---- The DOOM 2 Benchmark! ( DEMO2 - 12Dec95) Jeff Rollason ---- This is a games benchmark based on the 3D game DOOM 2. It is a more useful means of assessing a machines potential to run 3D games than conventional benchmarks which just measure particular aspects video and processor performance. DEMO2 is used for the benchmark as it did not change between v1.66 and v1.9 of DOOM 2. The DOOM 2 engine may have changed a little between these versions, but the differences are small in performance. This release of the DOOMBENC results also provides the results of the tests run through the RATING program to test video card performance (See the end of this doc). Please submit your own results for this test. EVEN BETTER, SUBMIT RESULTS FOR YOUR MACHINE WITH DIFFERENT VIDEO CARDS etc.. Make sure that your machine is running at its full potential. Careful tweaking of the CMOS can yield a substantial improvement in machine performance (check that wait states are set to 0, hidden refresh is on, cache enabled etc..) 4.4 1991 386dx33 64 4 16isa Trident Tiny Computers 9.9 10/95 AmdDX100 256 16 ISA Paradise PVGA1A-JK *(3) 10.1 1991 486dx33 0 20 ISA Onboard ET4000 IBM PS/1 10.3 1989 CyrDX66u 64 8 ISA ATI Wonder(integrated) Horthgate slim386 10.5 1992 486sx50 ??? 12 local Paradise Toshiba 4400SX 11.2 1993 486SX33 256 8 VLB Genoa 8500VL Lucky (LSI) 11.2 1994 486dx100 256 20 ISA16 Trident 8900 *(Green)(10) 11.8 6/92 486sx33 256 4 VLB Cirrus CL-GD5422 * 12.2 4/92 486dx36 256 8 ISA9 ET4000 1MB DRAM *(Pacomp) 12.8 1993 486dx33 0 8 VLB Cirrus AVGA3 5424 Opus .12.8 1994 Nx586/90 256 16 ISA16 Tseng ET4000 Alaris/Nexgen(13) .12.8 1994 Nx586/90 256 16 16 ATI Graphics Wonder Alaris/Nexgen(13) 14.1 1994 486dx100 256 20 ISA16 Oak OTI-077 *(Green)(10) 14.8 1992 486dx66 256 16 VLB Diamond Stealth Viper * .15.2 1990 486dx50 256 8 ISA16 NCR 77C22E 2Mb DRAM * 16.5 1993 ibmDX75 256 16 VLB Herc'Graph'Pr0 2Mb Blue Lightning 17.0 1993 486d2-50 0 16 local Onboard 32bit S3 * 18.9 10/95 AmdDX100 256 16 VLB Diamond Stealth Viper *(3) 19.0 1992 486dx66 256 16 VLB ET4000/W32i DRAM *(2) 19.2 12/94 P90 256 32 ISA Tseng ET4000 1MB DRAM *(4) 19.7 1992 486d2-50 256 8 ISA Onboard ET4000 Compaq .21.5 ? CyrDX66 256 8 VLB Diamond Stealth SE * 21.5 6/92 486dx50 256 8 VLB Genoa 8500VL Generic 21.9 1992 486dx66 256 8 VLB Speedstar Pro * .23.3 12/92 486dx66 256 16 VLB ATI Mach32 Gateway 2000(12) 24.6 2/95 486dx66 256 8 PCI sis 86c201 (?) *SIS sg86c201 25.2 1994 486dx66 256 16 VLB SpeedStar Pro * .25.5 1994 Nx586/90 256 16 VLB Herc' Dyn' Pro (ET4) Alaris/Nexgen(13) 26.0 1992 486dx100 256 16 VLB ET4000/W32i DRAM *(2) 26.1 1993 486dx66 256 16 PCI ATI grap'ult' (mach32) GW2K 26.2 1992 486dx66 128 16 VLB Cirrus AVGA3 5428 * 27.1 1995? 486dx66 256 8 VLB ET4000/W32p DRAM * 27.5 4/94 486dx66 256 12 VLB ET4000/W32i DRAM Dan Technology(8) 27.7 3/94 486dx100 256 16 VLB Herc' Dyn' Power (ET4) * 27.8 10/94 486dx66 256 32 PCI ATI Mach64 2MB DRam * 28.2 6/94 AmdDX80 256 16 VLB Orchid Kelvin64 2-DRAM * .28.3 1994 486dx66 128 16 VLB Tseng ET4000 1MB Dell 466 MXV 28.4 4/94 486dx66 256 12 VLB Diamond Stealth32-VRAM Dan Technology(8) 29.2 95? P100 256 8 PCI Trident PT540 Atlantic Systems 29.3 1993 486dx100 256 16 VLB Cirrus 5426(integrated)AST Advantage Pro .30.2 12/92 P83ovdr 256 16 VLB ATI Mach32 Gateway 2000(12) 31.4 10/95 AmdDX100 256 16 VLB ET4000/W32i DRAM *(3) 31.6 1994 AmdDX100 256 16 VLB ATI Mach64 -2meg * 31.7 1995 P75 256 16 PCI Trident 9440 RM (Microstar mobo) 31.9 1994 P66 256 8 PCI ATI GX Mach 64 VRAM Hi-Grade-SIS(1) 31.9 1994 P66 256 8 PCI ET4000/W32p DRAM Hi-Grade-SIS(1) 32.5 12/94 P75 256 32 PCI33 Herc' Dyn' Power (ET4) *(4) 32.7 10/95 AmdDX100 256 16 VLB Herc' Dyn' Power (ET4) *(3) 33.0 1994 486dx100 256 16 VLB Diamond Stealth64-DRAM * 33.2 11/92 P90 512 32 VLB Herc' Dyn' Power (ET4) *Opti 33.3 1994 P66 256 8 PCI Diamond Stealth64-DRAM Hi-Grade-SIS(1) 34.0 1994 P90 256 8 PCI Diamond Stealth64-DRAM Vale Platinum 34.0 1994 486dx100 256 20 VLB ET4000/W32i DRAM *(Green)(10) 35.1 10/94 486dx100 256 12 VLB Diamond Stealth32(ET4) * 35.3 1994 486dx100 256 12 VLB Diamond Stealth64-DRAM Gateway486-33 36.0 1994 486dx100 256 16 PCI ATI WinTurbo Mach64-Vr * 35.8 12/94 P90 0 32 PCI30 Herc' Dyn' Power (ET4) *(4) 37.4 1/95 486dx100 512 16 VLB Herc' Dyn' Pro (ET4) * 38.3 1995 486DX100 256 16 PCI Diamond Stealth64-DRAM GatewayP4D(Saturn2) 38.4 1994 486dx100 256 16 VLB #9GXE64(S3-864) Dell Dimension .38.8 1995 P75 256 8 PCI ET4000/W32p DRAM Hi-Grade (TMC) .39.0 1995 P75 256b 16 PCI Cirrus Logic CL-GD5434 * SIS 39.0 1995 AmdDX120 256 20 PCI Diamond Stealth64-VRAM * 39.0 12/94 P90 256 32 PCI30 Herc' Dyn' Power (ET4) *(4) 39.3 5/95 P90 256 32 PCI ATI Graph'Pro Turbo/VR *(plato) 40.9 12/94 P90 256 16 PCI Diamond Stealth64-VRAM Zeos Pantera 41.4 12/94 P90 256 16 PCI Diamond Stealth64-DRAM Micron Po'Sta'+ 41.6 10/95 P75 256b 32 PCI Herc' Grap' Term' (S3) *(9) Triton 41.8 1992 P90 256? 32 PCI Spider Tarantula 64 *(Plato) 41.9 3/95 P90 512 16 PCI Diamond Stealth64-DRAM Micron 42.4 1992 P90 256 24 PCI Diamond Stealth64-VRAM Gateway 2000 43.0 11/92 P90 256 8 PCI #9GXE64Pro 2meg Dell Dimension 43.0 6/95 P90 256 16 PCI Diamond Stealth64-DRAM * 43.1 12/94 P100 256 32 PCI33 Herc' Dyn' Power (ET4) *(4) 43.4 10/94 P90 256 8 PCI Alpine CL-GD5434 Packard Bell .45.0 ??/95 P133 256b 16e PCI Cirrus .. Atlantic Coast 45.3 12/94 P100 256 32 PCI33 Herc' Grap' Term' (S3) *(4) 47.5 P100 6/95 256 16 PCI ATI Mach64-DRAM Quantex Microsy' 48.8 11/94 P100 256 16 PCI Hercules Stingray 64/V *(plato) 49.2 8/94 P90 256 16 PCI Diamond Stealth64-DRAM * 49.3 05/95 P100 256b 16 PCI Matrox Millenium 2m WR *(5) 49.8 10/95 P90 256b 32 PCI Herc' Grap' Term' (S3) *(9) Triton 50.7 1995 P100 512 32 PCI Diamond Stealth64-VRAM * 50.1 1995 P100 256 16 PCI ET4000/W32p DRAM 2-meg Hi-Grade TMC+Triton 51.0 05/95 P100 256b 16 PCI Diamond Stealth64-VRAM *(5) 52.1 3/95 P100 256b 16e PCI #9 Motion 771 2m Dell Dimension 52.7 10/94 P100 256b 32 PCI33 Diamond Stealth64-VRAM Supermicro P55CMS 53.2 95? P100 256 16 PCI Diamond StealthPr-VRAM *(Plato) 54.0 1995 P120 256b 8e PCI ATI Mach64 2Mb DRAM Gateway(7*) 54.3 1995 P100 256b 16e PCI Diamond Stealth64-DRAM Micron Millenia 54.6 1995 P120 256b 16e PCI ATI Mach64 2Mb VRAM Gateway(7*) 54.7 10/95 P100 256b 32 PCI Herc' Grap' Term' (S3) *(9) Triton 55.1 05/95 P100 256b 16 PCI Hercules Stingray 64/V *(5) 55.4 5/95 P120 256 64e PCI Diamond Stealth64-VRAM Micron Millenia 55.8 3/95 P120 256b 16e PCI Diamond Stealth64-VRAM Micron Millenia120 56.6 3/95 P133 256 16e PCI ATI Mach64-VRAM 2m Micron P5-133(6) 56.8 1995 P133 256 32e PCI Number 9 Motion 771 *ASUSTeK mobo (11) 56.8 1995 P133 256 32e PCI Matrox Millenium 2m WR *ASUSTeK mobo (11) 57.5 10/95 P120 256b 32 PCI Herc' Grap' Term' (S3) *(9) Triton 59.8 1995 P133 256 32e PCI Diamond Stealth64-DRAM *ASUSTeK mobo (11) 59.8 1995 P133 256 32e PCI D.Stealth Vid64-DRAM *ASUSTeK mobo (11) 60.8 5/95 P133 256b 16e PCI Diamond Stealth64-VRAM Micron Millenia133 60.9 3/95 P133 256b 24e PCI Diamond Stealth64-VRAM Micron P5-133(6) 61.1 8/95 P133 256b 32 PCI Spea V7 P64Vid DRAM S3 *(9) Triton 62.5 8/95 P133 256b 32 PCI Diamond Stealth64-VRAM *(9) Triton 62.9 8/95 P133 256b 32 PCI Herc' Grap' Term' VRAM *(9) Triton 63.4 8/95 P133 256b 32 PCI Spea V7 P32 DRAM S3 *(9) Triton 63.4 8/95 P133 256b 32 PCI Herc' Dyn' Power (ET4) *(9) Triton 63.6 8/95 P133 256b 32 PCI Herc' Grap' Term' (S3) *(9) Triton 64.1 3/95 P133 256 16e PCI ATI Mach64-VRAM Gateway 2000 P5 .64.1 ?/95 P133 256b 48 PCI Diamond Stealth64V-VRM Ten Tech (Triton) 69.0 ?/95 P133 256b 16e PCI Hercules Stingray 64/V *Supermicro Mobo Results preceded by "." are new results. "?" = details to follow or unknown. "(n)" indicates same machine tested with differing components. Cache figures with "b" suffix indicate "burst" (e.g. 256b). Ram with "e" suffix indicate EDO ram. ISA16 indicates 16-bit ISA card instead of 8-bit. PCI33 indicates PCI running at 33MHz (instead of 30 MHz). ISA can be run at 7.5MHz or 10MHz etc. "*" for PC make indicates "Home brew". The figures above assume the fastest ISA rate available. "local" indicates a proprietry local bus, as used in some portables and earlier desktop machines. Figures should be for machine in normal configuration (not clean boot), running from DOS (not a DOS Window). Win95 allows Windows to be shut down to plain DOS. Please distinguish Stealth32 from Stealth64. They are completely different cards! -------- How to run the Benchmark --------- The frames per second is obtained by running: DOOM2 -NOSOUND -TIMEDEMO DEMO2 This prints out gametics and realtics to run the entire demo (2). Ignore the "cannot open demo2.lmp" message. DEMO 2 is now used as this is the only DEMO which has not changed with DOOM version number. fps = (gametics/realtics)*35 Figures are for "no sound", screen size the largest with status bar and full detail. L2che is the external cache, not the on-chip cpu cache. The BIOS date is displayed when your PC starts up (the BIOS date almost certainly does not matter, but it is an easy indication of the age of the motherboard). -------- How do I get my results in the listing? -------- Post your figures to me directly on CompuServe 100031,3537 (or internet 100031.3537@compuserve.com), otherwise I will miss them. --------- Conclusions --------- Buying a machine for 3D games is hard without advice or experience. Published benchmarks may not be helpful, for example the notorious 3DBENCH program looks convincing but when applied to 2 machines that differed in their DOOM fps speed by 2.4 : 1, 3DBENCH only managed to show a variance of 5% between them You may want to interpret the information above as you choose, but I have tried to glean a few conclusions. ------- Notes and Wisdoms(?) 1. It seems that the choice and age of the motherboard is evidently critical for the 486 based systems. The variations in performance are substantial. 2. Overdrive processors with old motherboards are probably a waste of time. The figure 26.0 above was an Intel 486dx100 overdrive, which I returned and exchanged for an AmdDx100 with a new motherboard (actually much cheaper!), yielding a somewhat improved 31.4 and leaving me with a spare motherboard. 3. Video Cards: The Trident PT540 looks like a card to avoid! The P100 with this card passed the cpu benchtests but did very badly in the fps rating. The home brew(3) machine above was tried with 4 video cards, showing a 3.2 : 1 variation in fps performance, confirming the importance of the choice of video card. In particular DOOM ran half as fast with a Diamond Stealth Viper-VRAM than a old cheap generic DRAM ET4000/W32. Note that Diamond keep changing chipsets: the Viper to Stealth32 to Stealth64 used Weitek, ET4000/W32 then S3. Note that cards that use interleaved RAM almost certainly do not offer extra benefits for DOS games (e.g. Hercules Dynamite Power, whose interleaving needs 2 meg of ram and whose benefits are only available to Windows drivers etc.). 4. ISA bus? Don't even think about using an ISA-bus video card!! 5. PCI or VLB? The conventional wisdom is that PCI is better, with its technical advantages. However I have examined 4 magazine reviews with compare the same cards based on VLB and PCI, and all (to their surprise) concluded that the VLB cards were generally faster, sometimes by up to 30%. This may be attributable to the fact that the processor actually sits on the VLB whereas PCI has some buffered connection with a separate processor bus. Also VLB will run at the speed that the processor runs (usually 33MHz) whereas PCI usually runs at 30MHz. It appears that P66, P100 and P133 machines run PCI at 33MHz. Maybe all this will change, but at the moment (aug95) the 486 VLB machines may have an advantage that the Pentium PCI machines do not have. A later input on VLB and PCI was provided by Paul Schmitz [71776,337] Sep95 who tried a three PCI and three VLB boards when choosing his DX4-based machine. He found that the PCI boards were all faster the the VLB boards, which suggests that perhaps PCI has now sorted out any problems it did have. 6. The Intel 486dx100 machines above seem to be some 10% faster than the AmdDX100 machines. This could be attributed to the larger 16k cache in the Intel chip compared to the 8k of the Amd version. The Intel chip also has a beefed-up integer multiply which is 3-5 times faster than a DX2-100 would be. Amd now have a 120MHz version out with two types of cache. The theoretically faster Write-Back version of the latter seems to be actually between 1% slower and 6% faster. 8. Memory? Almost all machines above have 8meg+. None seem to show any obvious advantage having 16meg. One user reports though that he has 32 megs of RAM and lets DOOM sit in his 14meg Smartdrive cache! This undoubtably kills those minor jerks as DOOM loads each new scene. 9. What should I buy? With limited funds a $160 AmdDx100 with motherboard looks an economic base for a machine. If funds are less tight the P100 looks good value at the moment (Aug95). Wait until the P6 is out! The Pentium price will be bound to drop. If funds really are tight, then a 486dx66 on a newer motherboard will still perform well. If you have the money for a Pentium then the Supermicro and Micronics motherboards look good. 10. NexGen P90? These look very cheap, but I was warned off buying one because of user reported compatibility problems. NexGen now have a P133. From a bench result received 8dec95 it looks as if a VLB NexGen P90 is about the same as a 486DX66 in performance! Also the board does not seem to run Heretic, but works with DOOM and Descent. That looks like a firm thumbs down to me! 11. Windows 95? It seems that Windows 95 shutdown to DOS is as fast as a pure DOS machine, whereas Windows 95 running a DOS session will probably be 20% slower (provide figures for shutdown Windows). 12. Asynchronous or pipelined burst cache? Steffen Itterheim [100606,2141] had the patience to ty out a number of different hardware, clock and video combinations for these tables. For a P90, the pipelined cache added about 5% on the fps performance. For a P133 this increased to 6%. 13. Two very consistent independent tests looking at fps performance against clock rate have shown the following results. These came from (1) Steffen above and (2) and (3) Timothy Zakharov [73773,3047]. These results looked at Pentium P75, P90, P100, P120 and P133 with (1) Graphite Terminator, (3) Hercules Stingray + Triton and (3) Stingray + Supermicro motherboard with pipeline burst cache. clock fps[1] fps[2] fps[3] ratio ratio ratio ratio P75 1.0 1.0 1.0 1.0 P90 1.2 1.2 1.19 1.19 P100 1.33 1.31 1.32 1.32 P120 1.6 1.38 1.38 1.41 P133 1.77 1.53 1.53 1.55 P120->133 1.11 1.11 1.11 1.10 From these tests above it can be seen that the P120 offers a poor performance return for the clock rate. This is (I believe) because the 120 uses a 60MHz bus whereas the P100 and P133 both use 66 MHz. Unless you can get a very favourable price for a P120, then maybe only the only upgrade from 100 is the 133. Thats' enough for now. Jeff Rollason Author of the Shareware "PocketD Plus" - The tiny but advanced file manager. "BEST UTILITY" award winner 92, 94 and 95. Download from CIS as POCK.ZIP. ------------------------Results of RATING program------------------------- This uses the RATING program and provides a way of comparing large numbers of inter-related results. An earlier version of this test included all instances where cards were tested on similar machine types. This proved to be too vulnerable to varying performance of one 486/Pentium compared to another, so I have restricted the input to examine results where groups of video cards were tested on a common physical machine. This comparision is only valid where all results can be linked by a chain of common video cards or the same testing platform. Video card Rating 1. Hercules Stingray64/V 1087 2. Diamond Stealth32-VRAM 1032 3. Hercules Graphite Terminator 1028 4. Diamond Stealth64-DRAM 1027 5. Diamond Stealth64-Video64 1025 6. Spea V7 P32-DRAM 1020 7. Hercules Dynamite Power 1018 8. Hercules Graphite Terminator-VRAM 1012 9. Diamond Stealth64-VRAM 1009 10. ET4000/W32 991 11. Spea V7 P64-DRAM 983 12. ATI Mach64 DRAM 976 13. Number-9 Motion-771 974 14. Matrox Millenium 973 15. ATI Mach64 VRAM 964 16. Diamond Stealth Viper 596 ---------------------- File used as input to RATING ---------------------- This is provided with this release as DOOMBENC.RAT. The RATING program and instuctions can be downloaded as RATING.ZIP.