Interware Booster Clone 30-SE50F Accelerator

[Ubik]

Just a quick note with results on the new Interware Clone "Booster 30-SE50F" 50Mhz 030 w 68882 FPU accelerator in my SE/30 purchased from zigzagjoe on 68kMLA. Installed in the PDS without issues, requires no software, and performed a bit over twice as fast than stock (based on Speedometer 3). No issues with the BMOW Mac ROM-inator II ROMs and System 7.1. It's a modest speed up, and while likely not as fast as the Diimo 030 50Mhz, was a drama-free, cost-effective upgrade.

 

[JDW]

@Ubik

I read in that 68kMLA thread there were audio issues that pertained to the 1st edition Boosters like yours. Would you mind testing the following 2 games to see if the stock SE/30 audio (without Booster installed) sounds the same or different than with the Booster installed?

1. Spectrum Holobyte TETRIS
2. Prince of Persia

We used those games in our testing of WarpSE by @Zane Kaminski because they reveal if audio is pristine or not.

I'd also like to ask @zigzagjoe if Booster v2.0 produces pristine audio with the above 2 games when compared to a stock SE/30.

I don't have a Booster but am in the planning stages of doing a video review of the Synchr030/S RAM card, by @YMK , which is a 256MB accelerated RAM card (PDS) that I've been told plays nicely with the Booster, primarily because it lacks an L2 cache.

Thank you.

 

[zigzagjoe]

Original boosters and my V1 boosters have identical to stock audio in Tetris and prince of persia when run under System 7.1. Installing Sound Manager 3 or 7.5 will cause the audio glitches. V2 boosters have correct audio under both OSes.

Audio works differently compared to the SEs; The issue with SE/30/IIx sound lies in the hardware not giving enough wait states for the ASC FIFO. Not a problem at stock speeds as minimum cycle time on a 16mhz 68030 avoids the issue, but faster CPUs (ie. accelerators) are able to violate the timings it needs. Ditto with SWIM.

Note on Synchr030 compatiblity: Booster PDS works as long as it's the first card in the slot. Booster PDS will not work in the passthrough slot without modifications, and Socket booster will not work without modifications. A strong clock pullup can allow the otherwise nonworking cases to work, but YMMV, be prepared to tinker and it may still not work as it's a hack.

 

[JDW]

Installing Sound Manager 3 or 7.5 will cause the audio glitches. V2 boosters have correct audio under both OSes.

Thank you very much for your kindness in confirming that not only do Tetris and PoP sound great, but that Sound Manager is no longer an issue on the v2 Booster!

Note on Synchr030 compatiblity: Booster PDS works as long as it's the first card in the slot. Booster PDS will not work in the passthrough slot without modifications, and Socket booster will not work without modifications. A strong clock pullup can allow the otherwise nonworking cases to work, but YMMV, be prepared to tinker and it may still not work as it's a hack.

I am aware of the 180Ω pull-up resistor mod (for the Booster) that is necessary to put your Booster card on top of the Synchr030. Wow. 180Ω. A very strong pull-up indeed! Is a 1/8W rated resistor adequate for the current flow?

The reason that resistor mod is critically important (in my mind, anyway) is because when the Booster is the first/bottom card attached to the PDS slot, the Synchr030 sits atop it and then comes dangerously close to the Flyback suction cup. And while I've heard about "metal shield mods," the 180Ω resistor mod is without a doubt easier than making a metal shield. Indeed, a metal shield can be a short circuit hazard if done wrong. Adding one tiny resistor to the Booster makes the most sense (in my mind) for most people who intend to install both cards in their SE/30, if indeed that resistor works 100% of the time.

I haven't purchased a Booster in large part because I already own a vintage CPU-socketed PowerCache and a DiiMO (both 50MHz). But without a Booster, I can only read the experiences of others and then pass that info along to people in my video about Synchr030.

The Booster lacks an L2 cache, but when the 50MHz DiiMO (for example) is installed with Synchr030, it's cache becomes disabled, so performance SHOULD BE more on par with your Booster card.

​

Speedometer 3.23 results​

RIGHT: DiiMO-only (DiiMO Control Panel loaded)
LEFT: DiiMO w/Control Panel & Synchr030 w/Extension

(People always ask me how I get these "in color" and the answer is that I save the Machine Records, then open them on a modern Mac using Basilisk II, then take the screenshot.)

The opening post by @Ubik says that Speedometer "3" was used, but knowing the precise version may matter. (It absolutely matters between Speedometer 3 & 4 results.) But Ubik's CPU score of 9.83 is pretty close to my DiiMO (cache disabled) score of 10.42.

 

[zigzagjoe]

Booster is faster than the Diimo (when cache is disabled) as the Diimo has to latch data to/from the card-local data bus. For this reason the Booster is always faster at accessing the system bus compared to cached accelerators. Of course, cached accelerators typically have a ~ 90% cache hit rate meaning they run faster except when on cache misses, when accessing a PDS card / IO device, or memory writes.

1/8W should be fine.

Officiall,y my position is that the only "supported" configuration for Booster + Synchr030 is with the Booster PDS in the LB slot and Synchro on top. Other configurations may work, such as by bodging in a resistor or swapping cards around but it's not something I can guarentee or recommend except as something to tinker with.

Also, it is best for accelerators (if present) to be first in the PDS slot for signal integrity reasons. I've run into situations where accelerators will fail to work on passthrough slots due to this.

 

[JDW]

...the Booster is always faster at accessing the system bus compared to cached accelerators.

"System bus" via CPU data transfers, you mean?
I ask that question because in terms of the opening post's CPU score of 9.83 (in Speedometer 3.??) and my DiiMO's score of 10.42 (in Speedometer 3.23), I am curious if my DiiMO being faster than the Booster boils down to my use of Synchr030, or something else. (The opening poster is using only a Booster.)

Separately from that...
I am aware of some Booster benchmarks on the 68kMLA, but unfortunately, it seems those are not using Speedometer 3.23, nor they using same version of Norton System Info 3.5.3 I always use, which sets the PM6100 as the reference machine.

1/8W should be fine.

Thank you for confirming!

... it is best for accelerators (if present) to be first in the PDS slot for signal integrity reasons. I've run into situations where accelerators will fail to work on passthrough slots due to this.

Very interesting. Thank you for that. But of course, that basically leaves people who buy the Booster and Synchr030 some very hard choices. For example, nobody wants a zap from the CRT suction cup to fry all the PDS cards and possibly the motherboard too. But at the same time, I think a lot of people would be turned off at hearing they need to make a "metal shield" and figure out how to properly mount it too. And yet, your mention about "signal integrity" illustrates the importance of keeping the Booster as the bottom card. So yeah... There are some difficult choices to make.

In any case, thank you for the detailed and helpful info. And thank you for your hard work on the Booster itself! Hopefully, this thread will prove useful for those seeking more info on the Booster and how it plays with other cards such as the Synchr030.

 

[zigzagjoe]

Here is Diimo at 50mhz without cache vs the booster. These are (old) Speedometer 4.01 numbers. I believe this is after I modified the socket Diimo for 50mhz FPU operation (normally it is 25mhz), not that speedometer is an accurate FPU benchmark. I avoid speedometer these days as it's a mediocre benchmark.... System info 3.2.1 is what I normally use. However, I don't have my CPLD Diimo set up right now to get numbers with the cache disabled.

Note that Sieve scores are nearly identical as this routine fits almost entirely in internal CPU caches thus it scales nearly linearly with clock speed. Meanwhile Dhrystone is an extremely memory heavy trace, so the booster gets an advantage here. Booster technically is clocked at 47mhz so the diimo even has a slight edge on CPU clock.

When I say better at accessing system bus, I am referring to any CPU activity which results in an external bus cycle - an access not serviced by the CPU internal caches. This could be an instruction fetch, a data or write, I/O read or write, FPU access, etc. Cached accelerators may cache instruction fetches and data reads in which case those are usually serviced in two clock cycles (@accelerator clock speed). Very fast, compared to the system bus.

However on a cache miss a bus retry cycle is required (+ a few cycles penalty), then logic on the accelerator must coordinate the transceivers on the card to read/write from the system address and data busses as needed - incurring a delay that happens regardless of cache. For non-cachable accesses the bus retry is avoided, but the penalty of accessing off-card bus remains.

As the booster has the CPU directly attached to the system bus, there's not any differentiation between accessing hardware on the card vs. on the logic board (except in the special case of the onboard FPU), so those penalties are avoided. Additionally, the booster cuts timing a little in a few other places which results in a little additional performance on top of that.

Regarding multiple cards, realistically, apple designed these machines for one card; with anything more than one card nothing is guaranteed as every machine is different especially after 30+ years. With a machine this old one needs to be prepared for some degree of DIY, even if it's just bending chassis mounts to fit the second card.

Personally I took the approach of making metal support brackets which mount to the HDD bracket and give a nice path to ground. I posted about this some time ago at the MLA. Unfortunately, no one-size-fits-all solution here, the best possible I think would be a sheet metal bracket picking up the right-side CRT mounts top and bottom with an insulated side presented to any PDS cards.

 

[JDW]

@zigzagjoe
Thank you for another detailed and extremely helpful reply!

Your mention of using Speedometer 4.01 and Norton System Info 3.2.1 will allow me to use those two versions with my PDS-edition DiiMO and Synchr030 setup (and later my CPU-socketed PowerCache), so I can make better sense of my own benchmark results relative to yours.

Honestly, I think most highly of MacBench 3.0 (4.0 won't run on an SE/30). But most of the time, it crashes on my SE/30 not long after I start the tests (be that Main Tests or All Tests). Very frustrating, especially because the Arcs test best shows Synchr030 performance. Wish I could figure out why it does that!

Your 2-card stack photo reminds me of why I've not yet tested Synchr030 with my vintage Micron Xceed MacroColor 30HR video card. The Xceed card would need to be on top but its wire harness is too short for that! And making a long version of all those wires just for testing purposes isn't really something I'm enthusiastic about doing! Which means I won't. (Too much twisting required too.)

Anyway, you have a very nice metal shield, properly mounted. The main reason I don’t intend to show something like that in my video is because I try to make videos with the unspoken motto: “if I can do it, you can too!” I'm not especially adept or cunning, which helps people feel more comfortable repeating what I often show them. If a given thing gets too complex (for the average Joe), then people might only appreciate the video as a form of entertainment, but that’s about it. That goes against my goal of trying to get people to actually try something, rather than just watch me have all the fun. Even getting people to do a super simple thing like install a custom ROM SIMM (e.g., ROM-inator II) can be challenging. Some people in our vintage Mac community are simply afraid they'll mess up their beloved machine, and I can somewhat understand that sentiment. Hey... I'm the guy who once turned off his thinking cap and plugged in his precious 40MHz Daystar Turbo040 directly into the top slot of the TS Adapter, rather than the cache slot on the side. Power On -> Poof! (@Bolle and @joethezombie brought that card back to life.)

Anyway, that's the basis for all my comments about the metal shield in this thread. Realistically speaking, most people just won’t install one, even if I showed them a step-by-step path to doing it. With that said, there is significant meaning for YOU to show it because there are hardcore DIY types out there who love to get their hands dirty, don't mind making mistakes and accidentally destroying things, and those people certainly appreciate seeing the kind of DIY awesomeness that you’ve put together. So thank you for that!

At the end of the day, you’re right. The SE/30 was designed with only a single card in mind. The Do-it-Yourselfer always needs to be mindful of that. Even so, single cards are so boring!

 

[JDW]

@zigzagjoe
I found your 68kMLA post where you mention "Speedometer 4.01" and then link to this page on Macintosh Garden, but that page only offers 4.0 and 4.02. So I will assume you meant to say "4.02" and use that in my testing.

Anyway, I did find Speedometer 4.01 on one of my old hard drives, so I will upload that to the Garden tomorrow. I tried 4.02 and it works, but it crashes when I try saving the Machine Record. v4.01 doesn't crash when saving the records.

 

[zigzagjoe]

Interesting. The speedometer version I use is 4.01 verified by version strings in ResEdit, so not sure where the disconnect is as it absolutely came from either MG or MR. Possibly one of their files is misnamed.

MacBench is fine for certain uses. CPU benchmark is OK as an application-type performance trace but poor if you're looking to understand differences of accelerators since it's a black-box. FPU benchmark is useless also but the detailed video performance is particularly useful.

The Harnesses are not fun to make. Least favorite part of making my grayscale video cards. I specifically designed them to be slightly longer for that reason, but on my original machine with an Xceed + MacCon I made an extension instead.

I was able to get System Info results without cache on my CPLD Diimo. It reflects the same, Booster is faster except in benchmarks that fit entirely in CPU internal caches. Ditto for video, both in raw bandwidth as well as algorithmic tests (ie. ovals, etc). Unfortunately I don't have the Synchro benchmarks handy (I had one on loan before shipping to its final destination) but as I recall it was essentially identical to uncached except for improved Bmove and Mem RW due to 25% faster memory cycles on the Synchro.

Heh. Point taken on the DIY, especially considering I posted a ethernet+grayscale combo card I made simply as I wanted it for one of my machines. Everyone has their own perspective and preferences, after all.

 

[JDW]

I just added Speedometer 4.01 with more useful Machine Records to the Garden page just now. I am using that 4.01 version to run my Benchmarks. That way I can compare my results to yours in a meaningful way.

Below are my Norton System Info 3.2.1 benchmarks, setup in the same way as yours (in your previous post), with the Q700 as the Reference System, and showing "CPU".

As you can see above:

1. A 40MHz Turbo040 gives an SE/30 the greatest boost (basically a Q700).
2. Just down from that is the PDS-DiiMO running at 50MHz with DiiMO Control Panel loaded, of course (HDD is a MacSD with maximum clock of 57MHz).
3. Below that is the Synchr030/S Rev.A (currently selling product), with its Extension DISABLED, and with the same DiiMO installed but with the DiiMO Control Panel loaded — so basically the same performance as the DiiMO without Synchr030 installed.
   NOTE: The PDS-slot edition DiiMO (vintage card) is only compatible when using the Synchr030's "/DSACK" jumper setting. In addition, the DiiMO must be installed in the Synchr030's top slot because in my testing, the SE/30 freezes with horizontal lines if the DiiMO is installed as the bottom card. Also, the /ECS Tap wire need not be connected when a slot-based accelerator card is used.
4. Just below the CPU-socketed 50MHz PowerCache result, we have another Synchr030 result. Performance is greatly reduced because when you load the Synchr030's "Extension," you get RAM acceleration but the L2 cache on the DiiMO is disabled. The DiiMO's Control Panel was loaded for that test.
5. Then below the IIci result, you have another Synchr030 result without the DiiMO installed — just a stock SE/30 with the Synchr030, but with its Extension loaded, but /ECS tap not connected. You can see it yields a slight boost (28.7) versus the stock SE/30's CPU result of 26.2. It would be better with /ECS Tap, but I've not run that test yet using System Info 3.2.1.

If you have System Info 3.2.1 results for your BOOSTER v2.0, could you please post that?​

Thanks!

 

[YMK]

Adding one tiny resistor to the Booster makes the most sense (in my mind) for most people who intend to install both cards in their SE/30, if indeed that resistor works 100% of the time.

Agreed and that's how I run my v1.2 Booster. While the resistor is a hack, so is any multi-card setup in an SE/30.

Booster installation couldn't be easier and it runs fairly cool. It's a great card.

 

[zigzagjoe]

Actually, now that I think of it, that speedometer version may have come from jag's house... might be why it was missing.

I found the image with the Synchr030 benchmarks and some other amusing things. OS versions were not controlled as these are a collection of older benchmarks but it shouldn't make a large difference in these routines.

Carrera 45 is a socket Carrera @ 45mhz with 128K of cache
Diimo 58 128K is a socket Diimo @ 58mhz with 128K of cache
SE/30 PowerCache 50 is Powercache on Daystar adapter, 50mhz, 32kb cache
Booster Synchro is booster + synchr030 at 47mhz, no cache. (This should be a V2 card, but V1.x vs V2.0 have identical performance)
Booster 47 is just a booster (probably V1.0) at 47mhz, no cache.

Strictly speaking the L2 cache isn't disabled with the Synchro and Diimo, the "fast" address space is located in an non-cacheable IO address as Diimo only will cache ROM and RAM addresses. I did briefly modify my CPLD diimo logic to add the super-slot space, but the Diimo doesn't have enough address lines to cache for more than 128MB of unique RAM address space so it did not work.

 

[JDW]

Please forgive me for the multiple confirmations below, but people know me here as a MAN OF DETAILS. I like to confirm the important things when relevant.

OS versions were not controlled as these are a collection of older benchmarks but it shouldn't make a large difference in these routines.

After having spent countless hours through the years with various benchmarks, I can attest to the fact that System Software version does influence the results. I've settled on System 7.1 as being my preferred OS of choice when doing benchmarks on 68K Macs. System 6 will yield results slightly better than 7.1, but it's not too significant. You'll see slower results as you move up from 7.1 and go to 7.5.5, 7.6.1 and then 8.1. But you can also FEEL the difference in 7.5.5 and above too. To me, they feel like FAT PIGS compared to 7.1.

So when viewing all the results I present, it's best to assume I am using System 7.1. Although in my more recent benchmarks, I try to save that OS version in the benchmark result name ("S7.1").

Diimo 58 128K is a socket Diimo @ 58mhz with 128K of cache

Based on your CPU score of 83.3 (noticeably higher than my DiiMO score of 73.4 for some reason), I must assume the "DiiMO" Control Panel was loaded at boot time when you tested, correct?

SE/30 PowerCache 50 is Powercache on Daystar adapter, 50mhz, 32kb cache

It seems you also had the "PowerCentral" Control Panel loaded at boot time too, correct?
(Mine is a CPU-socketed edition PowerCache, which doesn't use the PDS slot. My System Info score was 67.2, comparable to your 69.2)

Booster Synchro is booster + synchr030 at 47mhz, no cache. (This should be a V2 card, but V1.x vs V2.0 have identical performance)

I understand the Booster (any version) has no cache on the card at all, which is why no Control Panel or Extension is required, which is nice. But just to confirm, you had the Syncr030's "Extension" loaded at boot time, correct?
And did you also have the /ECS Tap wire connected?
(Use of /ECS Tap for better Syncr030 performance requires you to have soldered the little PCB to the underside of your SE/30 motherboard and then run the wire from that to the single pin header on the Synchr030 PDS card.)

Strictly speaking the L2 cache isn't disabled with the Synchro and Diimo, the "fast" address space is located in an non-cacheable IO address as Diimo only will cache ROM and RAM addresses. I did briefly modify my CPLD diimo logic to add the super-slot space, but the Diimo doesn't have enough address lines to cache for more than 128MB of unique RAM address space so it did not work.

Yes, @YMK educated me on the fact that accelerators have to be programmed to cache the fast memory range of 0xC0000000-0xCFFFFFFF, which is not something my PDS-slot DiiMO or my CPU-socketed PowerCache do. Unfortunately that the limited address lines prevented you from making a DiiMO mod. That would have been pretty neat. Even so, your Booster is neat as it is. Small, compact, no need for INITs, and Synchr030 compatible out-of-the-box, or so I'm told. (I don't yet have a Booster.)

 

[zigzagjoe]

In my experience system info CPU tests are not significantly influenced by OS version (between 7.1 and 7.5.5) as the tests are algorithmically heavy tests of specific CPU traces rather than an application benchmark (such as MacBench) that makes heavy use of Toolbox calls. As a result, I only see minor differences in performance between the two in this specific benchmark. Also, not loading the kitchen sink of extensions and control panels helps. I tend to use 7.1 these days as it does boot notably faster, useful when iteratively writing DeclROMs and stability testing.

See below for other answers.

Based on your CPU score of 83.3 (noticeably higher than my DiiMO score of 73.4 for some reason), I must assume the "DiiMO" Control Panel was loaded at boot time when you tested, correct? Yes. Performance difference is due to running at a higher clock frequency with additional cache vs the original design. Presented for curiosity value.

It seems you also had the "PowerCentral" Control Panel loaded at boot time too, correct? Yes

I understand the Booster (any version) has no cache on the card at all, which is why no Control Panel or Extension is required, which is nice. But just to confirm, you had the Syncr030's "Extension" loaded at boot time, correct? Yes

And did you also have the /ECS Tap wire connected? No, it does nothing productive on accelerators unless you can meet very specific timing requirements. It should not be connected except for a stock CPU.