ÉÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍ» º º º Classic Above Board Accelerator Board Compatibility: L-Z º º º ÈÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍͼ ORCHID'S JET 386 *CUSTOMER REPORTS* of limited compatibility with Above Boards and Matched Memory Classic. It is necessary to boot the system with the accelerator in 286 mode with an Above Board or Matched Memory Classic in the system. If you boot in 386 mode, EMM.SYS will give "Error Msg. 7" with older versions of EMM.SYS. MICROSOFT MACH 20 *CUSTOMER REPORTS* of compatibility with Above Boards and Matched Memory Classic. For Expanded memory to work with the Mach 20 board, the MACH20.SYS driver must appear BEFORE the EMM.SYS driver in the CONFIG.SYS. If the EMM.SYS driver is first, the expanded memory will not be recognized after the MACH20 driver has been invoked. MCT SPEED DEMON We have mixed customer reports about compatibility with Above Boards and the Matched Memory Classic. One report saw the EMM message "a switch incorrectly indicates that you have an 8087." This may have been due to the motherboard 8087 socket not being "plugged" (See the Accelerator General Information section). The board is compatible per another customer report. However, switch #5 on the MCT board needs to be turned off when using an expanded memory board. This is documented in the Speed Demon manual. Turning the switch off disables part of the cache memory. MICROWAY NUMBERSMASHER This accelerator was tested by Intel and found to be compatible with Above Boards and Matched Memory Classic. Note: This accelerator seems to be the exact same board as the UNIVATION PC TURBOCHARGER. MOUNTAIN RACECARD *CUSTOMER REPORTS* of compatibility with Above Boards and Matched Memory Classic. Note: This apparently is the same board as the PC TECHNOLOGIES 286 EXPRESS, which we have confirmed as being compatible with Intel Above Boards. ORCHID TURBO 186 Confirmed INCOMPATIBLE with Above Boards and Matched Memory Classic. This board is no longer being produced by Orchid. ORCHID TINY TURBO 286 *CUSTOMER REPORTS* of compatibility with current Above Boards and the Matched Memory Classic. We also have reports of compatibility with the discontinued Above Board 286, PS/286, PC, & PS/PC. Tech Notes: SETBOARD won't run in an XT with this board set in 286 mode. There is a toggle switch on the Tiny Turbo that disables the 286 and enables the 8088. Flip this switch and SETBOARD runs just fine. Some Tiny Turbos have a memory cache that must be disabled for an Above Board to work. NOTE: Be sure to tell SETBOARD and SOFTSET that you are in an 8088- based machine. This board has been used by 2 or 3 people in PCED and it works with the Above Board PC and PS/PC. There have been problems found when using a math coprocessor on the Turbo board and also using an Above Board. The problem can be solved by placing a dummy chip in the 8087 socket on the motherboard. Customers should call Orchid and ask for a dummy chip designed to work with the Tiny Turbo. ORCHID TURBO 286 *CUSTOMER REPORTS* of compatibility with current Above Boards and Matched Memory Classic. ORCHID TURBO 286E *CUSTOMER REPORTS* that this full length accelerator board, (with cache memory), is compatible with current Above Boards and Matched Memory Classic. We also have reports of compatibility with the discontinued Above Board 286, PS/286, PC, and PS/PC. Tech Notes: The old TESTABPC does not work when the cache memory is active. We have no reports with the current TESTAB program. PC TECHNOLOGIES 286 EXPRESS *CUSTOMER REPORTS* of compatibility with current Above Boards, Matched Memory Classic, and the discontinued Above Board 286 and PS/286. Tech Notes: If SETBOARD cannot find the Above Board, try removing the 286 express board and replace the 8088 temporarily. SETBOARD should then run correctly. The 286 Express can be forced to let the machine boot off of the 8088 which then allows SETBOARD to find the EEPROM. Change CONFIG.SYS to: DEVICE=EXPRESS.SYS 88 This causes the machine to boot in 8088 mode. Run SETBOARD, then drop the "88" parameter and the machine will boot 80286 again. QUADRAM QUADSPRINT Tested here and found to be compatible with current Above Boards and Matched Memory Classic. SOTA 286 TURBO BOARD *CUSTOMER REPORT* of compatibility with current Above Boards and Matched Memory Classic. STB PC ACCELERATOR *CUSTOMER REPORT* of compatibility with current Above Boards and Matched Memory Classic. We do have one report that the STB PC Accelerator software is INCOMPATIBLE with the optional QUIKBUF expanded memory print buffer. EMM.SYS installs without any problems. TITAN TECHNOLOGIES ACCELERATOR PC Tested at Intel and found to be compatible with current Above Boards and Matched Memory Classic. Tech Note: If the 8087 socket on the turbo board is used, the motherboard coprocessor switch should be in the ON position. UNIVATION'S TURBOCHARGER PC *CUSTOMER REPORTS* of compatibility with current Above Boards and Matched Memory Classic. Tech Note: It appears to be the same board as the MICROWAY NUMBERSMASHER which we have tested in the compatibility lab and found to work fine with Above Boards. VICTOR SPEED PACK 286 *CUSTOMER REPORTS* of compatibility with current Above Boards and Matched Memory Classic. Must disable their cache in order to run Setboard. Tech Notes: If the customer has a math co-processor installed on the Victor board, the math co-processor switch on the motherboard must be set for no math co-processor installed. Indicating that a math co- processor is there may cause EMM.SYS to not load and display "The math co-processor switch on your computer's system board is not set correctly". The math co-processor on the Victor board will still operate correctly (passes CHKCOP) with the motherboard MCP switch set to off. This board is apparently the same board as the PC TECHNOLOGIES 286 EXPRESS, confirmed compatible here. INTEL RAPIDCAD PERFORMANCE BRIEF Intel RapidCAD Engineering CoProcessor Performance Brief Table of Contents Introduction 1 The Intel RapidCAD Engineering CoProcessor 1 Intel RapidCAD Engineering CoProcessor Performance Summary 2 Test Configurations 2 Benchmark Tests 2 DOS Standard Benchmark Tests 3 UNIX Standard Benchmark Test 3 DOS Application Benchmark Tests 4 Table 1- DOS Application Benchmark Results 7 Table 2- DOS Standard Benchmark Results 10 Table 3- UNIX Benchmark Results 11 Introduction Benchmarks are intended to give a standard measure of performance that can be used to predict how well application code will execute. These benchmark programs should be representative of the intended applications. However, the performance measured is often the combined characteristic of a given computer architecture and many other tightly- coupled system software and hardware constituents. The memory and I/O subsystem design, as well as the operating system and the software development tools, may dominate the results and make the comparison difficult. This document contains performance measurements in both DOS and UNIX operating environment, which can be used as predictors of real application performance. The Intel RapidCAD Engineering CoProcessor The Intel RapidCAD Engineering CoProcessor, the newest member of Intel386 product family, is the highest performance floating-point upgrade for Intel386 DX microprocessor-based systems. Manufactured using high speed CHMOS V technology, the Intel RapidCAD Engineering CoProcessor is a two chip set: RapidCAD-1 and RapidCAD-2. The first chip, RapidCAD-1, replaces the Intel386 DX microprocessor. It is pin compatible with the Intel386DX microprocessor and integrates the central processing unit (CPU) and floating point unit (FPU) on the same silicon die reducing the inter-chip communication delays. Eliminating the communication overhead of transferring commands, data and results over the I/O bus between the CPU and the math coprocessor (MCP), enables exceptional floating-point performance. The second chip, RapidCAD-2, is installed in the Intel387 DX Math CoProcessor socket. It provides hardware compatibility with the unmasked floating-point exception reporting in standard Intel386 microprocessor-based architectures. The floating-point and binary coded decimal data formats fully conform to the ANSI/IEEE Standard 754-1985 for binary floating-point arithmetic. The Intel RapidCAD Engineering CoProcessor is binary compatible with the Intel386DX microprocessor and the Intel387 DX, Intel387SX, Intel287XL and 8087 Math CoProcessors. Intel RapidCAD Engineering CoProcessor Performance Summary Benchmark results confirm that the RapidCAD Engineering CoProcessor runs floating-point code from 56 to 146 percent faster than the Intel386 DX microprocessor with the Intel 387DX math CoProcessor. This exceptional floating-point performance translates into excellent performance improvement for applications which makes extensive use of the floating-point instruction set. Application benchmarks show performance improvement averaging 30 to 40 percent, and as high as 67 percent for 3D Studio and MathCAD. The Intel RapidCAD Engineering CoProcessor's exceptional floating-point performance translates into real time savings for the engineering professional using an Intel386 DX microprocessor-based system running CAD or scientific application software. Test Configurations The DOS tests were performed on a COMPAQ DeskPro 386/33MHz. Memory 640KB base and 7MB extended Video COMPAQ mother board VGA Disk 80MB IDE Operating Compaq Personal Computer DOS 3.31 System Windows 3.0 The UNIX tests were performed on a COMPAQ SystemPro 386/33MHz Memory 640KB base and 7MB extended Video COMPAQ mother board VGA Disk COMPAQ Disk Array Operating System AT&T UNIX System V/386 Release 4.0 Version 2.0 Benchmark Tests: Standard benchmark tests were used to separately evaluate integer and floating-point performance. DOS and UNIX standard benchmark tests were run. All applications tested run under DOS or Windows. Application performance was estimated using two kinds of tests. A subset of the BYTE Application Benchmark Version 2.0 and the AutoCAD Benchmark Test Series Distributed by the AutoCAD Users Group of San Diego Version 1.1, were used as part of the application benchmark testing. They run automatically, using the system clock to measure the execution time. For applications where an automatically running test was not available, such as AutoShade and 3D Studio, typical commands which use RapidCAD's floating-point capability were run and the execution time was measured with a stop watch. For each test there are two sets of results, one obtained with the Intel386 DX microprocessor and Intel387 DX Math CoProcessor, the other with Intel RapidCAD Engineering CoProcessor. A performance index is calculated in each case showing the relative execution speed delta using the Intel RapidCAD Engineering CoProcessor vs. the Intel 386 DX microprocessor and the Intel387 DX Math CoProcessor. DOS Standard Benchmark Tests Dhrystone is an industry-standard benchmark test designed to measure system programming performance. It includes weighted percentages of procedure calls, loops, integer assignments, integer arithmetic and logical operations. The result is CPU speed expressed in Dhrystones/sec. Sixteen-bit Dhrystone Version 2.0 and a 32-bit Dhrystones Version 2.1 were used. Whetstone is an industry-standard benchmark test designed to predict performance in a floating-point intensive enviroment. It is a synthetic mix of floating-point and integer arithmetic, transcendental functions, floating-point array computations, and floating-point subroutine calls, based on statistical analysis of scientific FORTRAN programs. The result is expressed in KIPS (kilo instructions per second). Single and double precision 16- and 32-bit Whetstones were used. In addition the Microway Whetstone benchmark was run to give a more comprehensive measure of floating-point performance in 32-bit protected mode. UNIX Standard Benchmark Test The SPEC benchmark Suite Release 1.0 consists of 10 FORTRAN and C benchmarks that are intended to be meaningful samples of applications which perform fixed- and floating-point logical and arithmetic operations as well as disk I/O in a technical environment. Many of these benchmarks have been derived from publicly available application programs. The benchmark suite may be divided in two separate benchmark suites to distinguish between the integer and floating-point performance. This allows for better performance prediction under different operating environments. The integer performance represents a more appropriate instruction mix for commercial applications in a business environment. The floating-point performance can be used to predict the system performance in a technical environment for scientific and engineering applications. The global SPEC index, SPECmark, is the geometric mean of all test results. The SPEC integer index, SPECint, represents the geometric mean of the results for the four C programs. The SPEC floating-point index, SPECfp, represents the geometric mean of the results of the six FORTRAN programs. DOS Application Benchmark Tests Generic 3D Drafting Version 1.1 The model BEARING.3DD was used to execute a perspective change (VIEW, Perspect VP) with the coordinates 0,0,-25 and 350,400,400. The elapsed time was measured with a stop watch. AutoCAD Release 11 The BYTE Application Benchmark Version 2.0 test and the Benchmark Test Series Distributed by The AutoCAD Users Group of San Diego were used. This series of tests execute a typical mix of commands that might be issued by an AutoCAD user. These tests measure the elapsed time using the system clock. AutoCAD Release 11 Advanced Modelling Extension (AME) A simple model was created (FLANGE.DWG) to test solids subtract, mesh and filled shade. The elapsed time was measured with a stop watch. AutoShade with RenderMan Release 2.0 The sample film KITCHEN.FLM was used to test full shade and RenderMan render. The elapsed time was measured with a stop watch. 3D Studio Release 1.0 The sample models CITY.3DS, RACECAR.3DS, and STILLIFE.3DS were used to test the render function, with the following setting: Shading limit = Phone; Anti-Alias= High; Shadows = On; Mapping = On; Hidden Geometry = Hide; Render Output = Display. The elapsed time was measured with a stop watch. Cadkey 386 Version 4 To measure performance with the standard drawing functions an array of 1000 ellipses was first drawn and then deleted. To test the performance of advanced solid functions the sample model SOLID4.PRT was used. A complex process performning solid boolean operations (solid subtraction and plane sectioning), mass properties and a smooth shading followed by the rendering (with Shading = Phong) of the resulting image were performed. The elapsed time was measured with a stop watch. MicroStation PC Version 4.0 The sample ORBITER.DGN model was used for hidden lines removal, smooth shading, phong shading, stereoscopic rendering and a zoom out. The elapsed time was measured with a stop watch. Upfront Version 1.0 The sample drawing LIBRARY.UPF was used and two view change tests were done: from initial view to Birdseye and back. The elapsed time was measured with a stop watch. Mathematica 2.0 for DOS 386/7 and Mathematica 2.0 for Windows The execution time of Plot3D[10 sin[x+Sin[y]], {x, -10, 10}, {y, -10, 10}, PlotPoints -> 80] was measured with a stop watch. MathCAD 2.50 The BYTE Application Benchmark Version 2.0 test was used. It calculates a convolution integral and evaluates an iterative function system. This test measures the elapsed time using the system clock. PC-Matlab Ver 3.5g The BYTE Application Benchmark Ver 2.0 test was used. This test performs a mix of matrix and signal processing operations. This test measures the elapsed time using the system clock. SPSS/PC + V4.0.1 A statistics example with 1473 cases was used for descriptive statistics (means) and a graphic representation with Harvard Graphics. The elapsed time was measured with a stop watch. STATGRAPHICS Ver 4.0 Three randomly gneerated 1000 samples series were used for summary statistics (STATS) and multiple regression. The elapsed time was measured with a stop watch. Lotus 1-2-3 Release 3 The BYTE Application Benchmark Ver 2.0 test was used. The test loads and recalculates a spreadsheet based on the Savage formula, then it runs a macro that performs a binary goal seek. Additionally, a large block of text data is loaded, copied and then saved. This test measures the elapsed time using the system clock. Excel Version 3.0 The BYTE Application Benchmark Ver 2.0 test was used. The test is similar to the one for Lotus 1-2-3. It loads and recalculates a spreadsheet based on the Savage formula, then it runs a macro that performs a binary goal seek. This test measures the elapsed time using the system clock. Table 1- DOS Application Benchmark Results DOS Application(1) Intel386DX Intel Percentage CPU and RapidCAD Performance Intel387DX Engineer- Improvement MCP ing Co- Processor Generic 3D Drafting Ver 1.1 33.46 25.53 31% AutoCAD Release 11 Byte Magazine Benchmark Test Redraw (sec) 6.03 5.60 8% Pan (sec) 38.50 30.03 28% Zoom (sec) 46.91 34.93 34% Hide (sec) 70.57 48.44 46% Regen (sec) 27.95 20.76 35% San Diego Benchmark Test Total time (sec) 339.00 295.44 15% Phase 1, draw (sec) 45.04 41.03 10% Phase 1, ZOOM (sec) 2.69 2.14 26% Phase 1, REGEN (sec) 5.00 3.84 30% Phase 2, draw (sec) 56.14 50.97 10% Phase 2, ZOOM (sec) 7.85 6.10 29% Phase 2, REGEN (sec) 9.83 7.58 30% Phase 3, draw (sec) 66.57 60.53 10% Phase 3, ZOOM (sec) 13.07 10.27 27% Phase 3, REGEN (sec) 20.93 16.25 29% 3D Module (sec) 12.75 11.42 12% 3D VPOINT (sec) 0.93 0.77 21% 3D HIDE (sec) 11.92 9.89 21% AutoLISP calculation (sec) 2.15 1.92 12% AutoCAD Release 11 AME Subtract (sec) 36.00 31.00 16% Mesh (sec) 30.86 21.34 45% Shade (sec) 9.52 6.12 56% AutoShade Release 2.0 Full shade (sec) 16.42 12.26 34% Render (sec) 178.49 109.18 63% 3D Studio Release 1.0 (Render) CITY.3DD (sec) 369.00 223.00 65% RACECAR.3DD (sec) 1244.00 746.00 67% STILLIFE.3DD (sec) 473.00 292.00 62% Cadkey 386 Version 4 Draw ellipses (sec) 19.16 13.79 39% Delete ALL (sec) 14.84 11.33 31% Complex solids process (sec) 88.00 58.09 51% Dashed smooth shading (sec) 85.00 59.86 42% Smooth shading display (sec) 64.00 39.87 61% MicroStation PC Render-Hidden lines(sec)130.00 98.00 39% -Smooth (sec) 106.00 77.00 38% -Phone (sec) 160.00 108.00 48% -Stereo (sec) 211.00 154.00 37% Zoom out (sec) 16.00 11.00 45% Mathematica 2.0 for DOS 386/7 Plot3D (sec) 119.10 103.38 15% Mathematica 2.0 for Windows Plot3D (sec) 114.53 89.17 28% MathCAD 2.50 BYTE Magazine Benchmark Test Convolve 41.66 36.67 14% IFS 23.77 19.65 21% PC-MATLAB Ver 3.5g BYTE Magazine Benchmark Test Matrix (sec) 8.34 5.49 52% Signal processing (sec) 41.06 24.54 67% SPSS/PC+ V4.0.1 Means (sec) 14.54 12.15 20% GRAPH (sec) 10.19 9.23 10% STATGRAPHICS Ver 4.0 Summary statistics(sec) 7.61 6.00 27% Multiple regression(sec)11.41 8.19 39% Lotus 1-2-3 R3.0 BYTE Magazine Benchmark Test Load Savage (sec) 8.00 7.00 14% Calc Savage (sec) 19.00 15.00 27% Run Goalseek (sec) 13.00 11.00 18% Load Block (sec) 8.00 7.00 14% Copy Block (sec) 28.00 25.00 12% Save Block (sec) 12.00 11.00 9% Excel Version 3.0 BYTE Magazine Benchmark Test Open Savage (sec) 18.00 17.00 6% Calc Savage (sec) 73.00 55.00 33% Run Goalseek (sec) 19.00 18.00 6% (1) All applications were run under Compaq Personal Computer DOS 3.31 Table 2- DOS Standard Benchmark Results DOS Benchmark Intel386DX Intel Percentage CPU and RapidCAD Performance Intel387DX Engineer- Improvement MCP ing Co- Processor 32-bit Performance Dhrystone (Dhrystone/sec) Version 2.1 15888.10 18274.90 15% Whetstone (KWhet/sec) Single Precision 3813.00 6120.00 61% Double Precision 3286.00 5299.00 61% Microway (KWhet/sec) Whetstone 3720.90 6481.00 74% WhetMat 733.94 1625.90 122% WhetScale 1422.20 3492.50 146% WhetTrans 1051.80 1733.20 65% 16-bit Performance Dhrystone (Dhrystone/sec) Version 2.0 12955.50 13704.50 6% Whetstone (KWhet/sec) Single Precision 2272.00 3571.00 57% Double Precision 2000.00 3125.00 56% Note: The loosely copuled internal architecture makes RapidCAD more sensitive to wait states than Intel386 DX CPU. Therefore in systems without cache and more than three wait states for memory accesses, the integer performance of RapidCAD becomes lower than the integer performance of Intel386 DX Microprocessor. However this is not likely to occur, since almost all Intel386 DX microprocessor-based PCs have cache in their configuration. Additional wait states in the MCP I/O cycles may considerably increase RapidCAD's floating-point performance relative to Intel386 DX microprocessor and Intel387 DX Math CoProcessor. Table 3- UNIX Benchmark Results UNIX Benchmark Intel386DX Intel Percentage CPU and RapidCAD Performance Intel387DX Engineer- Improvement MCP ing Co- Processor SPEC 1.0 001.gcc (sec) 239.00 210.00 14% 008.espresso (sec) 336.00 270.00 24% 013.spice2g6 (sec) 5227.00 3679.00 42% 015.doduc (sec) 684.00 328.00 109% 020.nasa7 (sec) 5847.00 3561.00 64% 022.li (sec) 803.00 671.00 20% 023.eqntott (sec) 245.00 207.00 18% 030.matrix300 (sec) 1120.00 672.00 67% 042.fpppp (sec) 1000.00 395.00 253% 047.tomcatv (sec) 1138.00 552.00 106% SPECmark 2.829 4.410 56% SPECint 2.827 3.364 19% SPECfp 2.830 5.283 87% ÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍ End of file Intel FaxBack # 1422 August 26,1992