ñAugust 13, 1999ð ñ-------------------------------------------------------------------------------ð ó Experiences with Centurbo II (Rev-B) and Centek.ð ó-------------------------------------------------------------------------------ð by Anders Eriksson (ae@atari.org) --------------------------------- First of all, this 'review' is very objective. I got my Centurbo II for free, and hence I might be overlooking some bad sides other people who actually spent money on it is bothered about. 1. Hardware 2. Unpacking 3. Installation 4. Using 5. Problems 6. Final thoughts ò1. Hardware.ð The Centurbo II is a circuitboard which fits to the internal expansionslot of the Falcon. Plus that it has the 'usual' Nemesis/Centurbo 1 motherboard over- clockings and patches. In a few lines, this is what your Falcon has after installing: - A new 68030 running at 50 MHz. - A socket on the Centurbo II board to put a new FPU in. It will be clocked at 50 MHz. - Overklocked DSP to 50 MHz. - Overklocked ST-RAM/BUS to 25 MHz. - IDE-Interface patches to allow newer and larger IDE- harddisks at higher speeds. - Overclocked Videl to 50 MHz. - 16, 32, 64 or 128 MB Fastram. Accessed at 50 MHz, 32-bit with burst. ó2. Unpacking.ð When first hearing of the Centurbo II, reading the news from Centek one would belive this is something they have put their entire soul into. Getting an absolute perfect product, in every detail. Not even god managed to create a perfect world, so I took those newsletters with a bit of salt. Later, about 8 months to be precise, the Error In Line'99 convention took place in Dresden/Germany. Their Falcon democompetition proudly stated a Centurbo II accelerator as first price. Not bad! I entered a little demo to the competition, and to my surprice it won, which means I got the Centurbo II card. Back home from the party, I unpacked it (ok ok, I unpacked it earlier on the way home too) and took a careful look of what was there, here's the list: - An antistatic bag with the Centurbo II card. - A plasticbag with a short wire, a few photcopied a4 pages, a switch, a resistor and a little veroboard peice. Oh! I almost forgot, you got a sticker saying "Powered by Centek" as well. The next thing to notice is that the installation manual really did not motivate me to actually have the card mounted. It was a bad photocopy, and unclear in most ways. The english is very badly checked, and can sometimes be confusing. This is what happens when you take French grammar and English words. If you have tested the Altavista Babelfish translator, you will feel at home with Centeks English as well. At this stage, I was lucky to have taken some salt with with the Centurbo II newsletters. There is zero feeling about professionalism once you receive the card. But maybe it's just a French way of doing things? Let's go on! The board itself looked robust and welldesigned. Except for a terrible cooling fan; it did not fit as it should and has a tendency to easily fall off. That help a great deal to enhance the "low-class" feeling of the entire product. If they can't even fit a fan, will the damn board really work? ò3. Installation.ð Here comes the horror part. Strip it naked, cut some legs, sue it together. Sounds like the Frankenstiens Monster movie, but it isn't. This is what happens during a Centurbo II install. If you dare, go on reading. All the hopeful promises from Centek about "easy install" quickly vanishes when you read what has to be done, or what about this: - Cut a leg from the existing 68030! - Cut several legs from other chips! - Cut two paths on the motherboard! - Cut your IDE cable! - Solder over a dozen points, including soldering directly to the motherboard and to the old 68030! Yes, and it really isn't any easier than it sounds. The pins of the 68030 are very tiny, and cutting tracks on the motherboard isn't the easiest thing. Especially not if you like it done with precision. For any Atarian with a weak heart, do not do this yourself, you'll fall into tears seeing the machine being butchered! Anyway, if you against all odds manage to get the soldering and cutting done, it's time to install the actual board. This is a peice of cake, 10 seconds and it's there. 5 secs more to fit the EDO SIMM module. Now the fun begins (hopefully) ! ò4. Using / performance.ð Having read the instructions, it is told that you should boot the machine, and run the "putflash" program to enable the Flashboot setup, and then also be able to enter the turbo modes. However, in my case the Flash was already loaded, and it started directly when I switched the machine on. And I mean directly here! From power-on until the Flash- bios menus are on screen it is like half a second delay! The next thing you notice is the extreme ugly colours and font of the flashbios! Jesus, did they hire a kindergarden kid with crayons to make this one? The flashbios is configurable, and the way you move around the menus is crap. Like a 5 min hack from a lazy programmer. Some would even suggest it's worse than those sh*t crap dhs musicdemo-menus ;-P In the flashbios you can toggle a few things, like the boot resolution, which harddisk driver to use, dsp and ide waitstates and a few things more, such as: Decide to boot in three ways: - Dolmen. This is an alternative OS from Centek, which unfortenly was not delivered. I wonder if they (Centek) ever got a Swedish mail about that name, it's sort of a synonym for penis. - TOS 4. This is the good old TOS, no differencies, except that the machine is running alot faster! Yep, the CPU is running in 50 MHz here, but there is no FastRam supported. - TOS 7. Yeah. This is fullspeed 50 MHz with Fastram! The TOS is patched and put into Fastram. There are other things such as a new error handler (replacing the old bombs..) which tells a bit more specific what has happened. You are also allowed to exit the program causing the error, you can try to continue (really useful!) or reboot the machine. A real improvement to the old bombs. So after having booted in TOS7 for the first time, you feel "so this was all". And yes, it actually is. The machine runs a bit quicker. But well then, how much faster does it run? This is a question without answer, as it performs differently from application to application. Depending if the app can make use of fastram, how dsp-intense it is and which videomode it runs in. If the application uses no dsp, and can't use fastram, the average speedup is 2-3 times. Depending a bit which video- mode and how good the code is (eg, cache optimized). On the other hand, if the application is pure DSP the increase stays at around 60%. The best case for noticeable speedup is if the application uses no dsp, has good fastram support and is running in an intense videomode. Then you can end up with some 6-7 times speedup. A good example of this is if you run a packer in GEM with a high screenmode. Here follows a few tests I've done, in the no-turbo modes, the extended resolutions have been made with BlowUP. Test: LHA packing Resolution: ST-High Normal Falc: 2 min 53 sec CT2 Falc: 0 min 44 sec Test: LHA packing (same file as above) Resolution: 896*704*4bit at 64Hz Normal Falc: 5 min 57 sec CT2 Falc: 0 min 47 sec As you can see from these two tests, the extended resolution almost killed the standard Falcon, while the Fastram on the CT2 saved the speed to almost the same level as with ST-High resolution. Check the test below, it's the same file again but with twice as heavy videomode. Test: LHA packing (same file as above) Resolution: 896*704*8bit at 64Hz Normal Falc: not possible CT2 Falc: 0 min 48 sec As you can see, still practically the same speed! Fastram does wonders! It isn't hard to create a demolike application that uses fastram all the way. These are the steps: A. Calculate how large the program is and add space for stack (in this case $1000 bytes). move.l 4(sp),a5 ;address to basepage move.l $0c(a5),d0 ;length of text segment add.l $14(a5),d0 ;length of data segment add.l $1c(a5),d0 ;length of bss segment add.l #$1000,d0 ;length of stackpointer add.l #$100,d0 ;length of basepage move.l a5,d1 ;address to basepage add.l d0,d1 ;end of program and.l #-2,d1 ;make address even move.l d1,sp ;new stackspace B. Do an mshrink() call to return all mem we don't need. Using the registers from previous calculations. move.l d0,-(sp) ;mshrink() - bytes to keep move.l a5,-(sp) ;address to block (basepage) move.w d0,-(sp) ;dummy move.w #$4a,-(sp) ; trap #1 ; lea.l 12(sp),sp ; C. Reserve memory with the mxalloc() call with ramtype '0' requested. This asks for stram only. Should be called to reserve memory for screenbuffers and sample-dma buffers. If d0.l returns '0' there wasn't enough free stram, and you should pterm(). move.w #0,-(sp) ;mxalloc() - stram only move.l #bytes+256,-(sp) ;number of bytes+256 move.w #$44,-(sp) ; trap #1 ; addq.l #8,sp ; tst.l d0 ;check if there was enough ram beq exit ;nope add.l #256,d0 ; clr.b d0 ;make screenaddress even move.l d0,screen_adr ;save address D. That's it. ó5. Problems.ð Centek promised alot. 100% compatibility, no errors with anything. Better sound than CT1 and Nemesis. New improved IDE interface that is much more compatible. Normal mode which would run anything. Easy installation. Great support. It was all hype. Nothing of the above feels familiar with my situation. Here's the real thing, how I experienced it. * Better sound than CT1 and Nemesis. I can't talk about CT1, but for sure the Nemesis machines I've tried had just as good sound as my CT2 machine. Eg, identical to a normal Falcon, not better, not worse. * New improved IDE interface. Bullshit. Perhaps it's faster, perhaps it can use newer harddisks. But better compatibility? No chance. My own IDE disk stopped working. * Normal mode runs anything. This was professional craptalk. Centek knew themself that they didn't wire the DSP-INT, not in turbo, not in normal modes. Some DSP programs are using DSP-INT, and will crash, Turbo or Normal mode doesn't matter. So how on earth they could say the machine was 100% compatible has no other answer than that they lied to get more customers. If you tried to talk to them, pointing out that the CT2 did have a problem here, they replied extremly arrogant in the style of "crappy coded software uses DSP-INT". Yes perhaps crappy programs does, but still, those crappy programs does not run with CT2! And, in my opinion some of these programs are not crappy at all. MegaPlayer is one example. If you upgrade to CT2, you can kiss the MegaPlayer goodbye. * Easy installation. As said above: If you are not extremly good at electronics and have professional tools; this is nothing for you! The easy install arguments was false, and I belive Centek themself realised it too. They no longer sell the cards without doing the fitting themself. * Great support. What a joke. Centek is known for being rude to their customers. Constantly "downranking" them. Have them feel like imbecills. Same thing happened to me. I was asked to read the FAQ, even though the FAQ didn't contain anything of what I asked for. I was asked to read french docfiles. I know no french! I asked them alot of questions, they might have answered 10% of them. It should be said I've only tried their E-mail service, not phone. However their e-mail service seems to be there to push down their customers so they don't dare asking anything more. My own Centurbo II problems are a history of themself. Below is a micro-version of what happened, and made me more or less Falconless for three months. After the actual hardware installation first had been made, I started to test some software. I quite quickly noticed that a certain type of applications didn't work at all: DSP-Audio ones. MP2 Audio, FlaySID, DSPMOD etc all crashed. And I knew these programs worked with a friends CT2 machine. Further problems were that the auto-folder text was a bit screwed, some letters had the wrong offset. My IDE disk didn't work. The 32mb EDO SIMM was seen as 64mb by the CT2. The standard mode was terrible. Crashing much more than the turbo mode. So I tried to e-mail Centek for the first time, to hear if they have any tips for me, how to solve these problems. The first reply I got was quite alright. They probably answered 50% of my questions. And not a single pointer to a FAQ or french docfile. The auto-folder textproblem was due to my blitter not keeping up with the 25 MHz speed correctly. It shouldn't have any other sideeffects regarding to Centek. Okey, phew.. one problem down. The IDE disk however was a downer. I was told my IDE was too slow. Great. Centek wanted to know the brand of my EDO simm and how many chips it had. Their tip about crappy standard mode was a bad 32MHz signal. The DSP-Audio problems was probably from an unstable SDMA. So I began composing a reply to their mail, telling the EDO-SIMM type, and here I made my first misstake; I forgot to count the number of chips on the SIMM. At this time I had written my own ram-test program that actually verified the SIMM to be a 64mb one. The dealer at the shop had given me the wrong one. Hurray, 32mb extra for free. I don't complain! However, Centek did. I got to know with alot of CAPITALS and '!!!!!!!!!!!!!!!!!!' marks that I forgot to tell them the number of chips on the SIMM. They didn't have any suggestion about the DSP and Normal- mode problems other than "Send the Falcon mb+ct2 to us, we will fix it. We perfectly know the Falcon". Yes, even the grammar misstakes were like that. Well, I decided that we (me and my friends who helped with the hardware installation) should make a last try to check everything before sending back the machine to Centek. It didn't work out at all. There was no difference made. In the meantime, I had given Centek the number-of-chips information about my EDO SIMM, and got to know from Centek it was a 16mb SIMM, not 32, not 64. Bahh what nonsense. At this time I was in the middle of fixing the 4col intro and waited with sending the machine until that was done. So, I shipped the Falc, the cost was approx 25 euro. I had most of luck getting a replacement Falcon while the other was in France. After approx 2 weeks idle, no word from Centek at all. I mailed them and asked if they had gotten the Falcon. No responce. A week later they wrote and told me the Falcon was fixed, and had been sent back. Yippie! However, they did send it with the slowest possible mail. So from the date it was sent, until I got it was nearly two weeks. It was interesting to see what modifcations they had done to our install. An example: In the instructions of fitting, it is clearly stated that you should not, absolutely not have any leads going over chips. And still, Centek themself had even taped a lead right on top of my videl! So, assembling the machine, and testing.. will it work now? Answer: No. The machine was stone-dead. Not even the Flash- bios menus appeared on screen. I shut the CT2 off, and what a surprice, the machine started. Alright, checking so the CT2 module is fitted well, and the EDO SIMM as well. Yep, seems to be alright, let's try again. Same thing, no life at all from the machine, except the CT2 led. Next was removing the EDO SIMM and see if that made any difference. And whoa! It did. The machine started, showed the flashbios. Damn, have Centek messed up the EDO controller on my CT2? Fixed anohter SIMM to test with. This one did however work! It was a slow old SIMM, and the CT2 was very very unstable with it. Not exactly surpricing as the CT2 needs 60ns SIMMs, and this one was a 80ns. So, it looked like that my 64mb EDO SIMM was damaged. I wrote Centek and asked what could have gone wrong. They replied that I should try downgrading the Flashbios, as they had put a beta version there! Cheez... Okey, downgraded the flash and the machine worked with my 64mb SIMM. Why Centek install beta flashes to their customers is a mystery, or why they reset the NVRAM another (my Falcon came back with French nvram setups!). Not a big deal if you are familiar with setting up the NVRAM. Okey, now I am supposed to have the machine working anyway. 100% as Centek claimed in their commercials. Let's test. Xboot and NVDI have a hard time to boot, one time of three they crash upon booting. Centek replies that they shouldn't. But gives no suggestion or other ideas why it does or how to fix it. If the machine comes thru Xboot and NVDI, it works pretty well however. The apps I expect to work, does work. The ones with DSP-IRQ fails, even with Centeks feeble attempt of a patch program for this. So, time to test old-time compatibility. Switch off the CT2, load some old stuff. It works like shit. The machine is way less stable than in turbo mode! I quickly realise that Centek can not have made any tests in normal-mode at all. Even though I did tell them about those problems. ò6. Final thoughts.ð At the moment, what hits my mind when saying "CT2" is nothing but "problems". Centek have produced 240 cards to date, and I feel sorry for all of the people who put money to get into trouble. Because it is trouble. Especially if you depend on the Falcon a bit, eg. use it for daily purposes, not only for watching a demo now and then. Considering the high price of the CT2, it cannot be said to be priceworthy at all. Especially not if you use the machine professionally, and loose money with every minute the machine is not working or away for fitting. Becuase with Centeks slow snailmailing, it takes you ~3 weeks to get the machine back. The speedincrease is what can be expected. Nothing over the ordinary, but still it's a nice feeling compared to a normal Falcon. The extra memory is great though. Especially if you run the Falcon with a bit more demanding operatingsystems such as Linux or MiNT. MiNTOS boots in a flash. N.AES starts in a kick. If you give the MiNT kernel a few megs memory for diskcache, it speeds up the disk operations a whole damn lot. Changing resolution in N.AES and having the diskcache up for some megs, the change is done within a couple of seconds. It hardly touches the harddisk, all goes from fastram diskcache. MP2-Audio goes down from average 80% DSP-Load to average 50% when playing 192kbit files. Plenty of time to add extra sound effects. There are many negative sides, and one that never escapes you is the damn ugly boot sequence. Starting in cyan/white colours, continues to blue/yellow and then sticks to the normal nv-ram bootsettings. If you enter the flashmenu, the colours are even worse. A few demos run fine with the CT2, those are easily counted though. And with running fine I mean those who actually take use of extra hardware, not just those that look the same or running too fast. The CT2 is almost a "vga only" product, so it's a big dissapointment almost all demos are rgb only. Again, there are some that both runs with vga and take use of extra hardware. The ALIVE demo by FUN is an excellent example. Okey, that's that. This review is my very personal opinion and other people might think different. -- ò Anders Erikssonð ñ ae@atari.orgð