Craig Finseth's Quick-and-Dirty HP95LX Timing Tests posted to Comp.sys.handhelds 22 May 1991 32SII added 17 November 1991 Part 1: as a calculator Problem 1: TVM functions, using built-in TVM, initial conditions: n 5 * 12 i ?/ PV 100,000 PMT -2,000 FV 0 solve for i [where applicable, it has been given a range of (0,0,0)] Problem 2: TVM functions, as above, but using this solver equation: (1-EXP(-n*LNP1(i/100)))*PMT*100/i+PV=-FV*EXP(-n*LNP1(i/100)) The 32SII uses the equation as given in the manual: P*100*(1-(1+I/100)^-N)/I+F*(1+I/100)^-N+B (Due to limits in the 32SII's solver, initial values for I are .0001) Problem 3: Solver equation X^3=3^X where applicable, X has been given a range of (50,50,50) ---------- Results time in seconds for problem model 1 2 3 -12C 12.6 -18C 2.1 7.1 41.3 -27S 1.3 4.3 27.0 -28S 7.9 17.8 [2] 11.6 26.0 32SII 9.7 21.7 -41CX[1] 17.1 48SX 3.1 15.1 [2] 3.8 25.3 95LX 2 3 16 [1] w/Advantage module [2] displaying intermediate results -------------------- Problem 4: Plot the equation from problem 3 using AUTO ranging, xmin of -2 and xmax of 4. (28S does not support AUTO, a Y range of -1 to 1 is used.) ---------- Results model time in seconds for problem 4 pixels s/pixel -28S, plotting as written 20.1 137 0.1467 plotting X^3-3^X 19.7 137 0.1438 48SX, plotting as written 26.4 131 0.2015 plotting X^3-3^X 23.6 131 0.1802 95LX 12 180est 0.0667 ------------------------------------------------------------ Part 2: as an IBM PC hardware: PC is an IBM PC/XT w/Microsoft Mach 10 card Tosh is a Toshiba T1000SE Measurement A: Central Point Software PCTools version 4.21 system Info 1.0 = "original IBM PC" Measurement B: "SYSINFO" program, origin unknown 1.0 = "original IBM PC" Measurement C1: Freya (Emacs-compatible text editor) CPU-intensive command (Help: building table of bindings; makes extensive use of pointer arithmetic, function calls, table searching, and string matching). Measurement is time in seconds. Measurement C2: Measurement C1 converted to an increasing, normalized scale. Numerically: 26.8 / C1 ---------- Results rating for measurement model A B C1 C2 average PC slow 0.9 1.4 26.8 1.0 1.1 PC fast 2.2 1.9 10.3 2.6 2.2 [1] Tosh slow 1.25 1.1 17.0 1.6 1.3 Tosh fast 2.55 2.2 8.7 3.1 2.6 [1] 95LX 2.5 18 1.5 2.0 [1] The doublings due to slow / fast change are consistent with the vendor's claims. In both cases, the difference between slow and fast modes is claimed to be a doubling of the clock speed. Conclusions: - Depending upon the problem, the 95LX is from 1 to 3 times the speed of a 48SX. - If you are doing CPU register-intensive work, the 95LX is indeed the claimed 2.5 times an original PC. If you are doing memory access-intensive work, the 95LX slows somewhat. This is possibly due to the very low power RAM chips having a slower cycle time.
Areas
General
Craig's Articles
Last modified Saturday, 2012-02-25T17:30:44-06:00.