Short: Benchmark with online ranking, QR upload Author: Thorsten Kadell Uploader: Thorsten Kadell Type: util/moni Version: 1.9 Replaces: util/moni/StoneBench.lha Requires: 68000+, Kickstart 2.04+, TCP/IP stack or a phone for upload Architecture: m68k-amigaos >= 2.0.4 Distribution: Aminet Kurz: Benchmark, Online-Rangliste, QR-Upload URL: https://sthd.de/stonebench/ StoneBench is a benchmark for all classic Amigas: from the A500 with 68000 to 68060 accelerators, PiStorm, Vampire, FPGA cores and UAE. It measures the machine and compares it with a fixed reference, an Amiga 1200 with 68060 at 50 MHz and AGA, which has 100 points. 200 means twice as fast. The result can be uploaded to the website, where it is ranked next to comparable machines (real hardware, accelerators, PiStorm, Vampire, FPGA and emulators each have their own list): https://sthd.de/stonebench/ No network on your Amiga? StoneBench shows the result as QR code on the screen. Scan it with a phone and the result goes to the website from there. Features -------- - One executable for 68000 to 68060 and 68080. CPU and FPU specific code is chosen at runtime. - Test groups: CPU (integer, mul/div, branches, ...), FPU, memory (chip, fast, slow ram, copy and read/write speed), chipset (blitter, c2p into an 8 bitplane AGA screen), graphics (OS drawing, RTG if present), disk (read, optional write) and OS (memory allocation, messages, task switches). - Three scores: Workbench (CPU, FPU, fast memory, graphics, OS), Oldschool (blitter, CPU, chip ram) and Newschool (CPU, FPU, fast memory, chunky output by c2p on AGA, an 8 bit graphics card screen (through the OS or straight into its memory) or the SAGA screen of a Vampire, in pictures per second). Disk is shown separately. - Detects CPU, FPU, MMU, clock, chipset, memory, Kickstart and Workbench version, RTG system and emulators, on real hardware and PiStorm the Amiga model (A500/A2000, A600, A1200, A3000, A4000, CD32, CDTV; for an A500/A2000 it asks before the upload whether it is an A500, A1000 or A2000) and on a Vampire the board (V500, V600, V1200, Firebird, Icedrake, Manticore, V4SA, A6000). - What can not be detected can be chosen in the window: the model, the emulator (WinUAE, FS-UAE, Amiberry, ...) or the FPGA system (MiSTer, MiST, Tang Nano, ...). Before an upload StoneBench asks for it once if nothing was detected or chosen; Skip uploads without it. - GadTools window or Shell mode with ReadArgs options. - Upload over any bsdsocket.library stack (Roadshow, AmiTCP, Miami, Genesis), plain http. Without network the result is shown as QR code and uploaded with a phone. - Installs nothing, writes nothing to your disks (unless you switch on the disk write test). The results go to RAM:StoneBench.txt. How it measures --------------- Every test is first calibrated, so that one run takes a fixed time on the tested machine. Then it is measured 5 times (3 times in the quick test) with timer.device (E-Clock) and the median is taken. Each test is divided by the value of the reference A1200/060, so every test has 100 points there. The group scores and the Workbench score are geometric means. Newschool and Oldschool are time models: a picture costs calculation time plus output time (chunky output or blitter), so the slower part limits the score. The scores are calculated on the server from the raw values, each result page shows every single value. By default multitasking is off (Forbid) during the cpu, fpu, memory and chipset tests, so the machine looks frozen for about a second per test. The reference values were measured the same way. NOFORBID or the switch in the window keeps multitasking on. The chip ram, slow ram and blitter tests run on an own grey screen with 1 bitplane, so the result does not depend on the screen mode of your Workbench: a Workbench screen with many colours or a high resolution takes chip ram time away and made them slower. The c2p test runs on AGA only and shows its picture on an own 320x200 screen with 8 bitplanes, like a game would. Upload without network ---------------------- Amigas without TCP/IP stack or network card can still take part. After the run press "QR code" in the window (or start StoneBench.exe with the QR option). The result is shown as QR code on an own screen. Scan it with the camera of a phone: the phone opens the website with the result, "Upload" there stores it. The code holds the complete result, signed like a normal upload. A click or a key closes the screen. If Upload in the window fails because there is no network, StoneBench offers the QR code by itself. If the phone reports a bar code instead (EAN, product code), use the normal camera app or switch off the 1D codes in the scanner app. Boot disk --------- StoneBench.adf is a bootable floppy disk image with StoneBench.exe and this text, for Amigas without hard disk (Kickstart 2.04 or newer). Write it to a floppy, boot from it and the window opens. Floppies are not used for the disk tests, the QR code works for the upload. Note: the floppy boots a bare system without SetPatch and without the CPU libraries (68040.library, 68060.library) of your Workbench. On accelerators the caches and the FPU may then not be set up like in daily use, so some results can be much worse than with your normal system. For results to compare, run StoneBench.exe from your normal Workbench; use the floppy for machines without hard disk. Without SetPatch an AGA machine starts in ECS mode; StoneBench switches AGA on itself (until the next reset), so c2p runs from the floppy too. Usage ----- Copy the StoneBench drawer (or just StoneBench.exe) anywhere and start it from the Shell or the Workbench, StoneBench.exe has its own icon. A window shows the detected system, Start runs the tests, Upload sends the result, QR code shows it for a phone. A full run takes 1 to 2 minutes on the reference A1200, longer on slow machines. The quick test needs a few seconds there. Shell template: StoneBench.exe ALL/S,QUICK/S,TESTS/K,DISK/K,DISKWRITE/S,NOFORBID/S, PLATFORM/K,MODEL/K,NAME/K,COMMENT/K,UPLOAD/S,QR/S, SERVER/K,WINDOW/S ALL all tests, without window QUICK quick test TESTS=cpu,mem,... only these groups (cpu,fpu,mem,chipset,gfx,disk,os) DISK=DH1: partition for the disk tests DISKWRITE also test writing (needs 4 MB free, cleaned up) NOFORBID multitasking stays on during all tests PLATFORM=... real, uae, pistorm, vampire or fpga MODEL=... real hardware and PiStorm: A500, A1000, A2000, A600, A1200, A3000, A3000T, A4000, A4000T, CD32, CDTV uae: WinUAE, FS-UAE, Amiberry, Amithlon, Other vampire: V500, V600, V1200, Firebird, Icedrake, Manticore, V4SA, A6000, Salamander, Other fpga: MiSTer, MiST, "Tang Nano", Chameleon, Replay, UnAmiga, AmiCube, Other NAME=, COMMENT= shown on the website UPLOAD upload after the run QR show the result as QR code, upload it with a phone WINDOW open the window with these options preset Examples: StoneBench.exe ALL NAME="Thorsten Kadell" UPLOAD StoneBench.exe QUICK TESTS=cpu,mem StoneBench.exe ALL QR Return codes: 0 ok, 5 stopped with Ctrl-C, 10 error or upload failed, 20 could not start. Privacy ------- An uploaded result contains the hardware, the measured values and the name and comment you entered, nothing else. The website does not store your IP address, only a salted hash to limit uploads. Developed with AI ----------------- StoneBench was written with the help of an AI (Claude by Anthropic), under my direction and with all design decisions made by me. It was tested on a real A1200 with 68060/50, on a PiStorm and under WinUAE. Please report anything that looks wrong. Archive contents ---------------- file_id.diz short description StoneBench.info drawer icon StoneBench/StoneBench.exe the program StoneBench/StoneBench.exe.info program icon StoneBench/StoneBench.readme this text StoneBench/StoneBench.adf boot disk image with the program Website, ranking and the latest version: https://sthd.de/stonebench/ Credits ------- StoneBench is written in C++ and 68k assembler, built with the m68k-amigaos GCC and vasm. The c2p routines are by Mikael Kalms, public domain. History ------- 1.9 - Newschool and Oldschool are calculated in a new way, as a time model: a picture costs calculation time plus output time and the slower part limits the score. Newschool: chunky output (40 %) and CPU, FPU, fast memory. Oldschool: blitter (50 %) and CPU, chip memory. A very fast CPU no longer makes up for a slow c2p or graphics card. All results on the website were calculated again. - The native screen is no part of Oldschool any more, it still counts as best screen (Workbench). - The native screen tests are skipped if the system hands out another screen for the native mode (mode promotion). - An FPU of a 68020/68030 that only exists as software emulation is shown as "emulated". - After an upload from an older version the website answers that a newer one is there. 1.82 - Graphics cards are scored with their fastest way too: the direct copy into the video memory (gfx.rtg8.direct, measured since 1.71) counts as chunky output (Newschool), like the fastest c2p on AGA and the SAGA copy on a Vampire. Results since 1.71 were calculated again on the website. 1.81 - AGA machines started without SetPatch (e.g. from the StoneBench floppy) ran in ECS mode, so c2p and the 8 bitplane test were skipped and there was no Newschool score. StoneBench now switches AGA on like SetPatch does. 1.8 - Vampire: the SAGA chunky write test copies a 320x200 picture like c2p and the graphics card test and counts as chunky output (Newschool). Vampire results of older versions keep their score. - On real hardware the cut down CPUs are named: 68EC020 in an A1200, 68LC040, 68EC/LC060 (from the PCR, now also sent with the result). An FPU of a 68040/68060 that only exists as software emulation is shown as "emulated". - A 68060 with its FPU switched off (PCR) is no longer shown with a 68060 FPU. - The QR code has a new format (6), the website still reads the codes of 1.7 and 1.71. 1.71 - New test gfx.rtg8.direct: the picture of gfx.rtg8.write copied by the CPU straight into the memory of the graphics card, as games and demos do it. It is only shown and counts in no score. 1.7 The website only takes results of 1.7 and newer, please update. - The blitter and slow ram tests run on the same own screen with 1 bitplane as the chip ram tests (the blitter ran with the display off before). New reference values for the blitter. - c2p runs on AGA only, into an own shown 320x200 screen with 8 bitplanes instead of hidden chip ram while the Workbench was shown. New reference value. - Chunky output: c2p and the 8 bit graphics card screen are both counted in pictures per second against the same reference. A graphics card got about 9 times the points for the same speed. - Best screen: graphics card screens are scored against the native screen, so the fastest screen counts. Before, a slower graphics card could win. - chip.dma_load1 is gone, it measured the same as mem.chip.read. 1.63 - The chip ram tests run on an own 320x200 screen with 1 bitplane instead of the Workbench screen, so the result no longer depends on its screen mode. New reference values for them. 1.62 - Vampire: the SAGA chunky write test now opens its own 320x240x8 P96 screen instead of switching SAGA directly. On a V500 the Workbench stayed garbled after the test. - The 8 bitplane chip DMA test only runs on real AGA chips. 1.61 - The A500, A1000 and A2000 have the same chips and are detected as "A500/A2000". Before the upload StoneBench now asks which one it is, the answer is shown as the model in the ranking. MODEL= takes A500, A1000 and A2000 too. - A3000T and A4000T can be chosen as the model of a detected A3000 or A4000, they are shown as the model in the ranking then. 1.6 - A PowerPC of a phase5 combo card (BlizzardPPC, CyberStorm PPC) is shown in its own PPC row below the FPU: cpu, card and the version of powerpc.library (WarpOS) or ppc.library (PowerUP). It is only shown and counts in no score. - The CPU row shows the measured clock in brackets, e.g. "50 MHz (49.8 measured)", the Measured row is gone. - RAM:StoneBench.json is not written any more, only RAM:StoneBench.txt. - The score of the task switch test is in line again. 1.55 - Fix: a machine with chip memory only (e.g. an A500 without expansion) has its chip memory scored against the fast memory of the reference for the Workbench and Newschool scores. Before, the chip memory points counted, which made it look far too fast. - The idle easter egg runs smoother: faster on a 68000, a new frame every vertical blank where the machine is fast enough. 1.54 - Calmer graphics tests: grey noise instead of colours, the fill alternates between two close greys instead of black and white, the text test shows "STONEHEAD * " black on grey as a slanted pattern. Same work, same results as 1.53. 1.53 - The clock is rounded to the usual one also when it is measured up to 6 % too low (a 68000 at 7 MHz measured 6.8), 75 MHz is new. 7, 14 and 28 MHz by PAL or NTSC of the machine: 7.09, 14.19, 28.38 or 7.16, 14.32, 28.64. 1.52 - The system in the window and in RAM:StoneBench.txt has the same rows in the same order as the result page of the website: measured clock, memory summary with the regions below, boards. - Shorter status line after the QR code. - The program is 5 KB smaller. - PiStorm with emulated AGA: an A600 is no longer shown as A1200 before the run, the model stays open. 1.51 - Runs on an Amiga with only 512 KB chip ram (A500 from the boot disk): it needs about 260 KB less memory at the start. 1.5 - New window: the three scores as columns with big numbers and bars, the StoneBench logo next to them. - Progress as a stone wall: one stone per test. - Icons for the program, the drawer and the boot disk. - Tighter option rows on 640x256 screens, so the list shows more lines. - Reminder before Upload or QR code when no name is entered. - Something happens when you leave it alone for a minute. - 92 MHz (Vampire V4 clock setting) is recognised as usual clock. 1.41 - The model comes from the chips: an A1200 or A4000 with AGA switched off in the early startup menu is no longer shown as A600 or A3000. - CD32 detected by its Akiko chip. 1.4 The website only takes results of 1.2 and newer, please update. - Groups are now the parts of the three scores (chip and fast memory, blitter, native screen, best screen, chunky output). - Blitter tests run with the display off, so the screen mode of the Workbench does not change the result. - Detects the Amiga model (A500/A2000, A600, A1200, A3000, A4000, CD32, CDTV) and the Vampire board, MODEL= to choose it. - On UAE the emulator can be chosen (WinUAE, FS-UAE, Amiberry, ...), Amithlon and the UAE version are detected. On FPGA systems the system can be chosen (MiSTer, MiST, Tang Nano, ...). - Asks for the model before an upload if nothing was detected. - A blitter far faster than a real one (immediate blitter of UAE, Amithlon) counts in no score. - Graphics write test and slow ram test now also on Kickstart 2.x. - New QR code format, the website reads the codes of 1.2 and newer. 1.0 First Aminet release. Contact ------- Bug reports, questions and results that look wrong: stonebench@gmx.de Please add RAM:StoneBench.txt or the number of the uploaded result.