SeriTek/1SM2

[Ataman Honcho Honchev]

Doing a huge overnight test now with the post-SeriTek/1SM2 (SImagic2LP) cardbus / PCMCIA card.
This test is done with the genuine SeriTek/1SEM2 hardware, but the 6.0.0 (completely new) driver.

The "victim" is the last series of Titanium G4. I got a similar 1GHZ machine from the local aftermarket as well, but it needs to be repaired / screen exchanged. This one is in fairly decent shape but is the slower, 867MHZ version.

The SeriTek/1SEM2 is using a popular Silicon Image 3512 chip which has a documented errata.
That errata wasn't known back in 2003. In 2023 the workaround is well-known and of course it is implemented since early "beta"-s of 6.0.0.

In 2003 I had to limit the simultaneous write access to both channels (that mitigated the problem).
With the errata workaround implemented (also applies to 3114!) I see no problem while writing to both channels the same time.

One is clear: the PCMCIA is not a speed demon, in particular on writes. The reads are OK, but the writes through PCMCIA - even if the simultaneous access is not a problem anymore is not going to be faster, than 30 MB/Sec.

Let's see, how it behaves overnight.

I am using two "disqualified" SSD-s: one from Kingston, one from AMD.
They are "disqualified" due long use times and "going to fail soon" status in SMART - but perfect to be abused.

These two 240GB SSD-s were set up in RAID-0 (SoftRAID 2.22) and now proceeding with FWB bench test, asked for a very long transfers and huge number of tests (guaranteed to keep the computer busy for a day).

Hope everything is OK. Other than that I will test the same with a 3112-based PCMCIA card and fix one "new" bug of mine when drives going to sleep - but the computer is thinking that the entire system is asleep.

If that is fixed and if the test is OK I will move one step forward. ;-)

 

[Ataman Honcho Honchev]

In the mean time the "(Intenso)®" SSD-s were rehabilitated: the problem was my newly "introduced" bug in the SIM.
Bug fixed yesterday.

Actually, Intenso was looking "bad" because it reacted "way to fast".

Writing on both channels the same time still lowers the performance, so on CardBus / PCMCIA cards the write to, say, channel 1 will wait until the write to the channel 0 is done.

In 2003 it was awful - but in 2023 even with non-SSD-s it is perfectly fine: the speed of the bus (ca. 100MB/Sec) saturates even by using a modern rotating disk. And now the CardBus can reach that speed.

So it is not that bad.

 

[Ataman Honcho Honchev]

As we see above, Apple System Profiler is not very kind to CardBus cards: the SCSI bus is not showing up properly anywhere. So let's fool it!

Now the SIM driver temporarily(!) patches the CardBus bridge properties the way that it appears as SCSI bus.
Of course only as long as the CardBus card is inserted. When I remove it, everything is back to normal.

The reasoning: as long as the CardBus card is inserted no one is going to use the CardBus bridge for anything, but the inserted CardBus "SCSI" I mean, SATA card. When the SATA card is removed, CardBus bridge shows up again as CardBus bridge. Yikes...

There it is, now in "Beta 37". I am slowly fed up with that SATA, MacOS "9" and PowerPC. Can't wait for NVMe, SAS, macOS X (and post-X) with Apple Silicon. No wonder, 37 "betas"...

You see the same PowerBook with CardBus magicaly now being reported as SImagic2LPCB (for SeriTek/1SM2 hardware). For 3112-based cardbus it is SImagic2CB.

The performance of 3512 and 3112 is about the same. And much-much higher than what the original SeriTek/1SM2 can do. It's good so: after 20 years some improvement is to be expected.

 

[Ataman Honcho Honchev]

And here is the ASP à la MacOS 9.

I don't think, Apple will ever fix their bugs, so I had to work around their.

- On the first picture (besides few samurai-s being upset due unusual features of a Mamiya NC1000s camera) you see the ASP with the SATA card inserted. The only way I could make that card appear in ASP properly was to "patch" some properties of the CardBus bridge.

- On the second picture the samurais aren't less upset - but at least all original properties of the CardBus bridge are back to normal (after the SATA card was removed).

I fully agree with samurai-s: as much controversy as the Mamiya NC1000s could mean, that Apple ASP "feature" is even more controversial.

 

[Ataman Honcho Honchev]

Update: found on fee-Bay an interesting card from Romania.
It uses an Initio chipset I am not familiar with.

Maybe the seller agrees to sell that card...

 

[B S Magnet]

Doing a huge overnight test now with the post-SeriTek/1SEM2 (SImagic2LP) cardbus / PCMCIA card.
This test is done with the genuine SeriTek/1SEM2 hardware, but the 6.0.0 (completely new) driver.

The "victim" is the last series of Titanium G4. I got a similar 1GHZ machine from the local aftermarket as well, but it needs to be repaired / screen exchanged. This one is in fairly decent shape but is the slower, 867MHZ version.

The SeriTek/1SEM2 is using a popular Silicon Image 3512 chip which has a documented errata.
That errata wasn't known back in 2003. In 2023 the workaround is well-known and of course it is implemented since early "beta"-s of 6.0.0.

In 2003 I had to limit the simultaneous write access to both channels (that mitigated the problem).
With the errata workaround implemented (also applies to 3114!) I see no problem while writing to both channels the same time.

One is clear: the PCMCIA is not a speed demon, in particular on writes. The reads are OK, but the writes through PCMCIA - even if the simultaneous access is not a problem anymore is not going to be faster, than 30 MB/Sec.

Let's see, how it behaves overnight.

I am using two "disqualified" SSD-s: one from Kingston, one from AMD.
They are "disqualified" due long use times and "going to fail soon" status in SMART - but perfect to be abused.

These two 240GB SSD-s were set up in RAID-0 (SoftRAID 2.22) and now proceeding with FWB bench test, asked for a very long transfers and huge number of tests (guaranteed to keep the computer busy for a day).

Hope everything is OK. Other than that I will test the same with a 3112-based PCMCIA card and fix one "new" bug of mine when drives going to sleep - but the computer is thinking that the entire system is asleep.

If that is fixed and if the test is OK I will move one step forward. ;-)

I might have missed it here, but how does the 1SEM2 differ from the older, 1SM2 — in terms of hardware and core functionality?

 

[Ataman Honcho Honchev]

I might have missed it here, but how does the 1SEM2 differ from the older, 1SM2 — in terms of hardware and core functionality?

By my typo only. I am getting older.
There is no 1SEM2. I am using the card on the picture and a common card with 3112 chip for the comparison.
Already corrected the typo, thanks.

 

[B S Magnet]

By my typo only. I am getting older.
There is no 1SEM2. I am using the card on the picture and a common card with 3112 chip for the comparison.
Already corrected the typo, thanks.

No worries! Ageing is a whole, pernicious, inevitable thing.