The Ultimate GPU Benchmark (2000 - 2002)

Published: 23rd February 2024 (last update 23rd February 2024 )

No more modern graphic cards, let's ge back a bit to revisit some trully famous pieces of history. In this article, we will get back to year 2000. Starting with Voodoo4 and up to GeForce4 Ti and Radeon 9000 Series.

Introduction

This project has a complicated history. It all started in 2012, yes, that's right - 12 years ago. At start the methodology was very time consuming. 12 settings per game and card across almost 30 games. Also at the time this was one very large project with very wide range of GPUs. The weakest ones were planned to be Rage 128 Pro / Riva TNT2 Ultra and Voodoo4. The cut-off point at the other end was meant to be Radeon X850 XT and GeForce 6800 Ultra. Later, I extended the top end to Radeon X1800 XL and GeForce 7800 GT, as these cards are not that different performance wise and in different articles they are completely in a shadow of GeForce 8 and Radeon HD Series.

Obviously, it is pretty much impossible to create a methodology that would cover such wide selection of hardware without CPU bottlenecks and using "modern" technologies like shaders. Therefore I selected games and settings that would be suitable for older GPUs and different set of games / settings for the more modern ones. Everything in one results spreadsheet. The weak HW would only run the less demanding games and the more powerful category would run everything. With the methodology pretty much complete, I started the benchmarking. You can imagine it was very time consuming. The years went by and the finish line was still far away. It took few dozen hours to complete one GPU... and there were 50 of them :D

I took many breaks over time, spent time on another projects and this one remained unfinished. During that time I worked on the succeeding series of GPU benchmarks starting at 2004 and up to 2013 (these are all complete and published). While doing the later benchmarks, I figured it would be good to backport some new ideas to the old methodology for AGP cards. 3DMark feature tests, AA/AF performance loss and even recording footage for GPU Duels (at my YT channel) - all this was added. Of course it required some amount of rebenching and increasing the time requirements even further. At one point, I decided to scrap the whole thing and leave it unfinished as the amount of time needed was insane.

But ultimately I went over the methodology again and did number of cuts to greatly reduce the amount of work to finish. Some games were cut out completely (inconsistent performance or difficult to measure). The two categories of GPUs got separated and less demanding games were removed from the faster HW group. There is not much point in benchmarking Quake 3 on GeForce 6800, is it? :D And the last measure that saved plenty of time - I didn't measure fps in games / settings that scored less than 10 fps. These steps helped me to finish the project in reasonable amount of time. This is the part one with GPUs up to GeForce4 Ti 4600 and Radeon 9500. Cut-off year is 2002. Later, there will be second article with DX8 and DX9 hardware.

Test System - Hardware

The test system was chosen to provide the best possible performance for AGP cards. ASRock 4CoreDual-SATA2 with overclocked Pentium E5700 should be more than enough. As a side effect, PCI-E cards can be tested on this system as well, albeit with slightly reduced performance (only 4 PCI-E lanes and slightly slower CPU). The CPU has BSEL mod applied to report FSB 266 (1066) as default. This allows memory divider 1:1 to be used and increases memory performance greatly. Apparently, the PT880 Ultra chipset is unstable at FSB > 266 MHz if PCI-E GPU is used. Also this board for some reason doesn't allow "half" multipliers. Therefore Fro PCI-E, 14×266 was used as a closest possible frequency.

Voodoo5 5500 AGP and Kyro 2 can't run on AGP 4×/8× board. And due to some incompatibility, neither can GeForce GT 610 PCI (but this one will be in the next article). I took Pentium 4 Northwood on SiS 651 chipset that is the fastest platform with 3.3V AGP support.

For AGP 4×/8× cards

Pentium E5700 @ 3.9 GHz (2C/2T, fixed clock)
ASRock 4CoreDual-SATA2
VIA PT880 Ultra Chipset (AGP 4×/8×)
2 × 1 GB DDR2 600 CL4-4-4-12-1T
SCSI 80 GB / 10000 rpm HDD
FSP 350W
Acer AL2021 (1600×1200)

For PCI-E cards

Pentium E5700 @ 3.73 GHz (2C/2T, fixed clock)
ASRock 4CoreDual-SATA2
VIA PT880 Ultra Chipset (PCI-E 1.0 ×4)
2 × 1 GB DDR2 533 CL4-4-4-12-1T
SCSI 80 GB / 10000 rpm HDD
FSP 350W
Acer AL2021 (1600×1200)

For Voodoo5 5500 AGP, Kyro 2, GeForce GT610 PCI

Pentium 4 2.8 Northwood @ 3.2 GHz (1C/1T, fixed clock)
Gigabyte GA-8SIMLH-P
SiS 651 Chipset (AGP 2×/4×)
2 × 512 MB DDR 400 CL2.5-3-3-6
Crucial BX500 120 GB
FSP 400W
Acer AL2021 (1600×1200)

Test System - OS and Drivers

Windows XP SP3
ForceWare 71.89 (for GeForce 256, 2, 3, 4, FX)
Catalyst 6.11 (for Radeon 7000, 8000, 9000-9200 Series)
Catalyst 10.2 (for Radeon 9500)
...
For all GPUs high quality filtering (if possible), vsync off

Test System - Games

Call of Duty [2003, OGL]
Doom 3 [2004, OGL]
Far Cry [2004, DX7/8/9]
FlatOut 2 [2006, DX7/8/9]
GTA 3 [2001, DX7/8]
GTA Vice City [2002, DX7/8]
Half Life 2 [2004, DX7/8/9]
Mafia: The City of Lost Heaven [2002, DX7]
Need for Speed: Underground [2003, DX7/8]
Quake 3 Arena [1999, OGL]
Return to Castle Wolfenstein [2001, OGL]
Serious Sam: SE [2002, OGL]

Tested Video Cards

	Voodoo4 4500 PCI	Voodoo4 4500 AGP	Voodoo4 4800 AGP	Voodoo5 5500 AGP	P650	Parhelia AGP 8×	Radeon 7000	Radeon SDR
GPU	VSA-100	VSA-100	VSA-100	2 × VSA-100	Parhelia-LX	Parhelia-512	RV100	R100
Architecture	VSA	VSA	VSA	VSA	MGA	MGA	R1xx	R1xx
Technology	250 nm	250 nm	250 nm	250 nm	150 nm	150 nm	180 nm	180 nm
Die Size	112 mm²	112 mm²	112 mm²	2 × 112 mm²	-	174 mm²	80 mm²	114 mm²
Transistor Count	14 mil.	14 mil.	14 mil.	2 × 14 mil.	80 mil.	80 mil.	-	30 mil.
Transistor Density	0.12 mil. / mm²	0.12 mil. / mm²	0.12 mil. / mm²	0.12 mil. / mm²	-	0.46 mil. / mm²	-	0.26 mil. / mm²
GPU Clock	166 MHz	166 MHz	183 MHz	166 MHz	230 MHz	250 MHz	183 MHz	160 MHz
ROPs	2	2	2	2 × 2	2	4	1	2
TMUs	2	2	2	2 × 2	8	16	3	6
Shaders	-	-	-	-	2 PS + 2 VS	4 PS + 4 VS	-	-
Memory	32 MB SDR	32 MB SDR	64 MB SDR	2 × 32 MB SDR	64 MB DDR	128 MB DDR	32 MB DDR	32 MB SDR
Memory Clock	166 MHz	166 MHz	183 MHz	166 MHz	460 MHz	600 MHz	366 MHz	160 MHz
Bus Width	128 bit	128 bit	128 bit	2 × 128 bit	128 bit	256 bit	64 bit	128 bit
Memory Bandwidth	2.7 GB/s	2.7 GB/s	2.9 GB/s	2 × 2.7 GB/s	7.4 GB/s	19.2 GB/s	2.9 GB/s	2.6 GB/s
Fillrate (Pixel)	332 MP/s	332 MP/s	366 MP/s	2 × 332 MP/s	460 MP/s	1000 MP/s	183 MP/s	320 MP/s
Fillrate (Texel)	332 MT/s	332 MT/s	366 MT/s	2 × 332 MT/s	1840 MT/s	4000 MT/s	549 MT/s	960 MT/s
Bus Type	PCI	AGP 4×	AGP 4×	AGP	AGP 8×	AGP 8×	AGP 4×	AGP 4×
TDP	15 W	15 W	15 W	30 W	-	-	-	30 W
DirectX	6/7	6/7	6/7	6/7	8.1	8.1	7	7
OpenGL	1.1	1.1	1.1	1.1	1.3	1.3	1.3	1.3
Launch Year	2000	2000	2000	2000	2003	2002	2001	2000

	Radeon DDR VIVO	Radeon 7500	Radeon 8500	Radeon 9000	Radeon 9000 Pro	Radeon 9100	Radeon 9200 SE	Radeon 9500
GPU	R100	RV200	R200	RV250	RV250	R200	RV250	R300
Architecture	R1xx	R1xx	R2xx	R2xx	R2xx	R2xx	R2xx	R3xx / 4xx
Technology	180 nm	150 nm	150 nm	150 nm	150 nm	150 nm	150 nm	150 nm
Die Size	114 mm²	84 mm²	140 mm²	81 mm²	81 mm²	140 mm²	81 mm²	218 mm²
Transistor Count	30 mil.	30 mil.	60 mil.	36 mil.	36 mil.	60 mil.	36 mil.	107 mil.
Transistor Density	0.26 mil. / mm²	0.36 mil. / mm²	0.43 mil. / mm²	0.44 mil. / mm²	0.44 mil. / mm²	0.43 mil. / mm²	0.44 mil. / mm²	0.49 mil. / mm²
GPU Clock	183 MHz	290 MHz	275 MHz	250 MHz	275 MHz	250 MHz	200 MHz	275 MHz
ROPs	2	2	4	4	4	4	4	4
TMUs	6	6	8	4	4	8	4	4
Shaders	-	-	4 PS + 2 VS	4 PS + 1 VS	4 PS + 1 VS	4 PS + 2 VS	4 PS + 1 VS	4 PS + 4 VS
Memory	64 MB DDR	64 MB DDR	64 MB DDR	64 MB DDR	64 MB DDR	128 MB DDR	128 MB DDR	64 MB DDR
Memory Clock	366 MHz	460 MHz	550 MHz	400 MHz	550 MHz	500 MHz	333 MHz	540 MHz
Bus Width	128 bit	128 bit	128 bit	128 bit	128 bit	128 bit	64 bit	128 bit
Memory Bandwidth	5.9 GB/s	7.4 GB/s	8.8 GB/s	6.4 GB/s	8.8 GB/s	8 GB/s	2.7 GB/s	8.6 GB/s
Fillrate (Pixel)	366 MP/s	580 MP/s	1100 MP/s	1000 MP/s	1100 MP/s	1000 MP/s	800 MP/s	1100 MP/s
Fillrate (Texel)	1098 MT/s	1740 MT/s	2200 MT/s	1000 MT/s	1100 MT/s	2000 MT/s	800 MT/s	1100 MT/s
Bus Type	AGP 4×	AGP 4×	AGP 4×	AGP 4×	AGP 4×	AGP 4×	AGP 8×	AGP 8×
TDP	30 W	-	-	-	-	-	-	-
DirectX	7	7	8.1	8.1	8.1	8.1	8.1	9.0
OpenGL	1.3	1.3	1.4	1.4	1.4	1.4	1.4	2.0
Launch Year	2000	2001	2001	2002	2002	2003	2003	2002

	GeForce 256 SDR	GeForce 256 DDR	Quadro SDR	GeForce2 MX	GeForce2 MX 400	GeForce2 GTS	GeForce2 Ultra	GeForce3 Ti200
GPU	NV10	NV10	NV10	NV11	NV11	NV15	NV15	NV20
Architecture	NV1x	NV1x	NV1x	NV1x	NV1x	NV1x	NV1x	NV2x
Technology	220 nm	220 nm	220 nm	180 nm	180 nm	180 nm	180 nm	150 nm
Die Size	139 mm²	139 mm²	139 mm²	65 mm²	65 mm²	89 mm²	89 mm²	145 mm²
Transistor Count	23 mil.	23 mil.	23 mil.	19 mil.	19 mil.	25 mil.	25 mil.	57 mil.
Transistor Density	0.17 mil. / mm²	0.17 mil. / mm²	0.17 mil. / mm²	0.29 mil. / mm²	0.29 mil. / mm²	0.28 mil. / mm²	0.28 mil. / mm²	0.39 mil. / mm²
GPU Clock	120 MHz	120 MHz	135 MHz	175 MHz	200 MHz	200 MHz	250 MHz	175 MHz
ROPs	4	4	4	2	2	4	4	4
TMUs	4	4	4	4	4	8	8	8
Shaders	-	-	-	-	-	-	-	4 PS + 1 VS
Memory	32 MB SDR	32 MB DDR	64 MB SDR	32 MB SDR	64 MB SDR	32 MB DDR	64 MB DDR	64 MB DDR
Memory Clock	166 MHz	300 MHz	166 MHz	166 MHz	200 MHz	333 MHz	460 MHz	400 MHz
Bus Width	128 bit	128 bit	128 bit	128 bit	128 bit	128 bit	128 bit	128 bit
Memory Bandwidth	2.7 GB/s	4.8 GB/s	2.7 GB/s	2.7 GB/s	3.2 GB/s	5.3 GB/s	7.4 GB/s	6.4 GB/s
Fillrate (Pixel)	480 MP/s	480 MP/s	540 MP/s	350 MP/s	400 MP/s	800 MP/s	1000 MP/s	700 MP/s
Fillrate (Texel)	480 MT/s	480 MT/s	540 MT/s	700 MT/s	800 MT/s	1600 MT/s	2000 MT/s	1400 MT/s
Bus Type	AGP 4×	AGP 4×	AGP 4×	AGP 4×	AGP 4×	AGP 4×	AGP 4×	AGP 4×
TDP	20 W	20 W	20 W	18 W	20 W	33 W	-	35 W
DirectX	7	7	7	7	7	7	7	8
OpenGL	1.2	1.2	1.2	1.2	1.2	1.2	1.2	1.3
Launch Year	1999	1999	1999	2000	2001	2000	2000	2001

	GeForce3 Ti500	GeForce4 MX440 SE	GeForce4 MX460	GeForce4 Ti4200	GeForce4 Ti4600	GeForce FX 5200 64-bit	GeForce FX 5200	GeForce FX 5600	Kyro 2
GPU	NV20	NV18	NV17	NV25	NV25	NV34	NV34	NV31	STG4500
Architecture	NV2x	NV1x	NV1x	NV2x	NV2x	NV3x	NV3x	NV3x	Kyro
Technology	150 nm	150 nm	150 nm	150 nm	150 nm	150 nm	150 nm	150 nm	180 nm
Die Size	145 mm²	65 mm²	65 mm²	150 mm²	150 mm²	91 mm²	91 mm²	135 mm²	71 mm²
Transistor Count	57 mil.	29 mil.	29 mil.	63 mil.	63 mil.	47 mil.	47 mil.	80 mil.	15 mil.
Transistor Density	0.39 mil. / mm²	-	-	0.42 mil. / mm²	0.42 mil. / mm²	0.52 mil. / mm²	0.52 mil. / mm²	0.59 mil. / mm²	0.21 mil. / mm²
GPU Clock	240 MHz	275 MHz	300 MHz	250 MHz	300 MHz	250 MHz	250 MHz	325 MHz	175 MHz
ROPs	4	2	2	4	4	4	4	4	2
TMUs	8	4	4	8	8	4	4	4	2
Shaders	4 PS + 1 VS	-	-	4 PS + 2 VS	4 PS + 2 VS	4 PS + 2 VS	4 PS + 2 VS	4 PS + 2 VS	-
Memory	64 MB DDR	64 MB DDR	64 MB DDR	64 MB DDR	128 MB DDR	128 MB DDR	128 MB DDR	128 MB DDR	64 MB SDR
Memory Clock	500 MHz	266 MHz	550 MHz	512 MHz	650 MHz	333 MHz	400 MHz	550 MHz	175 MHz
Bus Width	128 bit	64 bit	128 bit	128 bit	128 bit	64 bit	128 bit	128 bit	128 bit
Memory Bandwidth	8 GB/s	2.1 GB/s	8.8 GB/s	8.2 GB/s	10.4 GB/s	2.7 GB/s	6.4 GB/s	8.8 GB/s	2.8 GB/s
Fillrate (Pixel)	960 MP/s	550 MP/s	600 MP/s	1000 MP/s	1200 MP/s	1000 MP/s	1000 MP/s	1300 MP/s	350 MP/s
Fillrate (Texel)	1920 MT/s	1100 MT/s	1200 MT/s	2000 MT/s	2400 MT/s	1000 MT/s	1000 MT/s	1300 MT/s	350 MT/s
Bus Type	AGP 4×	AGP 8×	AGP 4×	AGP 4×	AGP 4×	AGP 8×	AGP 8×	AGP 8×	AGP 2×
TDP	43 W	-	25 W	33 W	45 W	25 W	25 W	37 W	10 W
DirectX	8	7	7	8.1	8.1	9.0	9.0	9.0	7
OpenGL	1.3	1.2	1.2	1.3	1.3	1.5	1.5	1.5	1.2
Launch Year	2001	2002??	2002	2002	2002	2003	2003	2003	2001