Path: menudo.uh.edu!usenet From: cld@wucs1.wustl.edu (Christopher L. Davis) Newsgroups: comp.sys.amiga.reviews Subject: REVIEW: Fusion Forty accelerator for Amiga 2000 Followup-To: comp.sys.amiga.hardware Date: 27 Apr 1993 02:18:03 GMT Organization: The Amiga Online Review Column - ed. Daniel Barrett Lines: 332 Sender: amiga-reviews@math.uh.edu (comp.sys.amiga.reviews moderator) Distribution: world Message-ID: <1ri54r$iin@menudo.uh.edu> Reply-To: cld@wucs1.wustl.edu (Christopher L. Davis) NNTP-Posting-Host: karazm.math.uh.edu Keywords: hardware, A2000, accelerator, 68040, commercial PRODUCT NAME Fusion Forty accelerator BRIEF DESCRIPTION The Fusion Forty (abbreviated as "F40" in this review) is a 68040 accelerator card for the Amiga 2000. It connects via the processor slot and may be populated with 4, 8, 16, 20, or 32 MB of 80ns (or faster) 32-bit RAM. My board has "Plug and Go" ROMs version 2.1. AUTHOR/COMPANY INFORMATION Name: RCS Management Address: 120 McGill Street Montreal, Quebec H2Y 2E5 Canada Telephone: (514) 871-4924 FAX: (514) 871-4926 BBS: (514) 871-9881 LIST PRICE $1170 (US) with no 32-bit RAM, when purchased directly from RCS. Street price: unknown. I got mine during a special for Amiga User Groups and paid $995 for the board and $175 for 4 MB RAM. At the time, other 68040 boards were selling for $2000 or more. SPECIAL HARDWARE AND SOFTWARE REQUIREMENTS HARDWARE An Amiga 2000. I recommend a hard drive. SOFTWARE Works under AmigaDOS 1.3 and higher. If you have AmigaDOS 2.1, all you need is the CPU command in the C: directory and LIBS:68040.library. For lower AmigaDOS versions, additional software is provided on the F40 install disk (SetPatch, SetFF, FFCache). I did not try the board under AmigaDOS 1.3, but the board does support it. COPY PROTECTION None. MACHINE USED FOR TESTING Amiga 2000 (ECS -- see below) IVS Trumpcard Pro SCSI Disk controller Supra 2000 Memory Card with 4 MB 16 bit Fast Ram DKB MegaChip 2000 (2 Meg Chip Ram) and Super Denise AmigaDOS 2.04 Fusion Forty Accelerator with 4 MB of 32-bit Fast RAM REVIEW The F40 is a 28 MHz 68040 accelerator board for the Amiga 2000. (The manual says 25MHz, but all my system measurement software says the processor is overclocked to 28 MHz.) It has a built in math coprocessor (FPU) and Memory Management Unit (MMU). And in case you want to go back to your 68000, there is a hardware switch on the back plate to disable the 040. This change should NEVER be made while the machine is running. The F40 is cleanly designed, with no traces or pins wired together. There are a couple of surface-mounted chips, but the rest are socketed. The board is 6-layer with separate ground and power planes. There are 3 expansion connectors on the board for future use. Before I got the F40, I hadn't really had much experience with hardware installation, but the board was quick and easy to install. Software installation was handled via Commodore's Installer program... very nice and easy to work with. The first thing I noticed was that everything was so quick. Things just jumped out onto the screen. Next, I set all the caches for maximum performance. I wanted to get the most out of the hardware. Some software broke because of this; I cover this topic in more depth in the BUGS section, below. Of course, one of the biggest benefits of the processor upgrade was in multitasking. I would experience pauses with my old 68000 while doing the most mundane things. With the F40, I have no such problems. I have a number of tasks running simultaneously with no discernible pauses and no noticeable slowdown. I have downloaded files at high speeds while compiling, working with a Digi-Paint picture, or processing JPEG graphics with HamLabPlus (a great program -- shameless plug for Ed Hamway). Just how fast is the board? I did some benchmarking with a pre-release version of AIBB 6.0 and with SysInfo. I chose AIBB because I believe the suite of tests is a pretty good cross-section of the computing we all do. It is composed of integer and floating point math, and some graphics tests, including a piece of a rudimentary raytracing algorithm. During the tests, I chose the A4000/040 as my base machine. All caches were active, and advanced code generation options were activated where applicable. Also, FPUs were utilized where they existed. The following table shows the results in the form of percentages faster or slower than the A4000/040. For example, a rating of 1.26 means "26% faster than an A4000/040," and a rating of 0.63 means "37% slower than an A4000/040." ----------------------------------------------------------------------- Tests | A4000-40 | A2000-F40 | A600 | A1200 | A3000-25 | | BASE | | | | | ----------------------------------------------------------------------- EmuTest | 1.00 | 1.26 | 0.06 | 0.12 | 0.31 | ----------------------------------------------------------------------- InstTest | 1.00 | 1.59 | 0.10 | 0.17 | 0.54 | ----------------------------------------------------------------------- EllipseTest| 1.00 | 0.61 | 0.18 | 0.44 | 0.43 | ----------------------------------------------------------------------- WritePixel | 1.00 | 0.43 | 0.07 | 0.19 | 0.23 | ----------------------------------------------------------------------- LineTest | 1.00 | 0.62 | 0.53 | 0.92 | 0.58 | ----------------------------------------------------------------------- Matrix | 1.00 | 1.47 | 0.06 | 0.23 | 0.63 | ----------------------------------------------------------------------- Sieve | 1.00 | 1.74 | 0.09 | 0.37 | 0.79 | ----------------------------------------------------------------------- IMath | 1.00 | 1.13 | 0.02 | 0.21 | 0.43 | ----------------------------------------------------------------------- Dhrystone | 1.00 | 1.13 | 0.05 | 0.11 | 0.29 | ----------------------------------------------------------------------- MemTest | 1.00 | 2.43 | 0.29 | 0.79 | 1.88 | ----------------------------------------------------------------------- Sort | 1.00 | 1.20 | 0.05 | 0.14 | 0.36 | ----------------------------------------------------------------------- TGTTest | 1.00 | 0.72 | 0.26 | 0.56 | 0.51 | ----------------------------------------------------------------------- Savage | 1.00 | 1.16 | <0.01 | 0.01 | 1.25 | ----------------------------------------------------------------------- Flops | 1.00 | 1.13 | <0.01 | <0.01 | 0.17 | ----------------------------------------------------------------------- FMath | 1.00 | 1.12 | <0.01 | <0.01 | 0.11 | ----------------------------------------------------------------------- TranTest | 1.00 | 1.53 | <0.01 | 0.02 | 0.95 | ----------------------------------------------------------------------- FMatrix | 1.00 | 1.73 | 0.05 | 0.12 | 0.37 | ----------------------------------------------------------------------- FTrace | 1.00 | 1.21 | <0.01 | 0.02 | 0.98 | ----------------------------------------------------------------------- BeachBall | 1.00 | 0.95 | <0.01 | 0.02 | 0.32 | ----------------------------------------------------------------------- CplxTest | 1.00 | 1.18 | 0.04 | 0.07 | 0.25 | ----------------------------------------------------------------------- Descriptions of the tests can be found by downloading AIBB version 6.00 when it is available. The descriptions and pseudo-code examples of the test can be found in the Documentation directory of the AIBB 6.0 archive. The F40 shows admirable benchmarks in all categories except the graphically oriented ones. I attribute this to the higher bandwidth and advanced capabilities of the AGA chipset. I can not figure out why the 3000-25 outscored both the 4000-40 and the F40 in the savage test: perhaps some advantage to the 030-882 combination? From the data collected above, the F40 equipped A2000 seems to exceed the performance of the A4000-40 in non-graphics tests by 12 to 143 percent. The fact that the F40 and the A3000 scored better on the memory test is not a surprise, considering that I have heard of some type of problem in the 4000 memory system. I don't know the exact nature of the problem, but it was discussed here on the net a while back if I remember correctly. Nic Wilson's SysInfo 3.11 agrees with the Dhrystone measurement above, with a 11% faster F40 compared to AIBB's 12%. It also rates it at approximately 21+ MIPS and 5+ MFLOPS. Take that rating (MIPS/MFLOPS) for what it is worth. ;) To test C compilation speed, I wanted results that could be utilized by the largest group of possible people interested. I decided to compile something that came on the SAS C 6.2 distribution (6.0 with 6.2 patch applied). I chose the "cback" example. I tested the compile by double clicking the Build icon for cback and stop-watching the compile. It actually creates two files, cback and schelp. I tested with my 68000, my F40 with caches and copyback off, and with caches and copyback on. Here are the results. 68000, plain: 137 seconds 68040, caches off: 26 seconds 68040, caches on: 18 seconds I enjoy video and have done some playing around with 3D Rendering, so I wanted to do a rendering benchmark. Again, like compiling, I wanted to do something that others could reproduce. I have Caligari 2 and chose to load an object that came with the package. I chose gobot.obj. Once the object is loaded, I clicked the render button. I ran it with the data cache and copyback, and then without any caches. I don't even think Caligari 2 supports just using the 68000; and in any case, I don't believe I have the patience to wait as long as I think it would take. Here are the results: without copyback and data cache: 87 seconds with copyback and data cache: 43 seconds This is by no means conclusive but provides you with some solid numbers that you can compare to your own system. For you Video Toaster users, the F40 is fully compatible, and I hear they are in regular use at NewTek (this was unconfirmed by NewTek -- have you tried reaching them for the last week or so?). According to Micro Times, the F40 was the accelerator used in the Amiga 2000's that produced the effects for Babylon 5. Their setup consisted of a dozen 2000's (8 of which did rendering) with F40s and 32 Meg of RAM. They were networked by a Novell network and used Oxxi's netware package for the Amiga. DOCUMENTATION A small 30 page manual is provided covering the F40 and its installation (hardware, software, and new RAM). It also covers enabling and disabling the 040 and how to use the cache controlling software. There are sections in the manual that deal with hardware and software considerations, how the F40 works, a Question/Answer section (short), and technical specifications. LIKES: Speed!!! What a screamer. Vendor -- See vendor section. Reliability -- Rock SOLID for over a year. DISLIKES One only really that I just discovered. I put in the new Plug and Go ROMS recently. They fixed a minor annoyance that called for a double boot at power up time. When I went to pry the ROMs out of the sockets, I found out that the sockets did not have the usual hard plastic bottoms. There was a mylar-like film running between the pin sockets (see diagram below): ____________<-Mylar ooooooooooooooo Side | |<-Pin Mylar here Top View View | | Sockets--->ooooooooooooooo Forget using the old screw driver pry up for this task. I had a chip puller handy for Fat Agnus chips, and it happened to fit just fine... big sigh of relief. Putting the new chips in was challenging, but this fumble-fingered hardware numbskull managed, so it wasn't too bad. Suggestions for the board: on-board SCSI (preferably SCSI-II) would be a welcome addition. See the section at the end of the review about other coming attractions. COMPARISON WITH OTHER PRODUCTS I have used no other 040 accelerators, but I would heartily recommend this product, without a shade of hesitancy, based on the reliability of my current card and the vendor's excellent attitude toward customer satisfaction (see vendor). BUGS I experienced a few bugs on my older ROM version. It had the annoying property of having to be double booted at power-up time (now fixed with Plug and Go). Quarterback 5.0 would hang if copyback was on. Talking to CCS/New Horizons provided a fix, and the bug became less frequent, but Plug and Go has corrected the problem totally. My SAS 6.2 upgrade version of CPR caused a similar type of problem that Quarterback did, and the Plug and Go ROMS fixed that problem, too. There may be some few bugs left (I have read articles on the net about problems with Emplant), but RCS works quite hard to maintain maximum compatibility. They just recently set up an upgrade BBS (see the phone number in the AUTHOR/COMPANY INFORMATION section, above), so we can download new software as it becomes available. VENDOR SUPPORT Superlative. This is one area you can't fault RCS on. At least I can't. I have called and talked with Sales and Technical Support ("TS") a number of times. I have talked to the same people for the last year or longer. TS is always helpful, and actively asks what types of things I am doing. Serge (TS) is knowledgable and quite forthcoming with information, so much so that it is hard to get all my questions in sometimes. Rischi (Sales) is quite helpful and is free with information and careful not to bad-mouth his competition. RCS seems very driven to satisfy the customer, and I am quite satisfied with their product and their support. WARRANTY One year, parts and labor. CONCLUSION The F40 is an excellent product. The quality of the board and the responsiveness of RCS combine into a winning solution. Two thumbs up here. COMING ATTRACTIONS I have specs on their new 33 MHz version board. The two things that have changed radically are: 1.) 2.0 AmigaDOS or Higher is required (no more 1.3). 2.) 256 Meg of RAM on the board. Another model may only support up to 128 Meg on the board. There was word of a Chip RAM accelerator, but nothing has come of it yet. It has apparently been shelved for now, but it is not dead. The new version of the board is being given top priority. They also have a networking card that currently works with the existing F40. They also promise some new and exciting stuff (that I wish I knew more about) soon, like an extremely cost-effective 24-bit graphics card that plugs into the F40. COPYRIGHT NOTICE This article is in the Public Domain. --- 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