PowerCache Card in IIci -- Why So Slow?

[This Does Not Compute]

I'm testing out a DayStar Universal PowerCache card in my IIci. It's the 50MHz 68030 version, so it *should* be wicked fast. And it's certainly faster than the IIci in stock form, but the benchmarks suggest it's not performing as it should.

The IIci has 20MB of RAM and is running System 7.0.1 with 32-bit addressing enabled. I have the PowerCache control panel installed, and both the cache and FPU are enabled, and it says the 50MHz 030 is active. The thing is, MacBench 2 says that the system (with card) only performs at 60% of what a IIfx does (that one has a 40MHz 030), though FPU scores are comparable. Speedometer rates the CPU at .74, while an entry for "50MHz SE/30" (there's no saved score for a IIfx) has a CPU rating of .87. (A stock IIci is rated at .40.) Some individual tests (whetstones, etc) are faster with the PowerCache, but not all.

For a card that was advertised as making a IIci faster than a IIfx, what gives? Is this a bus limitation in the IIci that the IIfx doesn't have, or does this card not benchmark well?

 

[Garrett]

What version of Speedometer are you using? I've got this same card, and could run some benchmarks on my end to compare.

 

[This Does Not Compute]

I'm running Speedometer 4.0.2. I might try upgrading to System 7.1 next, to see if that makes a difference. Also of note is that the machine has a BlueSCSI v1 installed, but I'd be surprised if that was a factor.

 

[Fizzbinn]

Are you using onboard video? Using a NuBus video provides a nice ~10% CPU bump as the CPU doesn't have to compete with the onboard video for access to bank A RAM.

 

[retr01]

That is correct. NuBus video card redraws video faster, especially an accelerated NuBus video card such as Radius or Apple's 8*24 card.

 

[trag]

I don't know if it affects your situation, but I'll throw out that the IIfx has faster DRAM access, at least for writes, because it buffers the Write operations immediately to free up the CPU to continue about its business.

That wouldn't explain a discrepancy between IIci and SE/30 results, though. I like the above hypothesis of on-board video slowing the system down.

 

[This Does Not Compute]

For clarity, this 50MHz card was advertised as being *faster* than a IIfx, which was 40MHz. Given that the IIfx was likely better optimized for speed than the IIci was, I'd expect the two to maybe be similar in benchmarks, and interestingly their FPU results are indeed very close. But the CPU benchmarks are significantly lower than a IIfx, and only 30 percentage points higher than a stock IIci -- that just doesn't seem right, and I'm having a hard time believing that the onboard video is the culprit. Here's an example of my results (there are two entries for the PowerCache card, showing that System 7.0.1 vs 7.1.2 has negligible impact):

Out of curiosity, I turned off the cache in the DayStar control panel and the results got noticeably worse, so it does appear that (at least part of) the cache is working. The CPU on the card is labeled as being a 50MHz part, and I have no reason to believe it's been tampered with or swapped. The control panel and Gauge Pro both report it as a 50MHz CPU. So I'm curious to see the results @Garrett comes up with for his card!

 

[ApfelKlassik.de]

Hi Colin,

if talking about system score my guess would be a combination of the "advanced IIfx memory handling" plus the bus speed difference - the IIfx also sports a 40MHz bus.

Cheers!

 

[retr01]

The IIfx is

• almost twice as powerful as a base 25 MHz Mac IIci
• about 60% faster than the IIci with a 32 KB L2 cache
• two 6502 processors to manage the floppy drive(s), ADB ports, and serial ports
• reducing the CPU load

(Paraphrased from Low End Mac’s “1990: The “Wicked Fast’ IIfx and the First Consumer Macs” article)

It looks like it’s how the IIfx and IIci boards were designed. IIfx was intended to take the load off the CPU as much as possible, utilizing secondary processors to handle various tasks that would have otherwise loaded the Motorola 68030 processor.

After that, it seems the Quadras blew the IIfx out.

By the way, I have the 8•24GC that accomplished specific tasks 5 to 30 times faster than previous Apple video cards. That takes off more burden from the CPU. The kicker is that it has an onboard AMD 29000 RISC processor--Apple’s first RISC product.

 

[Fizzbinn]

If you don't have a NuBus video card and need to use the built-in video you can speed up a IIci by using up the all the memory in Bank A (which is used by the built-in video) with Disk Cache and/or a RAM disk via the memory control panel. That forces programs to run out of Bank B which is un-encumbered by the built-in system.

Hopefully you have 4MB in Bank A and 16MB in Bank B, if not, swap your SIMMs. Its a waste of memory but it did make a notable difference in my testing a while back. I put 4 256KB SIMMs in bank A when I tested but don't see why you couldn't just eat up 4MB vs. 1MB.

 

[trag]

two 6502 processors to manage the floppy drive(s), ADB ports, and serial ports

Unfortunately, in Mac OS, Apple never took advantage of the IOPs on the ports. There's a pass-through software path that ignores the IOPs and just goes direct to the serial and SCSI chips, etc. and that's what the Mac OS uses. Supposedly, AIX had support for the IOPs.

 

[ApfelKlassik.de]

I assume you wanted to write: A/UX

 

[retr01]

If you don't have a NuBus video card and need to use the built-in video you can speed up a IIci by using up the all the memory in Bank A (which is used by the built-in video) with Disk Cache and/or a RAM disk via the memory control panel. That forces programs to run out of Bank B which is un-encumbered by the built-in system.

Hopefully you have 4MB in Bank A and 16MB in Bank B, if not, swap your SIMMs. Its a waste of memory but it did make a notable difference in my testing a while back. I put 4 256KB SIMMs in bank A when I tested but don't see why you couldn't just eat up 4MB vs. 1MB.

Interesting. I have 128 MB in the IIci. It doesn't help with video because it is not accelerated and still burdens the CPU no matter what. Even with the 8*24GC helping out, the video redraws with the photo thumbnail preview program weren't much faster. So, if the program was poorly coded, it might have been slow on the IIci or IIfx, no matter the acceleration.

Back then, programmers may not have coded to take advantage of this and that, much like programmers today who do not take advantage of multi-cores.

Hmmm...

 

[retr01]

Unfortunately, in Mac OS, Apple never took advantage of the IOPs on the ports. There's a pass-through software path that ignores the IOPs and just goes direct to the serial and SCSI chips, etc. and that's what the Mac OS uses. Supposedly, AIX had support for the IOPs.

And that's an example of poorly designed and/or executed coding, IMO.

 

[trag]

I assume you wanted to write: A/UX

Yes. I felt like I was wrong there, but couldn't remember where I was going wrong. AIX, I think was what the ANS ran. Anyway, yes, I meant A/UX.

 

[Fizzbinn]

Interesting. I have 128 MB in the IIci. It doesn't help with video because it is not accelerated and still burdens the CPU no matter what. Even with the 8*24GC helping out, the video redraws with the photo thumbnail preview program weren't much faster. So, if the program was poorly coded, it might have been slow on the IIci or IIfx, no matter the acceleration.

Back then, programmers may not have coded to take advantage of this and that, much like programmers today who do not take advantage of multi-cores.

Hmmm...

The issue is if you use the RAM based built-in Video on the iici you lower the system's CPU performance, as RAM Bank A becomes slower to access by the CPU. BUT this is ONLY IF you also use RAM Bank A for running system/applications. If you tie up the memory in Bank A (above that which is used by the RAM based built-in Video) with other uses like Disk Cache, RAM Disk, or some program that grabs that memory and does nothing with it, your System and Applications will be forced to run out of RAM BANK B which has no RAM based built-in Video memory performance leach.

From the Iici Dev Note, page 23:

"Video accesses affect only Bank A memory access because the data bus between the RAM banks can be disconnected by an F245 buffer as shown in Figure 3-2. This allows the RBV to fetch data from Bank A without interrupting CPU access to Bank B or I/O devices. Each bank of RAM is accessed independently by the MDU, so it can decode addresses for the CPU and the RBV at the same time without interference."​

​

So if you are using IIci RAM based built-in Video, using the trick above increases system performance similarly to using a Nubus Video card. It doesn't do anything for video performance, which using a NuBus card likely would (and of course potentially support higher resolutions and bit depths), but its noticeable.

I don't have easy access to my Iici at present (about to move) but my recollection was about a 10% CPU benchmark increase, which is comparable to the performance increase you get with the Apple IIci L2 cache card.

I think folks are more familiar with this trick on the IIsi which has the same memory, RAM based video setup but with BANK A being 1MB soldered to the logic board vs. the SIMM slots the IIci has for Bank A (they both have SIMM slots for Bank B).

For this thread, this might all be moot if @This Does Not Compute is using a NuBus video card, but he's not and doesn't have one he can likely get a nice bump with it.

 

[retr01]

Thank you @Fizzbinn, for clearing that up.

 

[Garrett]

I tried a preliminary test with my hardware last night. MacBench 2 shows it slower than a IIfx for me as well - scored almost the same as Colin from what I can tell. System 7.1, internal BlueSCSI v1.1, 20MB RAM (16MB in bank B), no video card. Something seemed buggy though, because the system kept apparently running out of RAM, and I was never able to disable the card. I have some more tinkering to do!

 

[ScutBoy]

If you don't have a NuBus video card and need to use the built-in video you can speed up a IIci by using up the all the memory in Bank A (which is used by the built-in video) with Disk Cache and/or a RAM disk via the memory control panel. That forces programs to run out of Bank B which is un-encumbered by the built-in system.

Hopefully you have 4MB in Bank A and 16MB in Bank B, if not, swap your SIMMs. Its a waste of memory but it did make a notable difference in my testing a while back. I put 4 256KB SIMMs in bank A when I tested but don't see why you couldn't just eat up 4MB vs. 1MB.

Wasn't there an app called something like "Memory gobbler" or "grabber" or "eater" that did this for you on the IIci?

 

[Fizzbinn]

Wasn't there an app called something like "Memory gobbler" or "grabber" or "eater" that did this for you on the IIci?

IIsi RAM-Muncher INIT:

IIsi RAM-Muncher INIT - Macintosh Garden

This extension is for Macintosh IIsi users, running System 7 or greater, only. The IIsi does not have VRAM, and video circuitry and the CPU must take turns to access the motherboard-mounted 1Mb that contains the screen image. However, this screen image can only be up to around 330K, leaving 700...

macintoshgarden.org

I would think it could work on a IIci but probably only if you have 1MB of RAM in Bank A.