Installing and testing a NewerTech Quadra Cache card on a Centris 650

[Lockjaw8Y9]

In the early 1990s when I was in graduate school, I bought a Centris 650 new. Along the way, I purchased a NewerTech Quadra Cache card to get some additional performance. I also modded that machine back in the day to accept a faster clock chip, per my chats with Marc Schrier and Marlin Prowell. They produced seminal documentation for overclocking these machines, which I see cited frequently here.

I decided to install that very card in my current Centris 650, which I had modded to report as a Quadra 650 like I had done using Marlin's instructions. I made two videos about the process, one for the logic board mod (which turns the Centris 650 into a Quadra 650), and one which measures performance with the cache card. I hope others might find them useful.

I recall having stability issues when I installed an Apple 601 PDS card in that Centris 650 while the cache card was installed, especially when overclocking. Does anyone else have this experience?

 

[register]

Everytime (back in the day, when I used a 68040 Mac as my daily work machine) I researched for 68040 cache upgrades I found information hinting towards stability issues of several kinds, no matter if even a PPC upgrade was involved. Thus I never invested into any of those cards, because I valued the excellent stability of my system. By the way: would one expect in the system running a 601 cpu any performance difference wether the 68040 cache option was installed or not?

 

[beachycove]

Interesting stuff. Some time back I searched in vain for what must have been years for an 040 PDS cache card. In the end I came up with a Daystar Quad 040 PDS card (40mhz and I think 128k on said card), where there have been no stability implications at all running in a Q650, or before that, in a Q950.

 

[Lockjaw8Y9]

Everytime (back in the day, when I used a 68040 Mac as my daily work machine) I researched for 68040 cache upgrades I found information hinting towards stability issues of several kinds, no matter if even a PPC upgrade was involved. Thus I never invested into any of those cards, because I valued the excellent stability of my system. By the way: would one expect in the system running a 601 cpu any performance difference wether the 68040 cache option was installed or not?

No, I would not expect that any cache attached to the '040 would have an impact on the 601. I would expect that it would be disabled, since the '040 was disabled while the 601 was enabled. The only reason I tried to keep both installed was because there were specific technical applications (Maple, Mathematica) that ran better in native '040 than in emulation on the 601. Of course, PPC native versions did come out, and there was no reason to keep the cache card installed at all.

 

[Lockjaw8Y9]

Interesting stuff. Some time back I searched in vain for what must have been years for an 040 PDS cache card. In the end I came up with a Daystar Quad 040 PDS card (40mhz and I think 128k on said card), where there have been no stability implications at all running in a Q650, or before that, in a Q950.

This cache installs on the CPU socket, so the PDS is still available.

 

[beachycove]

Yes, of course. The one you have has its advantages — above all the ability to use the 601 card as well.

On the cache question, you will know more than me, given your background. My understanding is that the cache holds data and instructions just outside the CPU so that it does not need to be fetched from RAM or (presumably less likely) the Hard Drive. Would benchmarking software of the period be designed to test for that? I would have thought that cache might speed up repetitive tasks but little else, giving (maybe?) something like a 10% boost in day-to-day computing. I think in the G1 and G2 PPCs, where cache could readily be added, you got something like that speed improvement with 256k cache cards.

 

[Lockjaw8Y9]

Yes, of course. The one you have has its advantages — above all the ability to use the 601 card as well.

On the cache question, you will know more than me, given your background. My understanding is that the cache holds data and instructions just outside the CPU so that it does not need to be fetched from RAM or (presumably less likely) the Hard Drive. Would benchmarking software of the period be designed to test for that? I would have thought that cache might speed up repetitive tasks but little else, giving (maybe?) something like a 10% boost in day-to-day computing. I think in the G1 and G2 PPCs, where cache could readily be added, you got something like that speed improvement with 256k cache cards.

The CPU and MMU managed the cache, so its presence allows the CPU to hold on to frequently accessed instructions, which it can then access at the speed of the CPU and not at the speed of the data bus. This allowed for a marginal improvement in performance, as instructions could be read and executed faster from the cache. Of course, the cache had management overhead, so the effective performance boost (according to the tests I just ran) was not more than 10%. Overclocking gave a way more significant performance boost, but the combination of an overclocked CPU and cache was worth it for us, when we were doing physics simulations. Note that overclocking does introduce more heat, so it is advised to add some airflow to the radiator that sits on top of the CPU.