Path: kernighan.cs.umass.edu!barrett From: dejong@stm.tudelft.nl (Maarten D. de Jong) Newsgroups: comp.sys.amiga.reviews Subject: REVIEW: Cyberstorm Mk II 040/40 ERC Followup-To: comp.sys.amiga.hardware Date: 27 Jul 1997 22:29:24 GMT Organization: The Amiga Online Review Column - ed. Daniel Barrett Lines: 508 Sender: amiga-reviews@math.uh.edu (comp.sys.amiga.reviews moderator) Distribution: world Message-ID: <5rgi44$cni@kernighan.cs.umass.edu> Reply-To: dejong@stm.tudelft.nl (Maarten D. de Jong) NNTP-Posting-Host: knots.cs.umass.edu Keywords: hardware, accelerator, 68040, A3000, A4000, commercial X-Review-Number: Volume 1997 Number 10 Originator: barrett@knots.cs.umass.edu PRODUCT NAME Cyberstorm Mk II 040/40 ERC BRIEF DESCRIPTION The Cyberstorm is a replacement for the CPU board found in any A3000, A3000T, A4000 and A4000T. It features a 40 MHz 68040 (full version), 4 SIMM slots which can be populated with 4, 8, 16 or 32 Mb SIMMs in any combination and a connector for a separately available FastSCSI-2 controller. AUTHOR/COMPANY INFORMATION Name: phase 5 Digital Products Address: In der Au 27 61440 Oberursel Germany Telephone: +49-6171-583787 Support: +49-6171-583788 FAX: +49-6171-583789 WWW: http://www.phase5.de/ FTP: ftp://ftp.phase5.de/ LIST PRICE DM 719,-- (approximately US$ 420) in April 1997. Prices of up to US$ 520 have been spotted, so shop wisely! SPECIAL HARDWARE AND SOFTWARE REQUIREMENTS HARDWARE Memory (SIMMs, 32 or 36 bits, 70 ns or faster, 4 MB or larger) to populate the board recommended but not required. SOFTWARE 68040.library, which you have to supply, is required. MACHINE USED FOR TESTING A4000/030 with 68EC030 and 68882 (both at 25 MHz) 2 MB Chip, 12 MB Fast memory Kickstart 3.0, Workbench 3.0 Harddisks: Quantum LPS270A (270 MB), Maxtor 7060 AT (60 MB) (both are IDE) Cybervision 64 (4 MB video memory, CyberGFX v40.64) GVP I/O Extender (v1.8) INSTALLATION Installing a new CPU module in a 4000 is tricky, because you'll have to remove quite a bit of hardware in order to have easy access to the CPU slot. I have no clue as to how the situation is with the other computers, but have your dealer or a qualified engineer install the board for you in case you're not comfortable about opening up the Amiga by yourself. The Mk II module differs substantially from its predecessor. The Mk I was made up of at least three PCBs: a carrier board, a CPU board and a memory board. The fourth board, containing the SCSI option, could be added separately. The carrier board featured a second-level cache of 512 KB. All of this made for a rather difficult install: all of the reviews I read mention the author rasping his knuckles or transfixing his fingers. The Mk II design is far simpler. In its basic configuration, the module consists of just one PCB, which houses both CPU and memory slots. The second level cache has vanished, since it was hardly ever used anyway. Only the separate SCSI module has remained, which will be dealt with in a separate review. Before you open up your Amiga, check if you have a 68040.library installed in your LIBS: directory. You can find one on any original 2.04 or higher Workbench disk set. (Since Kickstart 1.3- do not support the 68040, anything might happen if you try to use these versions of the OS. Be careful.) The installation process indeed looks simple too: only three pages of text and photographs are devoted to installing your new gadget in your computer. The process is only described for a 4000, so you are on your own when it comes to the 3000(T). However, as an addtional step I would recommend removing the front drive bay, since it facilitates access to the CPU-slot considerably, especially if you have big hands. There are a few things worth mentioning. According to the manual, a ventilator is included with the card; this is not the case anymore. The 68040 now only bears a cooling fin. This fin gets quite hot during operation; take care that no plastic cables touch it! Second, a leaflet has been provided stating that reliable operation of Zorro III DMA-devices in all 3000 models is not possible due to a flaw of the mainboard, and that DMA-capable additions to the Cyberstorm (in other words, the CyberSCSI Mk II module) will be affected by this problem as well. However, I have heard that you can solve this by upgrading the custom chips Buster and Gayle (or Gary?), but please check this out with your dealer or someone who knows the hardware of a 3000. In any case, I strongly recommend you read the manual thouroughly before opening up the Amiga. REVIEW The Hardware ~~~~~~~~~~~~ Let me start off by answering a few questions which have surely risen by now. The 'ERC' means 'Eco Recycling' -- basically this means the processor on the board has been used before. Not only does this help the environment by keeping an otherwise prefectly functional 68040 from going to the garbage dump, it also keeps the price of the chip down (and thus the price of the entire module). Of course, every chip is thoroughly tested by phase 5 before it is sold. It should be born in mind that 'ERC' has nothing to do with the familiar 'EC' and 'LC' abbreviations often found with Motorola CPUs. While on the subject, the CPU found on the Mk II module is a full 68040. That means that both the floating point unit (FPU) and the memory management unit (MMU) are present and functional. (The LC version omits the FPU, the EC version both FPU and MMU.) This means you can run programs like Enforcer and GigaMem or even the Unix clones NetBSD and Linux/m68k, which all require an MMU. With everyone turning to the 68060, you may wonder why I decided on a 68040. Simply put, I do not run the software which requires a 68060, like rendering or image processing programs. I program a lot, but I found that I spend far more time editing (and crashing) than compiling. And no 68060 will aid in speeding _me_ up. I can make quite an extensive list of what I do with my Amiga, but the bottom line remains that I was happy with the speed which the 68EC030 offered, 20% of the time it being too slow. In my opinion, buying a 68060 in this case is expensive overkill. Plus it helps the environment a little bit, but that of course does not take the power consumption of the 68040 into account... :-). After installation, I switched on the computer and waited. I could establish some speedup while the Amiga booted, but it was not a `WOW' experience. Various system information programs (SysInfo, AIBB 6.5, ShowConfig) showed the presence of a full 68040. Since the presence of an FPU could be established, I had to have a full MMU as well; this was confirmed by Lawbreaker. (Lawbreaker is a utilility to see if your Amiga is capable of running Enforcer. If you have an MMU, Lawbreaker generates 4 Enforcer hits, if not, less.) Memory Upgrades ~~~~~~~~~~~~~~~ I will deal with the software side of memory in the next section, but you should be aware of the fact that the SIMM slots might not be able to take any kind of double-sided SIMM. I was rather unpleasantly surprised when I found out the double-sided 8 MB SIMMs I borrowed from my mothers PC would not fit: the memory chips met the rim of the slot before the `click mechanism' could get a grip on the module. I did not have other SIMMs to see whether this problem would always occur, but I recommend to use 4 or 16 MB sizes when in doubt: they are populated at one side of the PCB only. Another point you should be aware of is that memory slot 4 is located under the front drive bay, and is completely inaccessible. In other words: if you want to populate this slot, you'll have to remove the drive bay (or the Cyberstorm). In my opinion, this is totally unnecessary, since there is plenty of space to move the four slots so that they take the room of three. AIBB Speed Tests ~~~~~~~~~~~~~~~~ Because I had forgotten to test the speed of my machine before I installed the Cyberstorm, I have no hard data relating to the old situation to present here. I dislike moving hardware around more than strictly necessary, and considering the rather tight fit of the card (it really snaps into place) I decided to leave it where it was. I did have a few AIBB modules of the A3000 (68030 at 25 MHz) and a Cyberstorm Mk I (68040 at 40 MHz) available, so I decided to use these. The A3000 is equivalent to the 4000/030, but tends to be a bit (usually about 5%) faster when access to memory is concerned. And it is of course slower in the graphics tests. TEST A3000 Cyberstorm Cyberstorm Mk I Mk II won Kickstart 3.1 3.0 3.0 by FPU mode 68882 68040 68040 EmuTest 1 5.79 5.68 I WritePixel 1 7.64 7.64 Sieve 1 3.50 2.87 I Dhrystone 1 5.37 5.37 Sort 1 4.73 4.63 I Ellipse 1 2.12 2.19 II Matrix 1 3.48 3.21 I IMath 1 3.73 3.73 MemTest 1 2.35 1.57 I TGTest 1 1.27 1.41 II LineTest 1 0.57(!) 0.68(!) II Savage 1 1.91 1.91 FMath 1 14.56 14.56 Beachball 1 6.51 6.67 II InstTest 1 4.29 3.90 I Flops 1 9.28 9.38 II TranTest 1 3.69 3.49 I FTrace 1 3.04 3.05 CplxTest 1 6.37 6.37 Memory Latency Indices Mainboard FastRAM 6.1 10.1 11.3 (smaller is Localbus FastRAM --- 7.1 8.1 better) Ok, let's discuss these (inherently questionable) values. The differences between the Mk I and II are in most cases completely irrelevant. I doubt that you will notice them in practice. That is not the case for memory-intensive jobs, however. According to these data, the speed at which the Mk II accesses its memory is decidedly lower than that of the Mk I. Since a lot of the work I do is memory-intensive I found these results to be rather disappointing. (The 'low' speed was confirmed by another program called Bustest, which measures the transfer speed to all types of memory, using byte, word and longword accesses. I have used both Bustest and AIBB on various systems, and their results are always close to eachother.) The ensuing discussion on Usenet showed that phase 5 always operate the SIMMs at 70 ns, regardless of the type installed. So I could in fact be measuring the difference between 60 and 70 ns memory here. However, since most SIMMs are 60 ns these days, I find it hard to accept that a company like phase 5 does not make use of this fact. A second reason the speeds differ so much could be due to the Mk I module using its second-level cache; AIBB does not store this information in its files. Whatever the reason, the Mk II definitely has a slower design, which is a shame. Despite this flaw, the other results are impressive. Of course, the increased clock speed and the CopyBack cache help to boost performance as well, but it's the result that counts. The integer speed has on average increased by a factor four. One can clearly notice this: the response of the Amiga is snappier and during CPU-intensive tasks a lot of the usual jerkiness (unfortunately still present on a 68030) has vanished. Another interesting fact is the floating point speed. A 68882 is not exactly slow, yet it is beaten time and time again by the 68040's internal FPU and the software emulation of unimplemented instructions. I therefore expect a `noticeable' speedup in floating point intensive programs. Real World Impressions ~~~~~~~~~~~~~~~~~~~~~~ The first application I tried is notorious for its demand for speed: Metafont. Metafont is a program which transforms a program-like description of a glyph (any character you see in print or on-screen) into a bitmap, so it can be printed. On average, a font (i.e., a complete set of glyphs) took about 90 seconds on my 68030. My new system requires about 30 seconds for the same font. Additionally, I expect the situation to improve a little when I switch to a SCSI-2 based filing system; Metafont requires a lot of file access and IDE can slow things down considerably. TeX was of course next. TeX is a typesetting program: it transforms plain ASCII text containing control sequences into typeset text. It's like a WYSIWIG word processor, but then with you _writing_ what the text should look like. I ran my Masters thesis (90 pages) through this program. Pages were generated at a `chee, this goes fast' rate. I got the feeling the computer was actually doing something; the 68030 definitely required more time to build a page. Pages containing lots of formulas still have a small pause, but this is hardly noticeable. The next program I tried was Ghostscript. Ghostscript is a screen-based PostScript-interpreter, meaning it shows your PostScript file on screen instead of on paper. Ghostscript requires a lot of calculations to build the display. The speed at which this now occurs was cause for the first `WOW'. Text-based PostScript files are built up faster than the eye can follow; graphics appear with a slight pause (the famous Tiger takes about 4 seconds on my Cybervision), but nevertheless at a very workable rate. At this point I decided to switch from OS-friendly to not-so OS-friendly programs. I unleashed my (small) collection of demos and games onto the Cyberstorm. As I expected, some of the demos crashed in record time; these included Origins Motion and Complex, Kefrens Desert Dream and Silents Demon Download (oldies, I know :)). Others kind-of worked (Faculty's Lethal Dose 2, Silents Ice, Andromeda's Nexus 7). And the few that did work did not contain many of my favourites. I was a bit disappointed at the modern stuff; are there really that little 68040's in Demoland? My games collection was already diminished to almost nothing after I bought my A4000/030; therefore I expected little problems. Populous 2, Rock`n'Roll, Civilization and Frontier (to name a few) worked ok. At this point I would like to point out that not all _old_ stuff fails to run: games like Defender of the Crown, Power Drift and Impact ran without even touching the caches. The rule here seems to be that if it can run on a 68030 (if necessary without caches) it will probably run on a 68040 as well. However, be prepared to reduce your (illegal??!) collection even further. (I wish to mention one strange exception here: Power Drift. On my 68030, I needed to Kickback into 1.3 to get this game working, with the 68040, it functions flawlessly under 3.0 -- it even has become much harder to play!) I already mentioned that Kickstart 1.3- and the 68040 do not like each other. I never got round to testing this, because none of the available softkickers and degrading tools succeeded in starting up the Amiga with 1.3. In other words: if the startup control screen fails, you can kiss the program goodbye. Next were the C-compilers I normally use: SAS/C 6.50 and gcc 2.7.2.1. (Compiling C is a process which is highly dependent on the speed of your hard disk as well, so the times presented are not solely the effect of the Cyberstorm.) SAS/C is a speedy compiler, but with a 68040 under the hood it really screams through sources. 130 KB of complex C-code (interpret.c from the MudOS v22b24 sources, ftp://ftp.imaginary.com/pub/LPC/driver/beta) was compiled in just under 20 seconds. It required 90 seconds on my 68030. With the relatively small programs (<50 KB of source) I normally write, the compile time is therefore reduced to almost nothing. gcc of course requires more time. The aforementioned MudOS sources compiled in 75 minutes on my 68030; the Cyberstorm system requires 35 minutes. When optimization is turned off, the required amount of time is reduced to 28 minutes. A lot of time is wasted because the compiler and the appropriate include files need to be reloaded time and time again. The compiler is also much slower than SAS/C: the interpret.c file requires 45 seconds without and 2 minutes and 15 seconds with optimisation to compile! At this time I decided it was time to test the GVP I/O Extender. The first connections with my provider (Term 4.7a 020+ and AmiTCP 4.0 demo) worked ok. There were a few errors during a ZModem download, but I contributed those to wrong buffersizes. However, I quickly changed my opinion when I found out that a direct 115k2 connection with my mothers PC lead to complete serial garbage and huge amounts of transfer errors when I tried to get something _to_ the Amiga. I spent hours trying to discover the cause of the problem, and eventually managed to track down the culprit: the good ole' CopyBack cache of the 68040. (This on-chip cache buffers writes to memory, as to speed up the processor. However, if a program requires that data and looks at the memory to find it, it will use _old_ data and thus go completely wrong.) When I turned off the cache, the problem vanished. This is not mentioned by either phase 5 or the manual of Term. After the cache problem had been solved, I could use the I/O Extender without further problems. Then it was time for some old-fashioned image processing using Photogenics Lite. This presented an unexpected problem. I never used the program because my 68030/68882 combination was far too slow for this type of work. (Hence the fact that I can't present data on other programs -- I don't have them :-)). The speed of the Cyberstorm proved to be adequate, but during the session I found out that image processing can be very fun. The hitherto impressive speed of the 68040 became a kind-of annoying pain, and for the first time I wished I had bought a 68060 or a (yet to be released) PowerPC. Serious image processing requires huge amounts of raw computing power, and in my opinion, the more the better. If your main application is image processing, the Cyberstorm 040 is _not_ your ideal purchase. I could go on and on about the various programs I tried, but I think I have written enough for now. The main conclusions which can be drawn from these tests is that the Cyberstorm gives your Amiga a solid and very noticeable performance boost. Software compatibility is very good if you stick to OS-legal stuff, and not so good if you don't. In some cases, other hardware may become the bottleneck, and it would be wise if you spent your money to remove these before buying more CPU power. DOCUMENTATION The documentation is good, if a bit short. There are several errors in the English text which are a little annoying considering the total amount of documentation. The documentation was written with the 68060 in mind and therefore some of the steps can be skipped: phase 5 indicate this on a separate leaflet which could have contained more information on the 68040 as well. The disk which is distributed with the 68060-version is not present. This is a bit annoying, since it contains a number of interesting utilities. phase 5 will give you a copy if you upgrade to a 68060, though. LIKES Speed, transparent presence, reliability, recycling of a 68040, simplicity of the module in comparison with its predecessor, price. DISLIKES AND SUGGESTIONS Slow memory interface, trouble with double-sided SIMMs, position of the fourth SIMM slot, lack of a CPU-cooler, short documentation, lack of disk. I didn't notice the `slow' memory interface in practice, because the speed of the card is still far above what I was used to. However, it is simply a shame to waste computing cycles in order to save a few bucks. I have one suggestion: get rid of these flaws! COMPARISON TO OTHER SIMILAR PRODUCTS I have never used other 68040-based cards, so a comparison based on experience is not possible. However, if I compare the literature with what I have seen and experienced so far, and neglect the SCSI-2 options found on all cards, the Mk II does not make it to the top three. Alas! This is simply because other cards (for example the Warp Engine, the Cyberstorm Mk I and the Apollo) have faster memory interfaces, which basically determine the maximum speed at which the CPU can process instructions. Only when used in combination with the accompanying SCSI-2 module does the Cyberstorm Mk II start to shine. Also, please note that the Mk I and the Warp Engine are not available any more. BUGS None found. Or perhaps one, but I cannot be sure. After a _real_ long day of hard work, my Amiga all of a sudden locked up when I wanted to redo an image in Photogenics Lite. I attribute this to a heat problem, though: it had been awfully warm in my room and my computer had been working at huge loads for the entire day. VENDOR SUPPORT Well, as we all know, phase 5 are not famous for their responsiveness concerning e-mail. During the test, I have sent the company several messages, all of which remained unanswered after two weeks. My advice would be to phone or fax them, but I have not tried this. WARRANTY phase 5 have devoted quite a lot of text to the issue of warranty. I will summarize the text in this document, but please refer to the original in case this becomes a necessity. I accept no responsibility when this summary and the original text differ! You get a 6 month guarantee, during which phase 5 will replace or repair your module free of cost if the card was broken due to material or production faults. Excluded are faults caused by `outside interference': improper usage, unauthorized repair or modifications and addition of new hardware and software. This includes upgrades of the _system_ hard- and software! phase 5 takes no warranty what so ever about the applicability of the Cyberstorm Mk II; the company also does not hold itself responsible in case of data loss, even when it has been warned about the possibility in advance. In case of problems you are instructed to turn to your dealer for assistance first. phase 5 will only accept returns when they have been assigned a so-called RMA-number; this number is given to you when the support department sees justifiable cause for you to return the module to phase 5. CONCLUSIONS The moment you've all been waiting for: the note. After a bit of thought I settled on 4 out of 5. The card itself is well-made and works as it should (and quite fast at that), but in my opinion there are too many small flaws in this design which could easily have been avoided. The card is not the ideal purchase for image processing freaks, but if you are looking for reliability and reasonable power at an affordable (even cheap-ish) price, this card is the one for you. COPYRIGHT NOTICE This review copyright 1997 by Maarten D. de Jong. Permission to distribute this text in any form is granted, as long as no modifications are made. The moderator of the comp.sys.amiga.reviews newsgroup is exempted from this requirement, but I trust him not to abuse this fact :-). If you distribute the text, I'd appreciate it if you dropped me a line informing me about the where and how. Reactions, comments and questions are welcome at my email address: M.D.deJong@stm.tudelft.nl. --- Accepted and posted by Daniel Barrett, comp.sys.amiga.reviews moderator Send reviews to: amiga-reviews-submissions@math.uh.edu Request information: amiga-reviews-requests@math.uh.edu Moderator mail: amiga-reviews@math.uh.edu Anonymous ftp site: math.uh.edu, in /pub/Amiga/comp.sys.amiga.reviews Web site: ftp://math.uh.edu/pub/Amiga/comp.sys.amiga.reviews/index.html