On the weekend I decided to examine my old PowerPC 604 system. The case is from a RS/6000 type 7248, but the actual system board is from a Power Series 830. This is the famous Carolina board (or “planar” in IBM-speak), which was shared by RS/6000 43P (7248) workstations and Personal Power Series 830/850 desktops (type 6050/6070). The only significant difference was that the RS/6000 boards included a SCSI controller and the Power Series boards did not.
The differences between the Power Series 830 and 850 models were again small. The 850 system case was larger (with more slots and bays, 5/5 vs. 3/3), and the 830 was only sold as a 100MHz version while the 850 was additionally available in 120 and 133MHz variants. The system board was in all cases essentially the same except for the CPU (which is soldered to the board) and L2 cache (256K on 100/120MHz models, 512K on 133MHz models, always separate module in its own socket).
Anyway, my Frankensystem has a board FRU 12H0818. According to the EPRM, that’s a type 6070 (Power Series 850) system board. Except for the part number, it’s not apparent what differences there might possibly be between the 830 system board and a 100MHz 850 board. At any rate, the real surprise was waiting under the heatsink:
That’s right—a 133MHz PowerPC 604 processor. It’s worth noting that the 100/120/133MHz boards were separate part numbers. The CPU frequency does not appear to be user modifiable which would explain the need for separate boards (even if all of them actually used the same CPU model).
A better question is why IBM sold high-end 133MHz processors as 100MHz. I have no explanation for that, although it is not an uncommon practice within the industry. Sometimes the yields are good and there’s a shortage of cheaper/slower chips, so faster chips are sold as slower ones. However, the CPUs are then typically labeled as the slower parts and nowadays also multiplier locked. In this case, IBM may not have bothered since the CPU is soldered and effectively locked.
Unfortunately that means I’m stuck with a nice 133MHz PowerPC 604 that’s been officially crippled by IBM to only run at 100MHz…
I can still remember hearing that some 486SX processors were functional DX parts, that had their math coprocessors ‘lasered’ out of working order turning them in SX’es as they were more than capable of making proper DX chips later on, but intel still liked to sell the cheaper/lower end SX part.
It’s possible the chip failed some validation at 133 but was fine at 100 and they didn’t bother remarking it since it was for “internal use” anyway.
It’s also possible the process matured faster than expected and it’s a genuine validated 133 MHz part because they weren’t getting enough failures to fill the slower bins. This happened on almost every major Intel CPU from the P5MMX onward because their manufacturing tech is just that good, although Intel did remark the chips as slower part numbers in that case.
The original rationale was using the same packaging for DX and SX and selling the chips with a faulty FPU as SX. Later on, they just flipped it around and sold some good DX chips as SX to cover demand…
From an economics theory POV it’s an interesting phenomenon: You have a higher-priced item A but sell it as a lower-priced item B because the customer wasn’t going to buy A anyway and would be more likely to buy competitor’s item C. Since the margin on A is outrageously high anyway, you still make a profit. The trick is just convincing the people who buy A that they’re getting something their money (Intel Extreme Edition CPUs, I’m looking at you) 🙂
I can’t imagine that the validation would be done after labeling the chips… the second explanation is IMO far more likely. Since IBM manufactured both the CPUs and the boards and the chips were soldered on anyway, there was probably no point in re-labeling it as a 100MHz part. The boards came with heatsinks or fans anyway so the customer wouldn’t be bothered by the sight of a 133MHz CPU in a 100MHz board.
When yields are too good, Intel would often do a pricing refresh and drop the price of the over produced fast chip and introduce a new faster speed bin to take advantage of the better than expected process. IBM inexplicably kept the PowerPC sold by IBM slower than the PowerPC 604s sold by other manufacturers with 180MHz being PowerComputing’s peak compared to the 150MHz allowed to RS/6000 or the 133MHz allowed to the better Power Series and keeping the poor 830 at a woeful 100MHz. It doesn’t make sense to me encouraging sales to go to the competition. IBM could have upgraded all the lines, made the same profit per unit, and slowed the Pentium onslaught.
There were rumors that the RS/6000 workstation folks really didn’t want the PC Co. to be all that successful. As is often the case, just because they worked for the same company didn’t stop them fighting each other. But yeah, it doesn’t make sense.