Path: menudo.uh.edu!menudo.uh.edu!usenet From: sjk@astro.as.utexas.edu (Scot Kleinman) Newsgroups: comp.sys.amiga.reviews Subject: REVIEW: Amiga 3000T-040 Followup-To: comp.sys.amiga.hardware Date: 21 Jan 1993 19:01:14 GMT Organization: The Amiga Online Review Column - ed. Daniel Barrett Lines: 444 Sender: amiga-reviews@math.uh.edu (comp.sys.amiga.reviews moderator) Distribution: world Message-ID: <1jmrtqINNbtl@menudo.uh.edu> Reply-To: sjk@astro.as.utexas.edu (Scot Kleinman) NNTP-Posting-Host: karazm.math.uh.edu Keywords: hardware, system, 3000T, 68040, tower, commercial PRODUCT NAME Amiga A3000T-040 [MODERATOR'S NOTE: This review was updated slightly by the author on May 3, 1993, to add a new timing result. - Dan] BRIEF DESCRIPTION Amiga A3000 in a tower case with pre-installed A3640 (68040) card. AUTHOR/COMPANY INFORMATION Name: Commodore Business Machines Address: 1200 Wilson Drive West Chester, PA 19380 USA (Varies for other countries) Telephone: (215) 431-9100 LIST PRICE The list price has recently been reduced to about $2700 (US). I got mine at a special price as a Commodore developer. SPECIAL HARDWARE AND SOFTWARE REQUIREMENTS The system came with a pre-installed A3640 card, 4MB of fast RAM and 1MB of chip RAM. It came with a preformatted (with Workbench 2.0.5) Seagate 240MB drive and one high-density (HD) floppy. I added a second internal HD floppy drive, an A2286 Bridgeboard, an A3070 tape drive and an A1960 monitor. I also transferred my Quantum 105MB drive and 14.4K Practical Peripherals modem from my A1000. In addition, I added 4MB of Fast RAM, another 1MB of Chip memory, and installed Workbench 2.1. BACKGROUND READING Since the Amiga 3000T-040 is an A3000T with a 68040 card, I recommend you read Dan Barrett's previous reviews of the A3000T. They are available in the c.s.a.reviews archives on math.uh.edu, and they have a lot of relevant material which I will not repeat here. WHY A3000T-040? My computing needs have recently outgrown my Amiga 1000, so I started exploring upgrade paths. My considerations were: an A4000, the A3000T-040, a '486 clone, and a SPARCClassic. I ruled out the '486 since that would require new software and learning a new architecture; and with the new pricing on the Amigas, the savings would not be much. The SPARCClassic was next to leave the list. While the new base machine is fairly inexpensive, the cost of hardware necessary to utilize it fully plus maintenance and repair proved too high. That narrowed it to an A4000 or an accelerated A3000. The only thing the A4000 offered that the A3000T-040 did not was the new chip set and higher graphics speeds (see benchmarks, below.) For the price of not having the latest machine, the A3000T-040 offers more expansion capability and a real SCSI port. The only other option I considered was waiting for the A4000T, but I really couldn't wait much longer, and I imagine the A4000T will be priced significantly more than the reduced A3000T-040. So, I placed my order for the A3000T-040. SETUP The A3000T-040's box contains a bag of parts labeled, "Open Immediately." Inside is a note that says to put the enclosed SCSI terminator on the back SCSI port before turning on the machine. It worries me that one might easily overlook this little warning and have trouble later. Also included was a note telling me to install the 68040.library immediately. A 68040 support disk was included with a neat install script that worked fine. After turning on the machine and making sure it all worked, I proceeded to take it apart and install my expansion devices. I used the hints in Dan's review to take off the front and side panels. (I recommend reading that review, since the manual does leave a few things out.) The first thing that I noticed is that the motherboard does not have the 68030 and 68882 that other A3000s have. The A3640 card is in the CPU slot and held down by two screws. This must be removed to install fast RAM. I made myself a groundstrap from a length of speaker wire and set to work. One must be careful, but the RAM installation is easily done. I then started to install the the second floppy drive. Here I faced two problems. First, the screws needed to hold the mounting bracket to the tower were not included in the bag of parts; and second, no instructions were included with the drive. So when I got the screws myself and installed the drive, I could not get the machine to recognize both drives at the same time. I fixed the problem by changing a jumper on the new drive from the DS0 position to the DS1 position. Note that there is a jumper on the motherboard marked DF1 and NO DF1, but it was recently posted to the net that this jumper is not relevant to the HD drives. Perhaps it is for the standard drives -- I don't know. Also missing from my bag were the screws for the Tower's feet and for the 5.25" bridgeboard drive. Commodore eventually sent me the feet screws along with the "Quick Installation" poster that I was missing. I managed to assemble the system without it. By the way, you must take both side panels off to install the floppy drives. The bridgeboard and drive went in easily, though I still can't get the floppy to work. I suspect it is a bad drive, as I did get a borrowed drive to function properly. Sigh. The biggest challenge, I thought, was going to be the installation of my old Quantum drive from my A1000. I was not sure if I would be able to take the drive off my A1000 controller and have it accessible on another. It actually turned out quite easy, though. I mounted it with the supplied "HD stacking brackets" over the internal Seagate drive and used my old mountlist entries to mount my three separate partitions. To my surprise, they mounted the first time (after changing the device name), and all my files were there. I then tested the A3070 by backing up the drives, and used HDToolBox to reformat them and the A3070 (via Quarterback 5.0.3) to restore them. It all worked fine. The only real tricky installation process is the installation of the hard drive shelf, as Dan mentions in his review. His hints, however, were quite helpful and sped up the process enormously. I made the mistake, though, of trying to figure it out myself before re-reading his hints. The installation of 2.1 over my existing 2.0.5 Workbench was simple. I just used the install script and it replaced everything it needed to, got rid of anything obsolete, and left everything else alone (especially SPEAK: and Say.) ADDRESSING DAN'S CRITICISMS (from his review) The internal fan: I don't find the fan noise intolerable at all. It is audible, but not disturbingly so. Either I have an improved fan or am not as sensitive as Dan. It is louder than my A1000's fan, but much quieter than the hunk-of-PC on my desk at work. The unit is standing on the floor facing me at the end of my desk -- about two feet away. The speaker: I have not really employed it much, so I can't comment on it yet. Locking up during crashes: I have had only a few crashes and all of them resolved with the three-finger salute. I've been quite happy at the fewer number of GURUs. I've had the machine on almost continuously since I got it. No major problems there, either. I guess the real test will be this summer when it is significantly hotter here in Texas. The machine is remarkably stable. (And Mindwalker still works!) SPEED AND BENCHMARKS Since one of my primary needs was a faster CPU, I set out to benchmark the system. My work entails a lot of floating-point intensive operations. I installed Manx C5.2a (they say version 5.3 will support inline math functions that the 68040 needs) and my old version of Absoft Fortran. Both seemed to produce valid code -- though the Fortran compiler is certainly not optimized for the new breed of Amiga CPUs. I used these compiler to run some of my standard codes and timed them compared to the SPARC stations I use at work. I would compare them to my A1000, but I have disassembled it, and the thought of setting up the compilers on a floppy-based system is no longer appealing. In the following benchmark reports, an asterisk ("*") denotes the A3000T-040. The tests were performed on a standard system. I tried to run all benchmarks with nothing else running on the system. (It's possible there were some Bridgeboard tasks running -- I am not familiar enough with it to be sure.) I have not yet tried any of the '040 hacks which speed things up by remapping Kickstart to RAM. If the doc files I read are correct, a substantial speed increase can be gained through some of these programs. I can post results later under some of these conditions if there is interest. DFT (Discrete Fourier Transform): I compiled my standard Discrete Fourier Transform program both with Aztec C and Absoft Fortran. I compare the speed to a SPARC 2 and a SPARC SLC both without any additional (non-root) processes running) using the native compilers optimized for floating point operations. The program reads in a 932 line data file and a 3 line control file, performs a DFT on the data, then writes the results back to disk. The small test produced a 600 point DFT; the big test used the same data but made a DFT with 60000 points. The Amiga C code was optimized by the compiler. The 5.3 version of Manx C should speed things up a bit when it is released. As I understand it, the 68040 FPU does not have the transcendental math functions built-in that the 68881/2 does. Instead, it generates traps to routines which emulate the functions. These traps take time. If, however, the compiler generates the emulated functions in the code itself, then the 68040 can execute these functions quicker than a similarly clocked 68881/2 can. Since the Fourier transform uses these functions extensively, I expect the inline optimization to make a noticeable difference. I think the Amiga FORTRAN timings say more about the compiler than the machine. System Small Timing (sec) Big Timing (min:sec) ------ ------------------ -------------------- * (C/Optimized -so) 54 / Not Run Not Run * (above but read/write to RAM:) 47 / 46 76:48 / 76:45 * (above with inline 040 code) Not Run / 37 Not Run / 60:27 * (Fortran RAM:) 113 224:50 SPARC 2 (C/Optimized -O2) 12 / 10 17:49 / 16:02 SPARC 2 (Fortran/Optimized -O2) 12 / 10 18:01 / 16:28 SPARC SLC (Fortran/Optimized -O2) 34 / 32 55:26 / 52:02 SPARC SLC (C/Optimized -O2) 35 / 33 55:13 / 53:04 CPUSpeed: Using the CPUSpeed program included with the operating system, I get the following numbers: Integer Math Floating Point Math ------------ ------------------- A2000 217K N/A A3000 1728K 616K * 3769K 4822K SysInfo: Using the Speed test on the main screen and the Speed test on the drives screen, I find: Dhrystones: 18093 Speed Comparisons (how much faster the tested system is) ----------------- A500 A600 STD 30.82 (e.g., 31 times faster than A500) B2000 Extra RAM 25.88 B2000 GVP A3001 3.04 A2500 A2620 8.80 A3000 25MHz 3.90 PP&S Mercury '040 0.90 MIPS 18.88 MFLOPS 4.79 Internal SCSI Drive Read Speed: 514K/sec. AIBB: (This is a NICE program -- thanks, LaMonte.) I ran all tests with multitasking disabled and the Screen Overlay option turned on. I chose the most optimized code options for each computer involved. In summary, the A3000T-040 is about 10-20 times faster than a stock A500 for non floating-point (FP) stuff, and about 200 times faster for FP stuff. It is consistently a few times faster than a stock 25MHz A3000 and equal to an A4000 for all but the graphics operations. I understand that running AmigaDOS 3.0 on the system even without the AGA chipset can speed up some graphic operations by a factor of two or so. Here is a summary of the AIBB log file: System CPU: 68040 System FPU: 68040 System MMU: 68040 CPU Clock Rate: 25.0 MHz FPU Clock Rate: 25.0 MHz Test Name: EmuTest This Machine : 16.43 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.02 (68000 | SC Math) A3000-25 : 5.53 (68020 | CP Math) A4000-25 : 16.09 (68020 | 40 Math) Test Name: InstTest This Machine : 10.40 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.01 (68000 | SC Math) A3000-25 : 5.90 (68020 | CP Math) A4000-25 : 10.08 (68020 | 40 Math) Test Name: Writepixel This Machine : 6.38 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.45 (68000 | SC Math) A3000-25 : 4.10 (68020 | CP Math) A4000-25 : 26.12 (68020 | 40 Math) Test Name: Sieve This Machine : 11.81 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.04 (68000 | SC Math) A3000-25 : 9.28 (68020 | CP Math) A4000-25 : 11.88 (68020 | 40 Math) Test Name: Dhrystone This Machine : 19.16 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.03 (68000 | SC Math) A3000-25 : 5.96 (68020 | CP Math) A4000-25 : 18.96 (68020 | 40 Math) Test Name: Sort This Machine : 19.91 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.03 (68000 | SC Math) A3000-25 : 7.13 (68020 | CP Math) A4000-25 : 19.67 (68020 | 40 Math) Test Name: Matrix This Machine : 16.14 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.02 (68000 | SC Math) A3000-25 : 10.14 (68020 | CP Math) A4000-25 : 16.22 (68020 | 40 Math) Test Name: IMath This Machine : 41.42 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.01 (68000 | SC Math) A3000-25 : 18.29 (68020 | CP Math) A4000-25 : 41.66 (68020 | 40 Math) Test Name: MemTest This Machine : 11.35 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.03 (68000 | SC Math) A3000-25 : 6.70 (68020 | CP Math) A4000-25 : 11.54 (68020 | 40 Math) Test Name: TGTest This Machine : 1.73 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.25 (68000 | SC Math) A3000-25 : 1.62 (68020 | CP Math) A4000-25 : 2.93 (68020 | 40 Math) Test Name: Savage This Machine : 229.91 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.01 (68000 | SC Math) A3000-25 : 207.15 (68020 | CP Math) A4000-25 : 247.66 (68020 | 40 Math) Test Name: FMath This Machine : 214.06 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.02 (68000 | SC Math) A3000-25 : 23.65 (68020 | CP Math) A4000-25 : 213.65 (68020 | 40 Math) Test Name: FMatrix This Machine : 19.54 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.02 (68000 | SC Math) A3000-25 : 7.14 (68020 | CP Math) A4000-25 : 19.33 (68020 | 40 Math) Test Name: BeachBall This Machine : 201.82 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 2.08 (68000 | SC Math) A3000-25 : 75.70 (68020 | CP Math) A4000-25 : 354.16 (68020 | 40 Math) Test Name: SWhetstone This Machine : 209.28 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.02 (68000 | SC Math) A3000-25 : 74.65 (68020 | CP Math) A4000-25 : 214.39 (68020 | 40 Math) Test Name: DWhetstone This Machine : 204.86 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.01 (68000 | SC Math) A3000-25 : 78.68 (68020 | CP Math) A4000-25 : 211.20 (68020 | 40 Math) Test Name: FTrace This Machine : 231.48 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.01 (68000 | SC Math) A3000-25 : 124.04 (68020 | CP Math) A4000-25 : 231.59 (68020 | 40 Math) Test Name: CplxTest This Machine : 29.39 A500-NFR : 1.00 (68000 | SC Math) Base System A2000-FR : 1.02 (68000 | SC Math) A3000-25 : 7.22 (68020 | CP Math) A4000-25 : 29.03 (68020 | 40 Math) DOCUMENTATION The documentation is as Dan describes in his review. The only 68040 specific item is the sheet of paper that comes with the 68040 support disk, and it has an error. It states that the included version of the 68040 does not include an FPU, but it does. LIKES AND DISLIKES I am very impressed with the speed of the machine and its reliability. I like AmigaDOS 2.1 a lot: it's a big jump up from a disk-based 1.3. I am equally happy with all the peripherals. The only thing I miss from my A1000 (besides a horribly crowded desktop) is a place to put a pen on the keyboard. Resting a pen against the top row of function keys is not quite the same. I was a bit disappointed that I did not receive the three-button mouse with the system. Instead, I got the standard "pregnant" two-button version. It handles quite well, though. I like the design of the tower and am happy with the A1960 monitor, though I had heard some bad reports about it. I certainly would prefer a larger 3rd party monitor; but without that kind of money, the A1960 seems to be a good bet. CONCLUSIONS I am very happy with the speed of the system. It is only a few times slower than the SPARC that I could have afforded instead (without peripherals of course.) My only worry is that the machine will be treated as some sort of hybrid and future support may be shaky. But then, I owned a Commodore CBM 2001 when everyone else was sporting Apples and TRS-80s, so I'm used to it. This machine should get as much support as any accelerated A3000 does, and perhaps more since it is Commodore-accelerated. I understand that AmigaDOS 3.0 will run on non-AGA machines; so once RTG graphics come along, this machine should be almost indistinguishable from an A4000. The machine really feels professional. For someone looking for a good deal on speed and who doesn't mind being slightly behind the current scene, I highly recommend this machine. I remain pleased with the machine and my productivity has gone up enormously. Processes that used to make me pour a glass of water, read the newspaper and walk the dog before they finished are now done before I can record in my logbook what I am doing. Now, if I only I could make it complete my logbook as well.... I would be happy to respond to any questions or comments. My E-mail address is printed below. COPYRIGHT NOTICE Copyright 1993 Scot Kleinman. All rights reserved. Permission is granted to copy and distribute this review free of charge, provided it is distributed unmodified and in its entirety. It may not appear in any commercial publication, in whole or in part, without the author's written permission. Scot Kleinman -- Astronomy Department, University of Texas at Austin, Austin TX 78712. -- sjk@astro.as.utexas.edu sjk@emx.cc.utexas.edu --- Daniel Barrett, Moderator, comp.sys.amiga.reviews Send reviews to: amiga-reviews-submissions@math.uh.edu Request information: amiga-reviews-requests@math.uh.edu Moderator mail: amiga-reviews@math.uh.edu