If you want to know how dominant a piece of computer hardware really was, don’t look at its launch. Launches are easy to game with marketing budgets and reviewer freebies. Look instead at what happens three years later, when the manufacturer has already moved on to the next thing, and ask: is anyone still bothering to build accessories for it?
By that measure, the SPARCstation 2 might be the most dominant Unix workstation that ever shipped.
The setup
The SPARCstation 2 came out on November 5, 1990. It was a pizza box that looked nearly identical to the SPARCstation 1+ that came before it, but inside it had a 40 MHz Cypress CY7C601 putting out about 28.5 MIPS — roughly double the 1+. Pricing started at $14,995 and ran up to about $50,000 fully loaded.
It sold like nothing Sun had ever made before. Nine thousand units in the first ten weeks. By June 1992, Sun had moved over 540,000 of them. In the first quarter of 1992 the SS2 alone accounted for forty percent of Sun’s total revenue. One industry analyst at the time called it “the heart of Sun’s product line,” and he wasn’t exaggerating.
If you went to a research university or an engineering school in the early 90s, the SS2 was the Unix workstation. The same silicon also showed up as the lunchbox SPARCstation IPX in July 1991, which used the very similar 40 MHz Weitek W8701. Same architecture, smaller box, integrated framebuffer. Between the two of them, by Sun’s own count, around a quarter million were still in active service worldwide by mid-1993.
That last number is the interesting one.
The SS10 was supposed to bury it
In May 1992, Sun announced the SPARCstation 10. Multiprocessor capable, MBus-based, faster everything. By December of that year they had shipped 19,000 SS10s, and another 80,000 in 1993. This was supposed to be the natural progression — SS2 owners would dutifully upgrade to the SS10 the way they had moved from the SS1 to the SS2.
Except they didn’t. Or at least not nearly as fast as Sun probably wanted.
There were good reasons. The SS2 worked. SunOS 4.1.x was solid, the userspace was familiar, the SBus cards in the slots were paid for. For a CAD seat or a grad student’s desk, twenty-eight MIPS was still plenty of compute in 1993. The only thing the SS10 really offered the average user was more speed for the same workload — and a five-figure bill to get it.
The market noticed. And someone in Sunnyvale decided to do something about it.
Enter Weitek
Weitek Corporation had been part of the Sun story since the mid-80s. They built FPUs for Sun starting with the early SPARC machines, including the Weitek 3170 and 3172 that lived next to the integer unit on the SS1 and SS1+. They were a SPARC licensee. They knew the architecture cold. And they had been watching the same install base numbers everyone else had.
On June 14, 1993, Weitek announced the SPARC POWER µP — pronounced “power up.” It was a pin-compatible drop-in replacement for the CPU sockets on the SPARCstation 2 and IPX. Fifteen hundred dollars. User-installable. They shipped it with a chip extraction tool and a little manual, and the whole upgrade was supposed to take a few minutes with the case open.
The trick was a phase-locked loop clock doubler. The chip ran its internal logic at 80 MHz while keeping the system bus interface at 40 MHz, so it dropped into the existing socket and worked with the existing motherboard, memory, and SBus subsystem with no other changes. It also brought 16KB of on-chip instruction cache and 8KB of data cache, which the original SS2 CPU did not have.
Weitek claimed up to 1.9x application performance. SPARC International certified it as fully compliant. The real-world numbers were closer to 50 to 60 percent improvement on average, because the bus and memory weren’t getting any faster — but on compute-bound workloads like CAD, FEA, and circuit simulation, you could see something close to that 1.9x peak.
For a thousand bucks less than a low-end SS10, you could nearly double your existing machine.
What this tells us
Aftermarket CPU upgrades for proprietary Unix workstations were not a thing that just happened. They required a chip company willing to spend real money on design and certification, motherboard documentation that was either licensed or reverse-engineered, and most importantly, a market large enough to make it back.
Weitek did this for exactly two machines: the SS2 and the IPX. Not the SS1. Not the SS10. Not the SLC or the ELC. Just the two best-selling sun4c boxes, which happened to share the same socket and the same 40 MHz bus. They named the press release “supercharge your SPARCstation” and pitched it directly at CAD shops, EDA users, and FEA shops who had real money tied up in their existing software stacks and absolutely zero interest in re-qualifying everything on a new platform.
That’s not a niche product. That’s a company looking at an installed base of 250,000 machines that the manufacturer had effectively stopped trying to grow, and saying “yeah, we’ll build to that.”
The chip is itself the most damning evidence of how stuck-in the SS2 was. Sun had launched a faster, more modern, multiprocessor-capable replacement more than a year earlier, and the market response was enthusiastic enough about staying on the old hardware that a third party could build a business on it.
Coda
The POWER µP didn’t end up being a big commercial win for Weitek. They never built a follow-on for later SPARC generations, and within a couple of years they had pivoted hard into the PC graphics market with the P9000. The SS2s and IPXs they had upgraded soldiered on through the 90s anyway, mostly running SunOS 4.1.3 or 4.1.4, eventually getting decommissioned not because they were slow but because the surrounding infrastructure — networks, storage, software — moved past them.
I have one of the upgraded chips sitting in my own SPARCstation 2 right now, paired with a DataRAM SBus card pushing the machine to 128MB. It is, by 1993 standards, the maximum-spec configuration of what was already the most popular Unix workstation in the world. And it’s the kind of thing you can only put together because, three decades ago, enough people refused to upgrade that an entire third-party ecosystem grew up around their refusal.
Sometimes the most interesting story about a computer isn’t how it was launched. It’s what other people built for it, after the company that made it had already stopped paying attention.