo3/3000 in all benchmarks except for Quake III Arena. This could be due to driver optimization for or within the game, or that the Matrox G400 lacks such optimization.
Conclusion
Looking back upon the whole range of benchmarks we’ve run, both the Matrox G400 and the 3dfx Voodoo3/3000 show little performance gain from a faster CPU. The sweet spot for both videocards seems to be around 700 MHz, to be able to keep the videocard as well as the system running at top-notch performance. We realize that we are running the new Pentium III on a BX chipset motherboard with PC-133 SDRAM. But if you look at the memory throughput, the BX chipset / PC-133 SDRAM combination is more than holding its own. Furthermore, we’ve run into no troubles, instability or other issues, while conducting our benchmarks.
In summary, Intel once again proved it could push the aging P6 core forwards with the release of its much-anticipated new Pentium III processor, codename Coppermine. They’ve designed a processor that combines the best of all previous generations, using a .18 micron process, Advanced Transfer Cache and Advanced System Buffering, and breathed new life into the P6 family. Combined with the versatility and proven reliability of the BX chipset, the new Pentium III proves to be a very fast CPU that runs faster, cooler and more efficiently than previous generations.
HardwareCentral would like to thank Intel Germany for donating us the Pentium III processor as used in the review and also Matrox and 3dfx for donating us their products for use in this and other reviews.
Sander Sassen.